Compare commits

..

4 Commits

253 changed files with 2341 additions and 597078 deletions

View File

@ -1,2 +1,2 @@
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-dynamic-vram
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --disable-smart-memory
pause

View File

@ -1,2 +0,0 @@
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
pause

View File

@ -1,103 +0,0 @@
#!/usr/bin/env bash
# Checks pull request commits for AI agent Co-authored-by trailers.
# Exits non-zero when any are found and prints fix instructions.
set -euo pipefail
base_sha="${1:?usage: check-ai-co-authors.sh <base_sha> <head_sha>}"
head_sha="${2:?usage: check-ai-co-authors.sh <base_sha> <head_sha>}"
# Known AI coding-agent trailer patterns (case-insensitive).
# Each entry is an extended-regex fragment matched against Co-authored-by lines.
AGENT_PATTERNS=(
# Anthropic — Claude Code / Amp
'noreply@anthropic\.com'
# Cursor
'cursoragent@cursor\.com'
# GitHub Copilot
'copilot-swe-agent\[bot\]'
'copilot@github\.com'
# OpenAI Codex
'noreply@openai\.com'
'codex@openai\.com'
# Aider
'aider@aider\.chat'
# Google — Gemini / Jules
'gemini@google\.com'
'jules@google\.com'
# Windsurf / Codeium
'@codeium\.com'
# Devin
'devin-ai-integration\[bot\]'
'devin@cognition\.ai'
'devin@cognition-labs\.com'
# Amazon Q Developer
'amazon-q-developer'
'@amazon\.com.*[Qq].[Dd]eveloper'
# Cline
'cline-bot'
'cline@cline\.ai'
# Continue
'continue-agent'
'continue@continue\.dev'
# Sourcegraph
'noreply@sourcegraph\.com'
# Generic catch-alls for common agent name patterns
'Co-authored-by:.*\b[Cc]laude\b'
'Co-authored-by:.*\b[Cc]opilot\b'
'Co-authored-by:.*\b[Cc]ursor\b'
'Co-authored-by:.*\b[Cc]odex\b'
'Co-authored-by:.*\b[Gg]emini\b'
'Co-authored-by:.*\b[Aa]ider\b'
'Co-authored-by:.*\b[Dd]evin\b'
'Co-authored-by:.*\b[Ww]indsurf\b'
'Co-authored-by:.*\b[Cc]line\b'
'Co-authored-by:.*\b[Aa]mazon Q\b'
'Co-authored-by:.*\b[Jj]ules\b'
'Co-authored-by:.*\bOpenCode\b'
)
# Build a single alternation regex from all patterns.
regex=""
for pattern in "${AGENT_PATTERNS[@]}"; do
if [[ -n "$regex" ]]; then
regex="${regex}|${pattern}"
else
regex="$pattern"
fi
done
# Collect Co-authored-by lines from every commit in the PR range.
violations=""
while IFS= read -r sha; do
message="$(git log -1 --format='%B' "$sha")"
matched_lines="$(echo "$message" | grep -iE "^Co-authored-by:" || true)"
if [[ -z "$matched_lines" ]]; then
continue
fi
while IFS= read -r line; do
if echo "$line" | grep -iqE "$regex"; then
short="$(git log -1 --format='%h' "$sha")"
violations="${violations} ${short}: ${line}"$'\n'
fi
done <<< "$matched_lines"
done < <(git rev-list "${base_sha}..${head_sha}")
if [[ -n "$violations" ]]; then
echo "::error::AI agent Co-authored-by trailers detected in PR commits."
echo ""
echo "The following commits contain Co-authored-by trailers from AI coding agents:"
echo ""
echo "$violations"
echo "These trailers should be removed before merging."
echo ""
echo "To fix, rewrite the commit messages with:"
echo " git rebase -i ${base_sha}"
echo ""
echo "and remove the Co-authored-by lines, then force-push your branch."
echo ""
echo "If you believe this is a false positive, please open an issue."
exit 1
fi
echo "No AI agent Co-authored-by trailers found."

View File

@ -1,19 +0,0 @@
name: Check AI Co-Authors
on:
pull_request:
branches: ['*']
jobs:
check-ai-co-authors:
name: Check for AI agent co-author trailers
runs-on: ubuntu-latest
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Check commits for AI co-author trailers
run: bash .github/scripts/check-ai-co-authors.sh "${{ github.event.pull_request.base.sha }}" "${{ github.event.pull_request.head.sha }}"

View File

@ -20,12 +20,29 @@ jobs:
git_tag: ${{ inputs.git_tag }}
cache_tag: "cu130"
python_minor: "13"
python_patch: "12"
python_patch: "11"
rel_name: "nvidia"
rel_extra_name: ""
test_release: true
secrets: inherit
release_nvidia_cu128:
permissions:
contents: "write"
packages: "write"
pull-requests: "read"
name: "Release NVIDIA cu128"
uses: ./.github/workflows/stable-release.yml
with:
git_tag: ${{ inputs.git_tag }}
cache_tag: "cu128"
python_minor: "12"
python_patch: "10"
rel_name: "nvidia"
rel_extra_name: "_cu128"
test_release: true
secrets: inherit
release_nvidia_cu126:
permissions:
contents: "write"
@ -59,20 +76,3 @@ jobs:
rel_extra_name: ""
test_release: false
secrets: inherit
release_xpu:
permissions:
contents: "write"
packages: "write"
pull-requests: "read"
name: "Release Intel XPU"
uses: ./.github/workflows/stable-release.yml
with:
git_tag: ${{ inputs.git_tag }}
cache_tag: "xpu"
python_minor: "13"
python_patch: "12"
rel_name: "intel"
rel_extra_name: ""
test_release: true
secrets: inherit

View File

@ -1,45 +0,0 @@
name: Tag Dispatch to Cloud
on:
push:
tags:
- 'v*'
jobs:
dispatch-cloud:
runs-on: ubuntu-latest
steps:
- name: Send repository dispatch to cloud
env:
DISPATCH_TOKEN: ${{ secrets.CLOUD_REPO_DISPATCH_TOKEN }}
RELEASE_TAG: ${{ github.ref_name }}
run: |
set -euo pipefail
if [ -z "${DISPATCH_TOKEN:-}" ]; then
echo "::error::CLOUD_REPO_DISPATCH_TOKEN is required but not set."
exit 1
fi
RELEASE_URL="https://github.com/${{ github.repository }}/releases/tag/${RELEASE_TAG}"
PAYLOAD="$(jq -n \
--arg release_tag "$RELEASE_TAG" \
--arg release_url "$RELEASE_URL" \
'{
event_type: "comfyui_tag_pushed",
client_payload: {
release_tag: $release_tag,
release_url: $release_url
}
}')"
curl -fsSL \
-X POST \
-H "Accept: application/vnd.github+json" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer ${DISPATCH_TOKEN}" \
https://api.github.com/repos/Comfy-Org/cloud/dispatches \
-d "$PAYLOAD"
echo "✅ Dispatched ComfyUI tag ${RELEASE_TAG} to Comfy-Org/cloud"

1
.gitignore vendored
View File

@ -21,5 +21,6 @@ venv*/
*.log
web_custom_versions/
.DS_Store
openapi.yaml
filtered-openapi.yaml
uv.lock

View File

@ -1,2 +1,2 @@
# Admins
* @comfyanonymous @kosinkadink @guill @alexisrolland @rattus128 @kijai
* @comfyanonymous @kosinkadink @guill

View File

@ -139,9 +139,9 @@ Example:
"_quantization_metadata": {
"format_version": "1.0",
"layers": {
"model.layers.0.mlp.up_proj": {"format": "float8_e4m3fn"},
"model.layers.0.mlp.down_proj": {"format": "float8_e4m3fn"},
"model.layers.1.mlp.up_proj": {"format": "float8_e4m3fn"}
"model.layers.0.mlp.up_proj": "float8_e4m3fn",
"model.layers.0.mlp.down_proj": "float8_e4m3fn",
"model.layers.1.mlp.up_proj": "float8_e4m3fn"
}
}
}
@ -165,4 +165,4 @@ Activation quantization (e.g., for FP8 Tensor Core operations) requires `input_s
3. **Compute scales**: Derive `input_scale` from collected statistics
4. **Store in checkpoint**: Save `input_scale` parameters alongside weights
The calibration dataset should be representative of your target use case. For diffusion models, this typically means a diverse set of prompts and generation parameters.
The calibration dataset should be representative of your target use case. For diffusion models, this typically means a diverse set of prompts and generation parameters.

View File

@ -1,7 +1,7 @@
<div align="center">
# ComfyUI
**The most powerful and modular AI engine for content creation.**
**The most powerful and modular visual AI engine and application.**
[![Website][website-shield]][website-url]
@ -31,21 +31,13 @@
[github-downloads-latest-shield]: https://img.shields.io/github/downloads/comfyanonymous/ComfyUI/latest/total?style=flat&label=downloads%40latest
[github-downloads-link]: https://github.com/comfyanonymous/ComfyUI/releases
<img width="1590" height="795" alt="ComfyUI Screenshot" src="https://github.com/user-attachments/assets/36e065e0-bfae-4456-8c7f-8369d5ea48a2" />
<br>
![ComfyUI Screenshot](https://github.com/user-attachments/assets/7ccaf2c1-9b72-41ae-9a89-5688c94b7abe)
</div>
ComfyUI is the AI creation engine for visual professionals who demand control over every model, every parameter, and every output. Its powerful and modular node graph interface empowers creatives to generate images, videos, 3D models, audio, and more...
- ComfyUI natively supports the latest open-source state of the art models.
- API nodes provide access to the best closed source models such as Nano Banana, Seedance, Hunyuan3D, etc.
- It is available on Windows, Linux, and macOS, locally with our desktop application or on our cloud.
- The most sophisticated workflows can be exposed through a simple UI thanks to App Mode.
- It integrates seamlessly into production pipelines with our API endpoints.
ComfyUI lets you design and execute advanced stable diffusion pipelines using a graph/nodes/flowchart based interface. Available on Windows, Linux, and macOS.
## Get Started
### Local
#### [Desktop Application](https://www.comfy.org/download)
- The easiest way to get started.
- Available on Windows & macOS.
@ -57,17 +49,11 @@ ComfyUI is the AI creation engine for visual professionals who demand control ov
#### [Manual Install](#manual-install-windows-linux)
Supports all operating systems and GPU types (NVIDIA, AMD, Intel, Apple Silicon, Ascend).
### Cloud
#### [Comfy Cloud](https://www.comfy.org/cloud)
- Our official paid cloud version for those who can't afford local hardware.
## Examples
See what ComfyUI can do with the [newer template workflows](https://comfy.org/workflows) or old [example workflows](https://comfyanonymous.github.io/ComfyUI_examples/).
## [Examples](https://comfyanonymous.github.io/ComfyUI_examples/)
See what ComfyUI can do with the [example workflows](https://comfyanonymous.github.io/ComfyUI_examples/).
## Features
- Nodes/graph/flowchart interface to experiment and create complex Stable Diffusion workflows without needing to code anything.
- NOTE: There are many more models supported than the list below, if you want to see what is supported see our templates list inside ComfyUI.
- Image Models
- SD1.x, SD2.x ([unCLIP](https://comfyanonymous.github.io/ComfyUI_examples/unclip/))
- [SDXL](https://comfyanonymous.github.io/ComfyUI_examples/sdxl/), [SDXL Turbo](https://comfyanonymous.github.io/ComfyUI_examples/sdturbo/)
@ -83,7 +69,6 @@ See what ComfyUI can do with the [newer template workflows](https://comfy.org/wo
- [Hunyuan Image 2.1](https://comfyanonymous.github.io/ComfyUI_examples/hunyuan_image/)
- [Flux 2](https://comfyanonymous.github.io/ComfyUI_examples/flux2/)
- [Z Image](https://comfyanonymous.github.io/ComfyUI_examples/z_image/)
- Ernie Image
- Image Editing Models
- [Omnigen 2](https://comfyanonymous.github.io/ComfyUI_examples/omnigen/)
- [Flux Kontext](https://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-kontext-image-editing-model)
@ -144,7 +129,7 @@ ComfyUI follows a weekly release cycle targeting Monday but this regularly chang
- Builds a new release using the latest stable core version
3. **[ComfyUI Frontend](https://github.com/Comfy-Org/ComfyUI_frontend)**
- Every 2+ weeks frontend updates are merged into the core repository
- Weekly frontend updates are merged into the core repository
- Features are frozen for the upcoming core release
- Development continues for the next release cycle
@ -200,15 +185,11 @@ If you have trouble extracting it, right click the file -> properties -> unblock
The portable above currently comes with python 3.13 and pytorch cuda 13.0. Update your Nvidia drivers if it doesn't start.
#### All Official Portable Downloads:
#### Alternative Downloads:
[Portable for AMD GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_amd.7z)
[Experimental portable for AMD GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_amd.7z)
[Portable for Intel GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_intel.7z)
[Portable for Nvidia GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia.7z) (supports 20 series and above).
[Portable for Nvidia GPUs with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs).
[Portable with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs).
#### How do I share models between another UI and ComfyUI?
@ -244,7 +225,7 @@ Put your VAE in: models/vae
AMD users can install rocm and pytorch with pip if you don't have it already installed, this is the command to install the stable version:
```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2```
```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.1```
This is the command to install the nightly with ROCm 7.2 which might have some performance improvements:
@ -287,7 +268,7 @@ Nvidia users should install stable pytorch using this command:
This is the command to install pytorch nightly instead which might have performance improvements.
```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu132```
```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu130```
#### Troubleshooting

View File

@ -8,7 +8,7 @@ from alembic import context
config = context.config
from app.database.models import Base, NAMING_CONVENTION
from app.database.models import Base
target_metadata = Base.metadata
# other values from the config, defined by the needs of env.py,
@ -51,10 +51,7 @@ def run_migrations_online() -> None:
with connectable.connect() as connection:
context.configure(
connection=connection,
target_metadata=target_metadata,
render_as_batch=True,
naming_convention=NAMING_CONVENTION,
connection=connection, target_metadata=target_metadata
)
with context.begin_transaction():

View File

@ -1,98 +0,0 @@
"""
Add system_metadata and job_id columns to asset_references.
Change preview_id FK from assets.id to asset_references.id.
Revision ID: 0003_add_metadata_job_id
Revises: 0002_merge_to_asset_references
Create Date: 2026-03-09
"""
from alembic import op
import sqlalchemy as sa
from app.database.models import NAMING_CONVENTION
revision = "0003_add_metadata_job_id"
down_revision = "0002_merge_to_asset_references"
branch_labels = None
depends_on = None
def upgrade() -> None:
with op.batch_alter_table("asset_references") as batch_op:
batch_op.add_column(
sa.Column("system_metadata", sa.JSON(), nullable=True)
)
batch_op.add_column(
sa.Column("job_id", sa.String(length=36), nullable=True)
)
# Change preview_id FK from assets.id to asset_references.id (self-ref).
# Existing values are asset-content IDs that won't match reference IDs,
# so null them out first.
op.execute("UPDATE asset_references SET preview_id = NULL WHERE preview_id IS NOT NULL")
with op.batch_alter_table(
"asset_references", naming_convention=NAMING_CONVENTION
) as batch_op:
batch_op.drop_constraint(
"fk_asset_references_preview_id_assets", type_="foreignkey"
)
batch_op.create_foreign_key(
"fk_asset_references_preview_id_asset_references",
"asset_references",
["preview_id"],
["id"],
ondelete="SET NULL",
)
batch_op.create_index(
"ix_asset_references_preview_id", ["preview_id"]
)
# Purge any all-null meta rows before adding the constraint
op.execute(
"DELETE FROM asset_reference_meta"
" WHERE val_str IS NULL AND val_num IS NULL AND val_bool IS NULL AND val_json IS NULL"
)
with op.batch_alter_table("asset_reference_meta") as batch_op:
batch_op.create_check_constraint(
"ck_asset_reference_meta_has_value",
"val_str IS NOT NULL OR val_num IS NOT NULL OR val_bool IS NOT NULL OR val_json IS NOT NULL",
)
def downgrade() -> None:
# SQLite doesn't reflect CHECK constraints, so we must declare it
# explicitly via table_args for the batch recreate to find it.
# Use the fully-rendered constraint name to avoid the naming convention
# doubling the prefix.
with op.batch_alter_table(
"asset_reference_meta",
table_args=[
sa.CheckConstraint(
"val_str IS NOT NULL OR val_num IS NOT NULL OR val_bool IS NOT NULL OR val_json IS NOT NULL",
name="ck_asset_reference_meta_has_value",
),
],
) as batch_op:
batch_op.drop_constraint(
"ck_asset_reference_meta_has_value", type_="check"
)
with op.batch_alter_table(
"asset_references", naming_convention=NAMING_CONVENTION
) as batch_op:
batch_op.drop_index("ix_asset_references_preview_id")
batch_op.drop_constraint(
"fk_asset_references_preview_id_asset_references", type_="foreignkey"
)
batch_op.create_foreign_key(
"fk_asset_references_preview_id_assets",
"assets",
["preview_id"],
["id"],
ondelete="SET NULL",
)
with op.batch_alter_table("asset_references") as batch_op:
batch_op.drop_column("job_id")
batch_op.drop_column("system_metadata")

View File

@ -67,7 +67,7 @@ class InternalRoutes:
(entry for entry in os.scandir(directory) if is_visible_file(entry)),
key=lambda entry: -entry.stat().st_mtime
)
return web.json_response([f"{entry.name} [{directory_type}]" for entry in sorted_files], status=200)
return web.json_response([entry.name for entry in sorted_files], status=200)
def get_app(self):

View File

@ -13,7 +13,6 @@ from pydantic import ValidationError
import folder_paths
from app import user_manager
from app.assets.api import schemas_in, schemas_out
from app.assets.services import schemas
from app.assets.api.schemas_in import (
AssetValidationError,
UploadError,
@ -39,7 +38,6 @@ from app.assets.services import (
update_asset_metadata,
upload_from_temp_path,
)
from app.assets.services.tagging import list_tag_histogram
ROUTES = web.RouteTableDef()
USER_MANAGER: user_manager.UserManager | None = None
@ -124,61 +122,6 @@ def _validate_sort_field(requested: str | None) -> str:
return "created_at"
def _build_preview_url_from_view(tags: list[str], user_metadata: dict[str, Any] | None) -> str | None:
"""Build a /api/view preview URL from asset tags and user_metadata filename."""
if not user_metadata:
return None
filename = user_metadata.get("filename")
if not filename:
return None
if "input" in tags:
view_type = "input"
elif "output" in tags:
view_type = "output"
else:
return None
subfolder = ""
if "/" in filename:
subfolder, filename = filename.rsplit("/", 1)
encoded_filename = urllib.parse.quote(filename, safe="")
url = f"/api/view?type={view_type}&filename={encoded_filename}"
if subfolder:
url += f"&subfolder={urllib.parse.quote(subfolder, safe='')}"
return url
def _build_asset_response(result: schemas.AssetDetailResult | schemas.UploadResult) -> schemas_out.Asset:
"""Build an Asset response from a service result."""
if result.ref.preview_id:
preview_detail = get_asset_detail(result.ref.preview_id)
if preview_detail:
preview_url = _build_preview_url_from_view(preview_detail.tags, preview_detail.ref.user_metadata)
else:
preview_url = None
else:
preview_url = _build_preview_url_from_view(result.tags, result.ref.user_metadata)
return schemas_out.Asset(
id=result.ref.id,
name=result.ref.name,
asset_hash=result.asset.hash if result.asset else None,
size=int(result.asset.size_bytes) if result.asset else None,
mime_type=result.asset.mime_type if result.asset else None,
tags=result.tags,
preview_url=preview_url,
preview_id=result.ref.preview_id,
user_metadata=result.ref.user_metadata or {},
metadata=result.ref.system_metadata,
job_id=result.ref.job_id,
prompt_id=result.ref.job_id, # deprecated: mirrors job_id for cloud compat
created_at=result.ref.created_at,
updated_at=result.ref.updated_at,
last_access_time=result.ref.last_access_time,
)
@ROUTES.head("/api/assets/hash/{hash}")
@_require_assets_feature_enabled
async def head_asset_by_hash(request: web.Request) -> web.Response:
@ -221,7 +164,20 @@ async def list_assets_route(request: web.Request) -> web.Response:
order=order,
)
summaries = [_build_asset_response(item) for item in result.items]
summaries = [
schemas_out.AssetSummary(
id=item.ref.id,
name=item.ref.name,
asset_hash=item.asset.hash if item.asset else None,
size=int(item.asset.size_bytes) if item.asset else None,
mime_type=item.asset.mime_type if item.asset else None,
tags=item.tags,
created_at=item.ref.created_at,
updated_at=item.ref.updated_at,
last_access_time=item.ref.last_access_time,
)
for item in result.items
]
payload = schemas_out.AssetsList(
assets=summaries,
@ -251,7 +207,18 @@ async def get_asset_route(request: web.Request) -> web.Response:
{"id": reference_id},
)
payload = _build_asset_response(result)
payload = schemas_out.AssetDetail(
id=result.ref.id,
name=result.ref.name,
asset_hash=result.asset.hash if result.asset else None,
size=int(result.asset.size_bytes) if result.asset else None,
mime_type=result.asset.mime_type if result.asset else None,
tags=result.tags,
user_metadata=result.ref.user_metadata or {},
preview_id=result.ref.preview_id,
created_at=result.ref.created_at,
last_access_time=result.ref.last_access_time,
)
except ValueError as e:
return _build_error_response(
404, "ASSET_NOT_FOUND", str(e), {"id": reference_id}
@ -263,7 +230,7 @@ async def get_asset_route(request: web.Request) -> web.Response:
USER_MANAGER.get_request_user_id(request),
)
return _build_error_response(500, "INTERNAL", "Unexpected server error.")
return web.json_response(payload.model_dump(mode="json", exclude_none=True), status=200)
return web.json_response(payload.model_dump(mode="json"), status=200)
@ROUTES.get(f"/api/assets/{{id:{UUID_RE}}}/content")
@ -345,31 +312,32 @@ async def create_asset_from_hash_route(request: web.Request) -> web.Response:
400, "INVALID_JSON", "Request body must be valid JSON."
)
# Derive name from hash if not provided
name = body.name
if name is None:
name = body.hash.split(":", 1)[1] if ":" in body.hash else body.hash
result = create_from_hash(
hash_str=body.hash,
name=name,
name=body.name,
tags=body.tags,
user_metadata=body.user_metadata,
owner_id=USER_MANAGER.get_request_user_id(request),
mime_type=body.mime_type,
preview_id=body.preview_id,
)
if result is None:
return _build_error_response(
404, "ASSET_NOT_FOUND", f"Asset content {body.hash} does not exist"
)
asset = _build_asset_response(result)
payload_out = schemas_out.AssetCreated(
**asset.model_dump(),
id=result.ref.id,
name=result.ref.name,
asset_hash=result.asset.hash,
size=int(result.asset.size_bytes),
mime_type=result.asset.mime_type,
tags=result.tags,
user_metadata=result.ref.user_metadata or {},
preview_id=result.ref.preview_id,
created_at=result.ref.created_at,
last_access_time=result.ref.last_access_time,
created_new=result.created_new,
)
return web.json_response(payload_out.model_dump(mode="json", exclude_none=True), status=201)
return web.json_response(payload_out.model_dump(mode="json"), status=201)
@ROUTES.post("/api/assets")
@ -390,8 +358,6 @@ async def upload_asset(request: web.Request) -> web.Response:
"name": parsed.provided_name,
"user_metadata": parsed.user_metadata_raw,
"hash": parsed.provided_hash,
"mime_type": parsed.provided_mime_type,
"preview_id": parsed.provided_preview_id,
}
)
except ValidationError as ve:
@ -420,8 +386,6 @@ async def upload_asset(request: web.Request) -> web.Response:
tags=spec.tags,
user_metadata=spec.user_metadata or {},
owner_id=owner_id,
mime_type=spec.mime_type,
preview_id=spec.preview_id,
)
if result is None:
delete_temp_file_if_exists(parsed.tmp_path)
@ -446,8 +410,6 @@ async def upload_asset(request: web.Request) -> web.Response:
client_filename=parsed.file_client_name,
owner_id=owner_id,
expected_hash=spec.hash,
mime_type=spec.mime_type,
preview_id=spec.preview_id,
)
except AssetValidationError as e:
delete_temp_file_if_exists(parsed.tmp_path)
@ -466,13 +428,21 @@ async def upload_asset(request: web.Request) -> web.Response:
logging.exception("upload_asset failed for owner_id=%s", owner_id)
return _build_error_response(500, "INTERNAL", "Unexpected server error.")
asset = _build_asset_response(result)
payload_out = schemas_out.AssetCreated(
**asset.model_dump(),
payload = schemas_out.AssetCreated(
id=result.ref.id,
name=result.ref.name,
asset_hash=result.asset.hash,
size=int(result.asset.size_bytes),
mime_type=result.asset.mime_type,
tags=result.tags,
user_metadata=result.ref.user_metadata or {},
preview_id=result.ref.preview_id,
created_at=result.ref.created_at,
last_access_time=result.ref.last_access_time,
created_new=result.created_new,
)
status = 201 if result.created_new else 200
return web.json_response(payload_out.model_dump(mode="json", exclude_none=True), status=status)
return web.json_response(payload.model_dump(mode="json"), status=status)
@ROUTES.put(f"/api/assets/{{id:{UUID_RE}}}")
@ -494,9 +464,15 @@ async def update_asset_route(request: web.Request) -> web.Response:
name=body.name,
user_metadata=body.user_metadata,
owner_id=USER_MANAGER.get_request_user_id(request),
preview_id=body.preview_id,
)
payload = _build_asset_response(result)
payload = schemas_out.AssetUpdated(
id=result.ref.id,
name=result.ref.name,
asset_hash=result.asset.hash if result.asset else None,
tags=result.tags,
user_metadata=result.ref.user_metadata or {},
updated_at=result.ref.updated_at,
)
except PermissionError as pe:
return _build_error_response(403, "FORBIDDEN", str(pe), {"id": reference_id})
except ValueError as ve:
@ -510,7 +486,7 @@ async def update_asset_route(request: web.Request) -> web.Response:
USER_MANAGER.get_request_user_id(request),
)
return _build_error_response(500, "INTERNAL", "Unexpected server error.")
return web.json_response(payload.model_dump(mode="json", exclude_none=True), status=200)
return web.json_response(payload.model_dump(mode="json"), status=200)
@ROUTES.delete(f"/api/assets/{{id:{UUID_RE}}}")
@ -579,7 +555,7 @@ async def get_tags(request: web.Request) -> web.Response:
payload = schemas_out.TagsList(
tags=tags, total=total, has_more=(query.offset + len(tags)) < total
)
return web.json_response(payload.model_dump(mode="json", exclude_none=True))
return web.json_response(payload.model_dump(mode="json"))
@ROUTES.post(f"/api/assets/{{id:{UUID_RE}}}/tags")
@ -627,7 +603,7 @@ async def add_asset_tags(request: web.Request) -> web.Response:
)
return _build_error_response(500, "INTERNAL", "Unexpected server error.")
return web.json_response(payload.model_dump(mode="json", exclude_none=True), status=200)
return web.json_response(payload.model_dump(mode="json"), status=200)
@ROUTES.delete(f"/api/assets/{{id:{UUID_RE}}}/tags")
@ -674,29 +650,7 @@ async def delete_asset_tags(request: web.Request) -> web.Response:
)
return _build_error_response(500, "INTERNAL", "Unexpected server error.")
return web.json_response(payload.model_dump(mode="json", exclude_none=True), status=200)
@ROUTES.get("/api/assets/tags/refine")
@_require_assets_feature_enabled
async def get_tags_refine(request: web.Request) -> web.Response:
"""GET request to get tag histogram for filtered assets."""
query_dict = get_query_dict(request)
try:
q = schemas_in.TagsRefineQuery.model_validate(query_dict)
except ValidationError as ve:
return _build_validation_error_response("INVALID_QUERY", ve)
tag_counts = list_tag_histogram(
owner_id=USER_MANAGER.get_request_user_id(request),
include_tags=q.include_tags,
exclude_tags=q.exclude_tags,
name_contains=q.name_contains,
metadata_filter=q.metadata_filter,
limit=q.limit,
)
payload = schemas_out.TagHistogram(tag_counts=tag_counts)
return web.json_response(payload.model_dump(mode="json", exclude_none=True), status=200)
return web.json_response(payload.model_dump(mode="json"), status=200)
@ROUTES.post("/api/assets/seed")

View File

@ -45,8 +45,6 @@ class ParsedUpload:
user_metadata_raw: str | None
provided_hash: str | None
provided_hash_exists: bool | None
provided_mime_type: str | None = None
provided_preview_id: str | None = None
class ListAssetsQuery(BaseModel):
@ -100,17 +98,11 @@ class ListAssetsQuery(BaseModel):
class UpdateAssetBody(BaseModel):
name: str | None = None
user_metadata: dict[str, Any] | None = None
preview_id: str | None = None # references an asset_reference id, not an asset id
@model_validator(mode="after")
def _validate_at_least_one_field(self):
if all(
v is None
for v in (self.name, self.user_metadata, self.preview_id)
):
raise ValueError(
"Provide at least one of: name, user_metadata, preview_id."
)
if self.name is None and self.user_metadata is None:
raise ValueError("Provide at least one of: name, user_metadata.")
return self
@ -118,11 +110,9 @@ class CreateFromHashBody(BaseModel):
model_config = ConfigDict(extra="ignore", str_strip_whitespace=True)
hash: str
name: str | None = None
name: str
tags: list[str] = Field(default_factory=list)
user_metadata: dict[str, Any] = Field(default_factory=dict)
mime_type: str | None = None
preview_id: str | None = None # references an asset_reference id, not an asset id
@field_validator("hash")
@classmethod
@ -148,44 +138,6 @@ class CreateFromHashBody(BaseModel):
return []
class TagsRefineQuery(BaseModel):
include_tags: list[str] = Field(default_factory=list)
exclude_tags: list[str] = Field(default_factory=list)
name_contains: str | None = None
metadata_filter: dict[str, Any] | None = None
limit: conint(ge=1, le=1000) = 100
@field_validator("include_tags", "exclude_tags", mode="before")
@classmethod
def _split_csv_tags(cls, v):
if v is None:
return []
if isinstance(v, str):
return [t.strip() for t in v.split(",") if t.strip()]
if isinstance(v, list):
out: list[str] = []
for item in v:
if isinstance(item, str):
out.extend([t.strip() for t in item.split(",") if t.strip()])
return out
return v
@field_validator("metadata_filter", mode="before")
@classmethod
def _parse_metadata_json(cls, v):
if v is None or isinstance(v, dict):
return v
if isinstance(v, str) and v.strip():
try:
parsed = json.loads(v)
except Exception as e:
raise ValueError(f"metadata_filter must be JSON: {e}") from e
if not isinstance(parsed, dict):
raise ValueError("metadata_filter must be a JSON object")
return parsed
return None
class TagsListQuery(BaseModel):
model_config = ConfigDict(extra="ignore", str_strip_whitespace=True)
@ -234,25 +186,21 @@ class TagsRemove(TagsAdd):
class UploadAssetSpec(BaseModel):
"""Upload Asset operation.
- tags: optional list; if provided, first is root ('models'|'input'|'output');
- tags: ordered; first is root ('models'|'input'|'output');
if root == 'models', second must be a valid category
- name: display name
- user_metadata: arbitrary JSON object (optional)
- hash: optional canonical 'blake3:<hex>' for validation / fast-path
- mime_type: optional MIME type override
- preview_id: optional asset_reference ID for preview
Files are stored using the content hash as filename stem.
"""
model_config = ConfigDict(extra="ignore", str_strip_whitespace=True)
tags: list[str] = Field(default_factory=list)
tags: list[str] = Field(..., min_length=1)
name: str | None = Field(default=None, max_length=512, description="Display Name")
user_metadata: dict[str, Any] = Field(default_factory=dict)
hash: str | None = Field(default=None)
mime_type: str | None = Field(default=None)
preview_id: str | None = Field(default=None) # references an asset_reference id
@field_validator("hash", mode="before")
@classmethod
@ -331,7 +279,7 @@ class UploadAssetSpec(BaseModel):
@model_validator(mode="after")
def _validate_order(self):
if not self.tags:
raise ValueError("at least one tag is required for uploads")
raise ValueError("tags must be provided and non-empty")
root = self.tags[0]
if root not in {"models", "input", "output"}:
raise ValueError("first tag must be one of: models, input, output")

View File

@ -4,10 +4,7 @@ from typing import Any
from pydantic import BaseModel, ConfigDict, Field, field_serializer
class Asset(BaseModel):
"""API view of an asset. Maps to DB ``AssetReference`` joined with its ``Asset`` blob;
``id`` here is the AssetReference id, not the content-addressed Asset id."""
class AssetSummary(BaseModel):
id: str
name: str
asset_hash: str | None = None
@ -15,14 +12,8 @@ class Asset(BaseModel):
mime_type: str | None = None
tags: list[str] = Field(default_factory=list)
preview_url: str | None = None
preview_id: str | None = None # references an asset_reference id, not an asset id
user_metadata: dict[str, Any] = Field(default_factory=dict)
is_immutable: bool = False
metadata: dict[str, Any] | None = None
job_id: str | None = None
prompt_id: str | None = None # deprecated: use job_id
created_at: datetime
updated_at: datetime
created_at: datetime | None = None
updated_at: datetime | None = None
last_access_time: datetime | None = None
model_config = ConfigDict(from_attributes=True)
@ -32,16 +23,50 @@ class Asset(BaseModel):
return v.isoformat() if v else None
class AssetCreated(Asset):
created_new: bool
class AssetsList(BaseModel):
assets: list[Asset]
assets: list[AssetSummary]
total: int
has_more: bool
class AssetUpdated(BaseModel):
id: str
name: str
asset_hash: str | None = None
tags: list[str] = Field(default_factory=list)
user_metadata: dict[str, Any] = Field(default_factory=dict)
updated_at: datetime | None = None
model_config = ConfigDict(from_attributes=True)
@field_serializer("updated_at")
def _serialize_updated_at(self, v: datetime | None, _info):
return v.isoformat() if v else None
class AssetDetail(BaseModel):
id: str
name: str
asset_hash: str | None = None
size: int | None = None
mime_type: str | None = None
tags: list[str] = Field(default_factory=list)
user_metadata: dict[str, Any] = Field(default_factory=dict)
preview_id: str | None = None
created_at: datetime | None = None
last_access_time: datetime | None = None
model_config = ConfigDict(from_attributes=True)
@field_serializer("created_at", "last_access_time")
def _serialize_datetime(self, v: datetime | None, _info):
return v.isoformat() if v else None
class AssetCreated(AssetDetail):
created_new: bool
class TagUsage(BaseModel):
name: str
count: int
@ -66,7 +91,3 @@ class TagsRemove(BaseModel):
removed: list[str] = Field(default_factory=list)
not_present: list[str] = Field(default_factory=list)
total_tags: list[str] = Field(default_factory=list)
class TagHistogram(BaseModel):
tag_counts: dict[str, int]

View File

@ -52,8 +52,6 @@ async def parse_multipart_upload(
user_metadata_raw: str | None = None
provided_hash: str | None = None
provided_hash_exists: bool | None = None
provided_mime_type: str | None = None
provided_preview_id: str | None = None
file_written = 0
tmp_path: str | None = None
@ -130,16 +128,6 @@ async def parse_multipart_upload(
provided_name = (await field.text()) or None
elif fname == "user_metadata":
user_metadata_raw = (await field.text()) or None
elif fname == "id":
raise UploadError(
400,
"UNSUPPORTED_FIELD",
"Client-provided 'id' is not supported. Asset IDs are assigned by the server.",
)
elif fname == "mime_type":
provided_mime_type = ((await field.text()) or "").strip() or None
elif fname == "preview_id":
provided_preview_id = ((await field.text()) or "").strip() or None
if not file_present and not (provided_hash and provided_hash_exists):
raise UploadError(
@ -164,8 +152,6 @@ async def parse_multipart_upload(
user_metadata_raw=user_metadata_raw,
provided_hash=provided_hash,
provided_hash_exists=provided_hash_exists,
provided_mime_type=provided_mime_type,
provided_preview_id=provided_preview_id,
)

View File

@ -45,7 +45,13 @@ class Asset(Base):
passive_deletes=True,
)
# preview_id on AssetReference is a self-referential FK to asset_references.id
preview_of: Mapped[list[AssetReference]] = relationship(
"AssetReference",
back_populates="preview_asset",
primaryjoin=lambda: Asset.id == foreign(AssetReference.preview_id),
foreign_keys=lambda: [AssetReference.preview_id],
viewonly=True,
)
__table_args__ = (
Index("uq_assets_hash", "hash", unique=True),
@ -85,15 +91,11 @@ class AssetReference(Base):
owner_id: Mapped[str] = mapped_column(String(128), nullable=False, default="")
name: Mapped[str] = mapped_column(String(512), nullable=False)
preview_id: Mapped[str | None] = mapped_column(
String(36), ForeignKey("asset_references.id", ondelete="SET NULL")
String(36), ForeignKey("assets.id", ondelete="SET NULL")
)
user_metadata: Mapped[dict[str, Any] | None] = mapped_column(
JSON(none_as_null=True)
)
system_metadata: Mapped[dict[str, Any] | None] = mapped_column(
JSON(none_as_null=True), nullable=True, default=None
)
job_id: Mapped[str | None] = mapped_column(String(36), nullable=True, default=None)
created_at: Mapped[datetime] = mapped_column(
DateTime(timezone=False), nullable=False, default=get_utc_now
)
@ -113,10 +115,10 @@ class AssetReference(Base):
foreign_keys=[asset_id],
lazy="selectin",
)
preview_ref: Mapped[AssetReference | None] = relationship(
"AssetReference",
preview_asset: Mapped[Asset | None] = relationship(
"Asset",
back_populates="preview_of",
foreign_keys=[preview_id],
remote_side=lambda: [AssetReference.id],
)
metadata_entries: Mapped[list[AssetReferenceMeta]] = relationship(
@ -150,7 +152,6 @@ class AssetReference(Base):
Index("ix_asset_references_created_at", "created_at"),
Index("ix_asset_references_last_access_time", "last_access_time"),
Index("ix_asset_references_deleted_at", "deleted_at"),
Index("ix_asset_references_preview_id", "preview_id"),
Index("ix_asset_references_owner_name", "owner_id", "name"),
CheckConstraint(
"(mtime_ns IS NULL) OR (mtime_ns >= 0)", name="ck_ar_mtime_nonneg"
@ -191,10 +192,6 @@ class AssetReferenceMeta(Base):
Index("ix_asset_reference_meta_key_val_str", "key", "val_str"),
Index("ix_asset_reference_meta_key_val_num", "key", "val_num"),
Index("ix_asset_reference_meta_key_val_bool", "key", "val_bool"),
CheckConstraint(
"val_str IS NOT NULL OR val_num IS NOT NULL OR val_bool IS NOT NULL OR val_json IS NOT NULL",
name="has_value",
),
)

View File

@ -1,7 +1,6 @@
from app.assets.database.queries.asset import (
asset_exists_by_hash,
bulk_insert_assets,
create_stub_asset,
get_asset_by_hash,
get_existing_asset_ids,
reassign_asset_references,
@ -13,7 +12,6 @@ from app.assets.database.queries.asset_reference import (
UnenrichedReferenceRow,
bulk_insert_references_ignore_conflicts,
bulk_update_enrichment_level,
count_active_siblings,
bulk_update_is_missing,
bulk_update_needs_verify,
convert_metadata_to_rows,
@ -33,21 +31,16 @@ from app.assets.database.queries.asset_reference import (
get_unenriched_references,
get_unreferenced_unhashed_asset_ids,
insert_reference,
list_all_file_paths_by_asset_id,
list_references_by_asset_id,
list_references_page,
mark_references_missing_outside_prefixes,
rebuild_metadata_projection,
reference_exists,
reference_exists_for_asset_id,
restore_references_by_paths,
set_reference_metadata,
set_reference_preview,
set_reference_system_metadata,
soft_delete_reference_by_id,
update_reference_access_time,
update_reference_name,
update_is_missing_by_asset_id,
update_reference_timestamps,
update_reference_updated_at,
upsert_reference,
@ -61,7 +54,6 @@ from app.assets.database.queries.tags import (
bulk_insert_tags_and_meta,
ensure_tags_exist,
get_reference_tags,
list_tag_counts_for_filtered_assets,
list_tags_with_usage,
remove_missing_tag_for_asset_id,
remove_tags_from_reference,
@ -82,8 +74,6 @@ __all__ = [
"bulk_insert_references_ignore_conflicts",
"bulk_insert_tags_and_meta",
"bulk_update_enrichment_level",
"count_active_siblings",
"create_stub_asset",
"bulk_update_is_missing",
"bulk_update_needs_verify",
"convert_metadata_to_rows",
@ -107,26 +97,20 @@ __all__ = [
"get_unenriched_references",
"get_unreferenced_unhashed_asset_ids",
"insert_reference",
"list_all_file_paths_by_asset_id",
"list_references_by_asset_id",
"list_references_page",
"list_tag_counts_for_filtered_assets",
"list_tags_with_usage",
"mark_references_missing_outside_prefixes",
"reassign_asset_references",
"rebuild_metadata_projection",
"reference_exists",
"reference_exists_for_asset_id",
"remove_missing_tag_for_asset_id",
"remove_tags_from_reference",
"restore_references_by_paths",
"set_reference_metadata",
"set_reference_preview",
"set_reference_system_metadata",
"soft_delete_reference_by_id",
"set_reference_tags",
"update_asset_hash_and_mime",
"update_is_missing_by_asset_id",
"update_reference_access_time",
"update_reference_name",
"update_reference_timestamps",

View File

@ -69,7 +69,7 @@ def upsert_asset(
if asset.size_bytes != int(size_bytes) and int(size_bytes) > 0:
asset.size_bytes = int(size_bytes)
changed = True
if mime_type and not asset.mime_type:
if mime_type and asset.mime_type != mime_type:
asset.mime_type = mime_type
changed = True
if changed:
@ -78,18 +78,6 @@ def upsert_asset(
return asset, created, updated
def create_stub_asset(
session: Session,
size_bytes: int,
mime_type: str | None = None,
) -> Asset:
"""Create a new asset with no hash (stub for later enrichment)."""
asset = Asset(size_bytes=size_bytes, mime_type=mime_type, hash=None)
session.add(asset)
session.flush()
return asset
def bulk_insert_assets(
session: Session,
rows: list[dict],
@ -130,7 +118,7 @@ def update_asset_hash_and_mime(
return False
if asset_hash is not None:
asset.hash = asset_hash
if mime_type is not None and not asset.mime_type:
if mime_type is not None:
asset.mime_type = mime_type
return True

View File

@ -10,7 +10,7 @@ from decimal import Decimal
from typing import NamedTuple, Sequence
import sqlalchemy as sa
from sqlalchemy import delete, select
from sqlalchemy import delete, exists, select
from sqlalchemy.dialects import sqlite
from sqlalchemy.exc import IntegrityError
from sqlalchemy.orm import Session, noload
@ -24,14 +24,12 @@ from app.assets.database.models import (
)
from app.assets.database.queries.common import (
MAX_BIND_PARAMS,
apply_metadata_filter,
apply_tag_filters,
build_prefix_like_conditions,
build_visible_owner_clause,
calculate_rows_per_statement,
iter_chunks,
)
from app.assets.helpers import escape_sql_like_string, get_utc_now
from app.assets.helpers import escape_sql_like_string, get_utc_now, normalize_tags
def _check_is_scalar(v):
@ -46,6 +44,15 @@ def _check_is_scalar(v):
def _scalar_to_row(key: str, ordinal: int, value) -> dict:
"""Convert a scalar value to a typed projection row."""
if value is None:
return {
"key": key,
"ordinal": ordinal,
"val_str": None,
"val_num": None,
"val_bool": None,
"val_json": None,
}
if isinstance(value, bool):
return {"key": key, "ordinal": ordinal, "val_bool": bool(value)}
if isinstance(value, (int, float, Decimal)):
@ -59,19 +66,96 @@ def _scalar_to_row(key: str, ordinal: int, value) -> dict:
def convert_metadata_to_rows(key: str, value) -> list[dict]:
"""Turn a metadata key/value into typed projection rows."""
if value is None:
return []
return [_scalar_to_row(key, 0, None)]
if _check_is_scalar(value):
return [_scalar_to_row(key, 0, value)]
if isinstance(value, list):
if all(_check_is_scalar(x) for x in value):
return [_scalar_to_row(key, i, x) for i, x in enumerate(value) if x is not None]
return [{"key": key, "ordinal": i, "val_json": x} for i, x in enumerate(value) if x is not None]
return [_scalar_to_row(key, i, x) for i, x in enumerate(value)]
return [{"key": key, "ordinal": i, "val_json": x} for i, x in enumerate(value)]
return [{"key": key, "ordinal": 0, "val_json": value}]
def _apply_tag_filters(
stmt: sa.sql.Select,
include_tags: Sequence[str] | None = None,
exclude_tags: Sequence[str] | None = None,
) -> sa.sql.Select:
"""include_tags: every tag must be present; exclude_tags: none may be present."""
include_tags = normalize_tags(include_tags)
exclude_tags = normalize_tags(exclude_tags)
if include_tags:
for tag_name in include_tags:
stmt = stmt.where(
exists().where(
(AssetReferenceTag.asset_reference_id == AssetReference.id)
& (AssetReferenceTag.tag_name == tag_name)
)
)
if exclude_tags:
stmt = stmt.where(
~exists().where(
(AssetReferenceTag.asset_reference_id == AssetReference.id)
& (AssetReferenceTag.tag_name.in_(exclude_tags))
)
)
return stmt
def _apply_metadata_filter(
stmt: sa.sql.Select,
metadata_filter: dict | None = None,
) -> sa.sql.Select:
"""Apply filters using asset_reference_meta projection table."""
if not metadata_filter:
return stmt
def _exists_for_pred(key: str, *preds) -> sa.sql.ClauseElement:
return sa.exists().where(
AssetReferenceMeta.asset_reference_id == AssetReference.id,
AssetReferenceMeta.key == key,
*preds,
)
def _exists_clause_for_value(key: str, value) -> sa.sql.ClauseElement:
if value is None:
no_row_for_key = sa.not_(
sa.exists().where(
AssetReferenceMeta.asset_reference_id == AssetReference.id,
AssetReferenceMeta.key == key,
)
)
null_row = _exists_for_pred(
key,
AssetReferenceMeta.val_json.is_(None),
AssetReferenceMeta.val_str.is_(None),
AssetReferenceMeta.val_num.is_(None),
AssetReferenceMeta.val_bool.is_(None),
)
return sa.or_(no_row_for_key, null_row)
if isinstance(value, bool):
return _exists_for_pred(key, AssetReferenceMeta.val_bool == bool(value))
if isinstance(value, (int, float, Decimal)):
num = value if isinstance(value, Decimal) else Decimal(str(value))
return _exists_for_pred(key, AssetReferenceMeta.val_num == num)
if isinstance(value, str):
return _exists_for_pred(key, AssetReferenceMeta.val_str == value)
return _exists_for_pred(key, AssetReferenceMeta.val_json == value)
for k, v in metadata_filter.items():
if isinstance(v, list):
ors = [_exists_clause_for_value(k, elem) for elem in v]
if ors:
stmt = stmt.where(sa.or_(*ors))
else:
stmt = stmt.where(_exists_clause_for_value(k, v))
return stmt
def get_reference_by_id(
@ -114,23 +198,6 @@ def get_reference_by_file_path(
)
def count_active_siblings(
session: Session,
asset_id: str,
exclude_reference_id: str,
) -> int:
"""Count active (non-deleted) references to an asset, excluding one reference."""
return (
session.query(AssetReference)
.filter(
AssetReference.asset_id == asset_id,
AssetReference.id != exclude_reference_id,
AssetReference.deleted_at.is_(None),
)
.count()
)
def reference_exists_for_asset_id(
session: Session,
asset_id: str,
@ -145,21 +212,6 @@ def reference_exists_for_asset_id(
return session.execute(q).first() is not None
def reference_exists(
session: Session,
reference_id: str,
) -> bool:
"""Return True if a reference with the given ID exists (not soft-deleted)."""
q = (
select(sa.literal(True))
.select_from(AssetReference)
.where(AssetReference.id == reference_id)
.where(AssetReference.deleted_at.is_(None))
.limit(1)
)
return session.execute(q).first() is not None
def insert_reference(
session: Session,
asset_id: str,
@ -284,8 +336,8 @@ def list_references_page(
escaped, esc = escape_sql_like_string(name_contains)
base = base.where(AssetReference.name.ilike(f"%{escaped}%", escape=esc))
base = apply_tag_filters(base, include_tags, exclude_tags)
base = apply_metadata_filter(base, metadata_filter)
base = _apply_tag_filters(base, include_tags, exclude_tags)
base = _apply_metadata_filter(base, metadata_filter)
sort = (sort or "created_at").lower()
order = (order or "desc").lower()
@ -314,8 +366,8 @@ def list_references_page(
count_stmt = count_stmt.where(
AssetReference.name.ilike(f"%{escaped}%", escape=esc)
)
count_stmt = apply_tag_filters(count_stmt, include_tags, exclude_tags)
count_stmt = apply_metadata_filter(count_stmt, metadata_filter)
count_stmt = _apply_tag_filters(count_stmt, include_tags, exclude_tags)
count_stmt = _apply_metadata_filter(count_stmt, metadata_filter)
total = int(session.execute(count_stmt).scalar_one() or 0)
refs = session.execute(base).unique().scalars().all()
@ -327,7 +379,7 @@ def list_references_page(
select(AssetReferenceTag.asset_reference_id, Tag.name)
.join(Tag, Tag.name == AssetReferenceTag.tag_name)
.where(AssetReferenceTag.asset_reference_id.in_(id_list))
.order_by(AssetReferenceTag.tag_name.asc())
.order_by(AssetReferenceTag.added_at)
)
for ref_id, tag_name in rows.all():
tag_map[ref_id].append(tag_name)
@ -440,42 +492,6 @@ def update_reference_updated_at(
)
def rebuild_metadata_projection(session: Session, ref: AssetReference) -> None:
"""Delete and rebuild AssetReferenceMeta rows from merged system+user metadata.
The merged dict is ``{**system_metadata, **user_metadata}`` so user keys
override system keys of the same name.
"""
session.execute(
delete(AssetReferenceMeta).where(
AssetReferenceMeta.asset_reference_id == ref.id
)
)
session.flush()
merged = {**(ref.system_metadata or {}), **(ref.user_metadata or {})}
if not merged:
return
rows: list[AssetReferenceMeta] = []
for k, v in merged.items():
for r in convert_metadata_to_rows(k, v):
rows.append(
AssetReferenceMeta(
asset_reference_id=ref.id,
key=r["key"],
ordinal=int(r["ordinal"]),
val_str=r.get("val_str"),
val_num=r.get("val_num"),
val_bool=r.get("val_bool"),
val_json=r.get("val_json"),
)
)
if rows:
session.add_all(rows)
session.flush()
def set_reference_metadata(
session: Session,
reference_id: str,
@ -489,24 +505,33 @@ def set_reference_metadata(
ref.updated_at = get_utc_now()
session.flush()
rebuild_metadata_projection(session, ref)
def set_reference_system_metadata(
session: Session,
reference_id: str,
system_metadata: dict | None = None,
) -> None:
"""Set system_metadata on a reference and rebuild the merged projection."""
ref = session.get(AssetReference, reference_id)
if not ref:
raise ValueError(f"AssetReference {reference_id} not found")
ref.system_metadata = system_metadata or {}
ref.updated_at = get_utc_now()
session.execute(
delete(AssetReferenceMeta).where(
AssetReferenceMeta.asset_reference_id == reference_id
)
)
session.flush()
rebuild_metadata_projection(session, ref)
if not user_metadata:
return
rows: list[AssetReferenceMeta] = []
for k, v in user_metadata.items():
for r in convert_metadata_to_rows(k, v):
rows.append(
AssetReferenceMeta(
asset_reference_id=reference_id,
key=r["key"],
ordinal=int(r["ordinal"]),
val_str=r.get("val_str"),
val_num=r.get("val_num"),
val_bool=r.get("val_bool"),
val_json=r.get("val_json"),
)
)
if rows:
session.add_all(rows)
session.flush()
def delete_reference_by_id(
@ -546,19 +571,19 @@ def soft_delete_reference_by_id(
def set_reference_preview(
session: Session,
reference_id: str,
preview_reference_id: str | None = None,
preview_asset_id: str | None = None,
) -> None:
"""Set or clear preview_id and bump updated_at. Raises on unknown IDs."""
ref = session.get(AssetReference, reference_id)
if not ref:
raise ValueError(f"AssetReference {reference_id} not found")
if preview_reference_id is None:
if preview_asset_id is None:
ref.preview_id = None
else:
if not session.get(AssetReference, preview_reference_id):
raise ValueError(f"Preview AssetReference {preview_reference_id} not found")
ref.preview_id = preview_reference_id
if not session.get(Asset, preview_asset_id):
raise ValueError(f"Preview Asset {preview_asset_id} not found")
ref.preview_id = preview_asset_id
ref.updated_at = get_utc_now()
session.flush()
@ -584,8 +609,6 @@ def list_references_by_asset_id(
session.execute(
select(AssetReference)
.where(AssetReference.asset_id == asset_id)
.where(AssetReference.is_missing == False) # noqa: E712
.where(AssetReference.deleted_at.is_(None))
.order_by(AssetReference.id.asc())
)
.scalars()
@ -593,25 +616,6 @@ def list_references_by_asset_id(
)
def list_all_file_paths_by_asset_id(
session: Session,
asset_id: str,
) -> list[str]:
"""Return every file_path for an asset, including soft-deleted/missing refs.
Used for orphan cleanup where all on-disk files must be removed.
"""
return list(
session.execute(
select(AssetReference.file_path)
.where(AssetReference.asset_id == asset_id)
.where(AssetReference.file_path.isnot(None))
)
.scalars()
.all()
)
def upsert_reference(
session: Session,
asset_id: str,
@ -851,22 +855,6 @@ def bulk_update_is_missing(
return total
def update_is_missing_by_asset_id(
session: Session, asset_id: str, value: bool
) -> int:
"""Set is_missing flag for ALL references belonging to an asset.
Returns: Number of rows updated
"""
result = session.execute(
sa.update(AssetReference)
.where(AssetReference.asset_id == asset_id)
.where(AssetReference.deleted_at.is_(None))
.values(is_missing=value)
)
return result.rowcount
def delete_references_by_ids(session: Session, reference_ids: list[str]) -> int:
"""Delete references by their IDs.

View File

@ -1,14 +1,12 @@
"""Shared utilities for database query modules."""
import os
from decimal import Decimal
from typing import Iterable, Sequence
from typing import Iterable
import sqlalchemy as sa
from sqlalchemy import exists
from app.assets.database.models import AssetReference, AssetReferenceMeta, AssetReferenceTag
from app.assets.helpers import escape_sql_like_string, normalize_tags
from app.assets.database.models import AssetReference
from app.assets.helpers import escape_sql_like_string
MAX_BIND_PARAMS = 800
@ -54,74 +52,3 @@ def build_prefix_like_conditions(
escaped, esc = escape_sql_like_string(base)
conds.append(AssetReference.file_path.like(escaped + "%", escape=esc))
return conds
def apply_tag_filters(
stmt: sa.sql.Select,
include_tags: Sequence[str] | None = None,
exclude_tags: Sequence[str] | None = None,
) -> sa.sql.Select:
"""include_tags: every tag must be present; exclude_tags: none may be present."""
include_tags = normalize_tags(include_tags)
exclude_tags = normalize_tags(exclude_tags)
if include_tags:
for tag_name in include_tags:
stmt = stmt.where(
exists().where(
(AssetReferenceTag.asset_reference_id == AssetReference.id)
& (AssetReferenceTag.tag_name == tag_name)
)
)
if exclude_tags:
stmt = stmt.where(
~exists().where(
(AssetReferenceTag.asset_reference_id == AssetReference.id)
& (AssetReferenceTag.tag_name.in_(exclude_tags))
)
)
return stmt
def apply_metadata_filter(
stmt: sa.sql.Select,
metadata_filter: dict | None = None,
) -> sa.sql.Select:
"""Apply filters using asset_reference_meta projection table."""
if not metadata_filter:
return stmt
def _exists_for_pred(key: str, *preds) -> sa.sql.ClauseElement:
return sa.exists().where(
AssetReferenceMeta.asset_reference_id == AssetReference.id,
AssetReferenceMeta.key == key,
*preds,
)
def _exists_clause_for_value(key: str, value) -> sa.sql.ClauseElement:
if value is None:
return sa.not_(
sa.exists().where(
AssetReferenceMeta.asset_reference_id == AssetReference.id,
AssetReferenceMeta.key == key,
)
)
if isinstance(value, bool):
return _exists_for_pred(key, AssetReferenceMeta.val_bool == bool(value))
if isinstance(value, (int, float, Decimal)):
num = value if isinstance(value, Decimal) else Decimal(str(value))
return _exists_for_pred(key, AssetReferenceMeta.val_num == num)
if isinstance(value, str):
return _exists_for_pred(key, AssetReferenceMeta.val_str == value)
return _exists_for_pred(key, AssetReferenceMeta.val_json == value)
for k, v in metadata_filter.items():
if isinstance(v, list):
ors = [_exists_clause_for_value(k, elem) for elem in v]
if ors:
stmt = stmt.where(sa.or_(*ors))
else:
stmt = stmt.where(_exists_clause_for_value(k, v))
return stmt

View File

@ -8,15 +8,12 @@ from sqlalchemy.exc import IntegrityError
from sqlalchemy.orm import Session
from app.assets.database.models import (
Asset,
AssetReference,
AssetReferenceMeta,
AssetReferenceTag,
Tag,
)
from app.assets.database.queries.common import (
apply_metadata_filter,
apply_tag_filters,
build_visible_owner_clause,
iter_row_chunks,
)
@ -75,9 +72,9 @@ def get_reference_tags(session: Session, reference_id: str) -> list[str]:
tag_name
for (tag_name,) in (
session.execute(
select(AssetReferenceTag.tag_name)
.where(AssetReferenceTag.asset_reference_id == reference_id)
.order_by(AssetReferenceTag.tag_name.asc())
select(AssetReferenceTag.tag_name).where(
AssetReferenceTag.asset_reference_id == reference_id
)
)
).all()
]
@ -120,7 +117,7 @@ def set_reference_tags(
)
session.flush()
return SetTagsResult(added=sorted(to_add), removed=sorted(to_remove), total=sorted(desired))
return SetTagsResult(added=to_add, removed=to_remove, total=desired)
def add_tags_to_reference(
@ -275,12 +272,6 @@ def list_tags_with_usage(
.select_from(AssetReferenceTag)
.join(AssetReference, AssetReference.id == AssetReferenceTag.asset_reference_id)
.where(build_visible_owner_clause(owner_id))
.where(
sa.or_(
AssetReference.is_missing == False, # noqa: E712
AssetReferenceTag.tag_name == "missing",
)
)
.where(AssetReference.deleted_at.is_(None))
.group_by(AssetReferenceTag.tag_name)
.subquery()
@ -317,12 +308,6 @@ def list_tags_with_usage(
select(AssetReferenceTag.tag_name)
.join(AssetReference, AssetReference.id == AssetReferenceTag.asset_reference_id)
.where(build_visible_owner_clause(owner_id))
.where(
sa.or_(
AssetReference.is_missing == False, # noqa: E712
AssetReferenceTag.tag_name == "missing",
)
)
.where(AssetReference.deleted_at.is_(None))
.group_by(AssetReferenceTag.tag_name)
)
@ -335,53 +320,6 @@ def list_tags_with_usage(
return rows_norm, int(total or 0)
def list_tag_counts_for_filtered_assets(
session: Session,
owner_id: str = "",
include_tags: Sequence[str] | None = None,
exclude_tags: Sequence[str] | None = None,
name_contains: str | None = None,
metadata_filter: dict | None = None,
limit: int = 100,
) -> dict[str, int]:
"""Return tag counts for assets matching the given filters.
Uses the same filtering logic as list_references_page but returns
{tag_name: count} instead of paginated references.
"""
# Build a subquery of matching reference IDs
ref_sq = (
select(AssetReference.id)
.join(Asset, Asset.id == AssetReference.asset_id)
.where(build_visible_owner_clause(owner_id))
.where(AssetReference.is_missing == False) # noqa: E712
.where(AssetReference.deleted_at.is_(None))
)
if name_contains:
escaped, esc = escape_sql_like_string(name_contains)
ref_sq = ref_sq.where(AssetReference.name.ilike(f"%{escaped}%", escape=esc))
ref_sq = apply_tag_filters(ref_sq, include_tags, exclude_tags)
ref_sq = apply_metadata_filter(ref_sq, metadata_filter)
ref_sq = ref_sq.subquery()
# Count tags across those references
q = (
select(
AssetReferenceTag.tag_name,
func.count(AssetReferenceTag.asset_reference_id).label("cnt"),
)
.where(AssetReferenceTag.asset_reference_id.in_(select(ref_sq.c.id)))
.group_by(AssetReferenceTag.tag_name)
.order_by(func.count(AssetReferenceTag.asset_reference_id).desc(), AssetReferenceTag.tag_name.asc())
.limit(limit)
)
rows = session.execute(q).all()
return {tag_name: int(cnt) for tag_name, cnt in rows}
def bulk_insert_tags_and_meta(
session: Session,
tag_rows: list[dict],

View File

@ -13,13 +13,12 @@ from app.assets.database.queries import (
delete_references_by_ids,
ensure_tags_exist,
get_asset_by_hash,
get_reference_by_id,
get_references_for_prefixes,
get_unenriched_references,
mark_references_missing_outside_prefixes,
reassign_asset_references,
remove_missing_tag_for_asset_id,
set_reference_system_metadata,
set_reference_metadata,
update_asset_hash_and_mime,
)
from app.assets.services.bulk_ingest import (
@ -339,7 +338,6 @@ def build_asset_specs(
"metadata": metadata,
"hash": asset_hash,
"mime_type": mime_type,
"job_id": None,
}
)
tag_pool.update(tags)
@ -428,7 +426,6 @@ def enrich_asset(
except OSError:
return new_level
initial_mtime_ns = get_mtime_ns(stat_p)
rel_fname = compute_relative_filename(file_path)
mime_type: str | None = None
metadata = None
@ -492,21 +489,9 @@ def enrich_asset(
except Exception as e:
logging.warning("Failed to hash %s: %s", file_path, e)
# Optimistic guard: if the reference's mtime_ns changed since we
# started (e.g. ingest_existing_file updated it), our results are
# stale — discard them to avoid overwriting fresh registration data.
ref = get_reference_by_id(session, reference_id)
if ref is None or ref.mtime_ns != initial_mtime_ns:
session.rollback()
logging.info(
"Ref %s mtime changed during enrichment, discarding stale result",
reference_id,
)
return ENRICHMENT_STUB
if extract_metadata and metadata:
system_metadata = metadata.to_user_metadata()
set_reference_system_metadata(session, reference_id, system_metadata)
user_metadata = metadata.to_user_metadata()
set_reference_metadata(session, reference_id, user_metadata)
if full_hash:
existing = get_asset_by_hash(session, full_hash)

View File

@ -77,9 +77,7 @@ class _AssetSeeder:
"""
def __init__(self) -> None:
# RLock is required because _run_scan() drains pending work while
# holding _lock and re-enters start() which also acquires _lock.
self._lock = threading.RLock()
self._lock = threading.Lock()
self._state = State.IDLE
self._progress: Progress | None = None
self._last_progress: Progress | None = None
@ -94,7 +92,6 @@ class _AssetSeeder:
self._prune_first: bool = False
self._progress_callback: ProgressCallback | None = None
self._disabled: bool = False
self._pending_enrich: dict | None = None
def disable(self) -> None:
"""Disable the asset seeder, preventing any scans from starting."""
@ -199,42 +196,6 @@ class _AssetSeeder:
compute_hashes=compute_hashes,
)
def enqueue_enrich(
self,
roots: tuple[RootType, ...] = ("models", "input", "output"),
compute_hashes: bool = False,
) -> bool:
"""Start an enrichment scan now, or queue it for after the current scan.
If the seeder is idle, starts immediately. Otherwise, the enrich
request is stored and will run automatically when the current scan
finishes.
Args:
roots: Tuple of root types to scan
compute_hashes: If True, compute blake3 hashes
Returns:
True if started immediately, False if queued for later
"""
with self._lock:
if self.start_enrich(roots=roots, compute_hashes=compute_hashes):
return True
if self._pending_enrich is not None:
existing_roots = set(self._pending_enrich["roots"])
existing_roots.update(roots)
self._pending_enrich["roots"] = tuple(existing_roots)
self._pending_enrich["compute_hashes"] = (
self._pending_enrich["compute_hashes"] or compute_hashes
)
else:
self._pending_enrich = {
"roots": roots,
"compute_hashes": compute_hashes,
}
logging.info("Enrich scan queued (roots=%s)", self._pending_enrich["roots"])
return False
def cancel(self) -> bool:
"""Request cancellation of the current scan.
@ -420,13 +381,9 @@ class _AssetSeeder:
return marked
finally:
with self._lock:
self._reset_to_idle()
def _reset_to_idle(self) -> None:
"""Reset state to IDLE, preserving last progress. Caller must hold _lock."""
self._last_progress = self._progress
self._state = State.IDLE
self._progress = None
self._last_progress = self._progress
self._state = State.IDLE
self._progress = None
def _is_cancelled(self) -> bool:
"""Check if cancellation has been requested."""
@ -637,18 +594,9 @@ class _AssetSeeder:
},
)
with self._lock:
self._reset_to_idle()
pending = self._pending_enrich
if pending is not None:
self._pending_enrich = None
if not self.start_enrich(
roots=pending["roots"],
compute_hashes=pending["compute_hashes"],
):
logging.warning(
"Pending enrich scan could not start (roots=%s)",
pending["roots"],
)
self._last_progress = self._progress
self._state = State.IDLE
self._progress = None
def _run_fast_phase(self, roots: tuple[RootType, ...]) -> tuple[int, int, int]:
"""Run phase 1: fast scan to create stub records.

View File

@ -23,8 +23,6 @@ from app.assets.services.ingest import (
DependencyMissingError,
HashMismatchError,
create_from_hash,
ingest_existing_file,
register_output_files,
upload_from_temp_path,
)
from app.assets.database.queries import (
@ -74,8 +72,6 @@ __all__ = [
"delete_asset_reference",
"get_asset_by_hash",
"get_asset_detail",
"ingest_existing_file",
"register_output_files",
"get_mtime_ns",
"get_size_and_mtime_ns",
"list_assets_page",

View File

@ -16,12 +16,10 @@ from app.assets.database.queries import (
get_reference_by_id,
get_reference_with_owner_check,
list_references_page,
list_all_file_paths_by_asset_id,
list_references_by_asset_id,
set_reference_metadata,
set_reference_preview,
set_reference_tags,
update_asset_hash_and_mime,
update_reference_access_time,
update_reference_name,
update_reference_updated_at,
@ -69,8 +67,6 @@ def update_asset_metadata(
user_metadata: UserMetadata = None,
tag_origin: str = "manual",
owner_id: str = "",
mime_type: str | None = None,
preview_id: str | None = None,
) -> AssetDetailResult:
with create_session() as session:
ref = get_reference_with_owner_check(session, reference_id, owner_id)
@ -107,21 +103,6 @@ def update_asset_metadata(
)
touched = True
if mime_type is not None:
updated = update_asset_hash_and_mime(
session, asset_id=ref.asset_id, mime_type=mime_type
)
if updated:
touched = True
if preview_id is not None:
set_reference_preview(
session,
reference_id=reference_id,
preview_reference_id=preview_id,
)
touched = True
if touched and user_metadata is None:
update_reference_updated_at(session, reference_id=reference_id)
@ -178,9 +159,11 @@ def delete_asset_reference(
session.commit()
return True
# Orphaned asset - gather ALL file paths (including
# soft-deleted / missing refs) so their on-disk files get cleaned up.
file_paths = list_all_file_paths_by_asset_id(session, asset_id=asset_id)
# Orphaned asset - delete it and its files
refs = list_references_by_asset_id(session, asset_id=asset_id)
file_paths = [
r.file_path for r in (refs or []) if getattr(r, "file_path", None)
]
# Also include the just-deleted file path
if file_path:
file_paths.append(file_path)
@ -202,7 +185,7 @@ def delete_asset_reference(
def set_asset_preview(
reference_id: str,
preview_reference_id: str | None = None,
preview_asset_id: str | None = None,
owner_id: str = "",
) -> AssetDetailResult:
with create_session() as session:
@ -211,7 +194,7 @@ def set_asset_preview(
set_reference_preview(
session,
reference_id=reference_id,
preview_reference_id=preview_reference_id,
preview_asset_id=preview_asset_id,
)
result = fetch_reference_asset_and_tags(
@ -280,47 +263,6 @@ def list_assets_page(
return ListAssetsResult(items=items, total=total)
def resolve_hash_to_path(
asset_hash: str,
owner_id: str = "",
) -> DownloadResolutionResult | None:
"""Resolve a blake3 hash to an on-disk file path.
Only references visible to *owner_id* are considered (owner-less
references are always visible).
Returns a DownloadResolutionResult with abs_path, content_type, and
download_name, or None if no asset or live path is found.
"""
with create_session() as session:
asset = queries_get_asset_by_hash(session, asset_hash)
if not asset:
return None
refs = list_references_by_asset_id(session, asset_id=asset.id)
visible = [
r for r in refs
if r.owner_id == "" or r.owner_id == owner_id
]
abs_path = select_best_live_path(visible)
if not abs_path:
return None
display_name = os.path.basename(abs_path)
for ref in visible:
if ref.file_path == abs_path and ref.name:
display_name = ref.name
break
ctype = (
asset.mime_type
or mimetypes.guess_type(display_name)[0]
or "application/octet-stream"
)
return DownloadResolutionResult(
abs_path=abs_path,
content_type=ctype,
download_name=display_name,
)
def resolve_asset_for_download(
reference_id: str,
owner_id: str = "",

View File

@ -37,7 +37,6 @@ class SeedAssetSpec(TypedDict):
metadata: ExtractedMetadata | None
hash: str | None
mime_type: str | None
job_id: str | None
class AssetRow(TypedDict):
@ -61,7 +60,6 @@ class ReferenceRow(TypedDict):
name: str
preview_id: str | None
user_metadata: dict[str, Any] | None
job_id: str | None
created_at: datetime
updated_at: datetime
last_access_time: datetime
@ -169,7 +167,6 @@ def batch_insert_seed_assets(
"name": spec["info_name"],
"preview_id": None,
"user_metadata": user_metadata,
"job_id": spec.get("job_id"),
"created_at": current_time,
"updated_at": current_time,
"last_access_time": current_time,

View File

@ -9,29 +9,23 @@ from sqlalchemy.orm import Session
import app.assets.services.hashing as hashing
from app.assets.database.queries import (
add_tags_to_reference,
count_active_siblings,
create_stub_asset,
ensure_tags_exist,
fetch_reference_and_asset,
get_asset_by_hash,
get_existing_asset_ids,
get_reference_by_file_path,
get_reference_tags,
get_or_create_reference,
reference_exists,
remove_missing_tag_for_asset_id,
set_reference_metadata,
set_reference_tags,
update_asset_hash_and_mime,
upsert_asset,
upsert_reference,
validate_tags_exist,
)
from app.assets.helpers import get_utc_now, normalize_tags
from app.assets.services.bulk_ingest import batch_insert_seed_assets
from app.assets.helpers import normalize_tags
from app.assets.services.file_utils import get_size_and_mtime_ns
from app.assets.services.path_utils import (
compute_relative_filename,
get_name_and_tags_from_asset_path,
resolve_destination_from_tags,
validate_path_within_base,
)
@ -71,7 +65,7 @@ def _ingest_file_from_path(
with create_session() as session:
if preview_id:
if not reference_exists(session, preview_id):
if preview_id not in get_existing_asset_ids(session, [preview_id]):
preview_id = None
asset, asset_created, asset_updated = upsert_asset(
@ -134,102 +128,6 @@ def _ingest_file_from_path(
)
def register_output_files(
file_paths: Sequence[str],
user_metadata: UserMetadata = None,
job_id: str | None = None,
) -> int:
"""Register a batch of output file paths as assets.
Returns the number of files successfully registered.
"""
registered = 0
for abs_path in file_paths:
if not os.path.isfile(abs_path):
continue
try:
if ingest_existing_file(
abs_path, user_metadata=user_metadata, job_id=job_id
):
registered += 1
except Exception:
logging.exception("Failed to register output: %s", abs_path)
return registered
def ingest_existing_file(
abs_path: str,
user_metadata: UserMetadata = None,
extra_tags: Sequence[str] = (),
owner_id: str = "",
job_id: str | None = None,
) -> bool:
"""Register an existing on-disk file as an asset stub.
If a reference already exists for this path, updates mtime_ns, job_id,
size_bytes, and resets enrichment so the enricher will re-hash it.
For brand-new paths, inserts a stub record (hash=NULL) for immediate
UX visibility.
Returns True if a row was inserted or updated, False otherwise.
"""
locator = os.path.abspath(abs_path)
size_bytes, mtime_ns = get_size_and_mtime_ns(abs_path)
mime_type = mimetypes.guess_type(abs_path, strict=False)[0]
name, path_tags = get_name_and_tags_from_asset_path(abs_path)
tags = list(dict.fromkeys(path_tags + list(extra_tags)))
with create_session() as session:
existing_ref = get_reference_by_file_path(session, locator)
if existing_ref is not None:
now = get_utc_now()
existing_ref.mtime_ns = mtime_ns
existing_ref.job_id = job_id
existing_ref.is_missing = False
existing_ref.deleted_at = None
existing_ref.updated_at = now
existing_ref.enrichment_level = 0
asset = existing_ref.asset
if asset:
# If other refs share this asset, detach to a new stub
# instead of mutating the shared row.
siblings = count_active_siblings(session, asset.id, existing_ref.id)
if siblings > 0:
new_asset = create_stub_asset(
session,
size_bytes=size_bytes,
mime_type=mime_type or asset.mime_type,
)
existing_ref.asset_id = new_asset.id
else:
asset.hash = None
asset.size_bytes = size_bytes
if mime_type:
asset.mime_type = mime_type
session.commit()
return True
spec = {
"abs_path": abs_path,
"size_bytes": size_bytes,
"mtime_ns": mtime_ns,
"info_name": name,
"tags": tags,
"fname": os.path.basename(abs_path),
"metadata": None,
"hash": None,
"mime_type": mime_type,
"job_id": job_id,
}
if tags:
ensure_tags_exist(session, tags)
result = batch_insert_seed_assets(session, [spec], owner_id=owner_id)
session.commit()
return result.won_paths > 0
def _register_existing_asset(
asset_hash: str,
name: str,
@ -237,8 +135,6 @@ def _register_existing_asset(
tags: list[str] | None = None,
tag_origin: str = "manual",
owner_id: str = "",
mime_type: str | None = None,
preview_id: str | None = None,
) -> RegisterAssetResult:
user_metadata = user_metadata or {}
@ -247,25 +143,14 @@ def _register_existing_asset(
if not asset:
raise ValueError(f"No asset with hash {asset_hash}")
if mime_type and not asset.mime_type:
update_asset_hash_and_mime(session, asset_id=asset.id, mime_type=mime_type)
if preview_id:
if not reference_exists(session, preview_id):
preview_id = None
ref, ref_created = get_or_create_reference(
session,
asset_id=asset.id,
owner_id=owner_id,
name=name,
preview_id=preview_id,
)
if not ref_created:
if preview_id and ref.preview_id != preview_id:
ref.preview_id = preview_id
tag_names = get_reference_tags(session, reference_id=ref.id)
result = RegisterAssetResult(
ref=extract_reference_data(ref),
@ -357,8 +242,6 @@ def upload_from_temp_path(
client_filename: str | None = None,
owner_id: str = "",
expected_hash: str | None = None,
mime_type: str | None = None,
preview_id: str | None = None,
) -> UploadResult:
try:
digest, _ = hashing.compute_blake3_hash(temp_path)
@ -387,8 +270,6 @@ def upload_from_temp_path(
tags=tags or [],
tag_origin="manual",
owner_id=owner_id,
mime_type=mime_type,
preview_id=preview_id,
)
return UploadResult(
ref=result.ref,
@ -410,7 +291,7 @@ def upload_from_temp_path(
dest_abs = os.path.abspath(os.path.join(dest_dir, hashed_basename))
validate_path_within_base(dest_abs, base_dir)
content_type = mime_type or (
content_type = (
mimetypes.guess_type(os.path.basename(src_for_ext), strict=False)[0]
or mimetypes.guess_type(hashed_basename, strict=False)[0]
or "application/octet-stream"
@ -434,7 +315,7 @@ def upload_from_temp_path(
mime_type=content_type,
info_name=_sanitize_filename(name or client_filename, fallback=digest),
owner_id=owner_id,
preview_id=preview_id,
preview_id=None,
user_metadata=user_metadata or {},
tags=tags,
tag_origin="manual",
@ -461,99 +342,30 @@ def upload_from_temp_path(
)
def register_file_in_place(
abs_path: str,
name: str,
tags: list[str],
owner_id: str = "",
mime_type: str | None = None,
) -> UploadResult:
"""Register an already-saved file in the asset database without moving it.
Tags are derived from the filesystem path (root category + subfolder names),
merged with any caller-provided tags, matching the behavior of the scanner.
If the path is not under a known root, only the caller-provided tags are used.
"""
try:
_, path_tags = get_name_and_tags_from_asset_path(abs_path)
except ValueError:
path_tags = []
merged_tags = normalize_tags([*path_tags, *tags])
try:
digest, _ = hashing.compute_blake3_hash(abs_path)
except ImportError as e:
raise DependencyMissingError(str(e))
except Exception as e:
raise RuntimeError(f"failed to hash file: {e}")
asset_hash = "blake3:" + digest
size_bytes, mtime_ns = get_size_and_mtime_ns(abs_path)
content_type = mime_type or (
mimetypes.guess_type(abs_path, strict=False)[0]
or "application/octet-stream"
)
ingest_result = _ingest_file_from_path(
abs_path=abs_path,
asset_hash=asset_hash,
size_bytes=size_bytes,
mtime_ns=mtime_ns,
mime_type=content_type,
info_name=_sanitize_filename(name, fallback=digest),
owner_id=owner_id,
tags=merged_tags,
tag_origin="upload",
require_existing_tags=False,
)
reference_id = ingest_result.reference_id
if not reference_id:
raise RuntimeError("failed to create asset reference")
with create_session() as session:
pair = fetch_reference_and_asset(
session, reference_id=reference_id, owner_id=owner_id
)
if not pair:
raise RuntimeError("inconsistent DB state after ingest")
ref, asset = pair
tag_names = get_reference_tags(session, reference_id=ref.id)
return UploadResult(
ref=extract_reference_data(ref),
asset=extract_asset_data(asset),
tags=tag_names,
created_new=ingest_result.asset_created,
)
def create_from_hash(
hash_str: str,
name: str,
tags: list[str] | None = None,
user_metadata: dict | None = None,
owner_id: str = "",
mime_type: str | None = None,
preview_id: str | None = None,
) -> UploadResult | None:
canonical = hash_str.strip().lower()
try:
result = _register_existing_asset(
asset_hash=canonical,
name=_sanitize_filename(
name, fallback=canonical.split(":", 1)[1] if ":" in canonical else canonical
),
user_metadata=user_metadata or {},
tags=tags or [],
tag_origin="manual",
owner_id=owner_id,
mime_type=mime_type,
preview_id=preview_id,
)
except ValueError:
logging.warning("create_from_hash: no asset found for hash %s", canonical)
return None
with create_session() as session:
asset = get_asset_by_hash(session, asset_hash=canonical)
if not asset:
return None
result = _register_existing_asset(
asset_hash=canonical,
name=_sanitize_filename(
name, fallback=canonical.split(":", 1)[1] if ":" in canonical else canonical
),
user_metadata=user_metadata or {},
tags=tags or [],
tag_origin="manual",
owner_id=owner_id,
)
return UploadResult(
ref=result.ref,

View File

@ -93,13 +93,12 @@ def compute_relative_filename(file_path: str) -> str | None:
def get_asset_category_and_relative_path(
file_path: str,
) -> tuple[Literal["input", "output", "temp", "models"], str]:
) -> tuple[Literal["input", "output", "models"], str]:
"""Determine which root category a file path belongs to.
Categories:
- 'input': under folder_paths.get_input_directory()
- 'output': under folder_paths.get_output_directory()
- 'temp': under folder_paths.get_temp_directory()
- 'models': under any base path from get_comfy_models_folders()
Returns:
@ -130,12 +129,7 @@ def get_asset_category_and_relative_path(
if _check_is_within(fp_abs, output_base):
return "output", _compute_relative(fp_abs, output_base)
# 3) temp
temp_base = os.path.abspath(folder_paths.get_temp_directory())
if _check_is_within(fp_abs, temp_base):
return "temp", _compute_relative(fp_abs, temp_base)
# 4) models (check deepest matching base to avoid ambiguity)
# 3) models (check deepest matching base to avoid ambiguity)
best: tuple[int, str, str] | None = None # (base_len, bucket, rel_inside_bucket)
for bucket, bases in get_comfy_models_folders():
for b in bases:
@ -152,7 +146,7 @@ def get_asset_category_and_relative_path(
return "models", os.path.relpath(os.path.join(os.sep, combined), os.sep)
raise ValueError(
f"Path is not within input, output, temp, or configured model bases: {file_path}"
f"Path is not within input, output, or configured model bases: {file_path}"
)

View File

@ -25,9 +25,7 @@ class ReferenceData:
preview_id: str | None
created_at: datetime
updated_at: datetime
system_metadata: dict[str, Any] | None = None
job_id: str | None = None
last_access_time: datetime | None = None
last_access_time: datetime | None
@dataclass(frozen=True)
@ -95,8 +93,6 @@ def extract_reference_data(ref: AssetReference) -> ReferenceData:
file_path=ref.file_path,
user_metadata=ref.user_metadata,
preview_id=ref.preview_id,
system_metadata=ref.system_metadata,
job_id=ref.job_id,
created_at=ref.created_at,
updated_at=ref.updated_at,
last_access_time=ref.last_access_time,

View File

@ -1,5 +1,3 @@
from typing import Sequence
from app.assets.database.queries import (
AddTagsResult,
RemoveTagsResult,
@ -8,7 +6,6 @@ from app.assets.database.queries import (
list_tags_with_usage,
remove_tags_from_reference,
)
from app.assets.database.queries.tags import list_tag_counts_for_filtered_assets
from app.assets.services.schemas import TagUsage
from app.database.db import create_session
@ -76,23 +73,3 @@ def list_tags(
)
return [TagUsage(name, tag_type, count) for name, tag_type, count in rows], total
def list_tag_histogram(
owner_id: str = "",
include_tags: Sequence[str] | None = None,
exclude_tags: Sequence[str] | None = None,
name_contains: str | None = None,
metadata_filter: dict | None = None,
limit: int = 100,
) -> dict[str, int]:
with create_session() as session:
return list_tag_counts_for_filtered_assets(
session,
owner_id=owner_id,
include_tags=include_tags,
exclude_tags=exclude_tags,
name_contains=name_contains,
metadata_filter=metadata_filter,
limit=limit,
)

View File

@ -1,18 +1,9 @@
from typing import Any
from datetime import datetime
from sqlalchemy import MetaData
from sqlalchemy.orm import DeclarativeBase
NAMING_CONVENTION = {
"ix": "ix_%(table_name)s_%(column_0_N_name)s",
"uq": "uq_%(table_name)s_%(column_0_N_name)s",
"ck": "ck_%(table_name)s_%(constraint_name)s",
"fk": "fk_%(table_name)s_%(column_0_name)s_%(referred_table_name)s",
"pk": "pk_%(table_name)s",
}
class Base(DeclarativeBase):
metadata = MetaData(naming_convention=NAMING_CONVENTION)
pass
def to_dict(obj: Any, include_none: bool = False) -> dict[str, Any]:
fields = obj.__table__.columns.keys()

View File

@ -6,7 +6,6 @@ import uuid
import glob
import shutil
import logging
import tempfile
from aiohttp import web
from urllib import parse
from comfy.cli_args import args
@ -378,15 +377,8 @@ class UserManager():
try:
body = await request.read()
dir_name = os.path.dirname(path)
fd, tmp_path = tempfile.mkstemp(dir=dir_name)
try:
with os.fdopen(fd, "wb") as f:
f.write(body)
os.replace(tmp_path, path)
except:
os.unlink(tmp_path)
raise
with open(path, "wb") as f:
f.write(body)
except OSError as e:
logging.warning(f"Error saving file '{path}': {e}")
return web.Response(

View File

@ -1,90 +0,0 @@
#version 300 es
precision highp float;
uniform sampler2D u_image0;
uniform float u_float0;
uniform float u_float1;
uniform float u_float2;
uniform float u_float3;
uniform float u_float4;
uniform float u_float5;
uniform float u_float6;
uniform float u_float7;
uniform float u_float8;
uniform bool u_bool0;
in vec2 v_texCoord;
out vec4 fragColor;
vec3 rgb2hsl(vec3 c) {
float maxC = max(c.r, max(c.g, c.b));
float minC = min(c.r, min(c.g, c.b));
float l = (maxC + minC) * 0.5;
if (maxC == minC) return vec3(0.0, 0.0, l);
float d = maxC - minC;
float s = l > 0.5 ? d / (2.0 - maxC - minC) : d / (maxC + minC);
float h;
if (maxC == c.r) {
h = (c.g - c.b) / d + (c.g < c.b ? 6.0 : 0.0);
} else if (maxC == c.g) {
h = (c.b - c.r) / d + 2.0;
} else {
h = (c.r - c.g) / d + 4.0;
}
h /= 6.0;
return vec3(h, s, l);
}
float hue2rgb(float p, float q, float t) {
if (t < 0.0) t += 1.0;
if (t > 1.0) t -= 1.0;
if (t < 1.0 / 6.0) return p + (q - p) * 6.0 * t;
if (t < 1.0 / 2.0) return q;
if (t < 2.0 / 3.0) return p + (q - p) * (2.0 / 3.0 - t) * 6.0;
return p;
}
vec3 hsl2rgb(vec3 hsl) {
float h = hsl.x, s = hsl.y, l = hsl.z;
if (s == 0.0) return vec3(l);
float q = l < 0.5 ? l * (1.0 + s) : l + s - l * s;
float p = 2.0 * l - q;
return vec3(
hue2rgb(p, q, h + 1.0 / 3.0),
hue2rgb(p, q, h),
hue2rgb(p, q, h - 1.0 / 3.0)
);
}
void main() {
vec4 tex = texture(u_image0, v_texCoord);
vec3 color = tex.rgb;
vec3 shadows = vec3(u_float0, u_float1, u_float2) * 0.01;
vec3 midtones = vec3(u_float3, u_float4, u_float5) * 0.01;
vec3 highlights = vec3(u_float6, u_float7, u_float8) * 0.01;
float maxC = max(color.r, max(color.g, color.b));
float minC = min(color.r, min(color.g, color.b));
float lightness = (maxC + minC) * 0.5;
// GIMP weight curves: linear ramps with constants a=0.25, b=0.333, scale=0.7
const float a = 0.25;
const float b = 0.333;
const float scale = 0.7;
float sw = clamp((lightness - b) / -a + 0.5, 0.0, 1.0) * scale;
float mw = clamp((lightness - b) / a + 0.5, 0.0, 1.0) *
clamp((lightness + b - 1.0) / -a + 0.5, 0.0, 1.0) * scale;
float hw = clamp((lightness + b - 1.0) / a + 0.5, 0.0, 1.0) * scale;
color += sw * shadows + mw * midtones + hw * highlights;
if (u_bool0) {
vec3 hsl = rgb2hsl(clamp(color, 0.0, 1.0));
hsl.z = lightness;
color = hsl2rgb(hsl);
}
fragColor = vec4(clamp(color, 0.0, 1.0), tex.a);
}

View File

@ -1,49 +0,0 @@
#version 300 es
precision highp float;
uniform sampler2D u_image0;
uniform sampler2D u_curve0; // RGB master curve (256x1 LUT)
uniform sampler2D u_curve1; // Red channel curve
uniform sampler2D u_curve2; // Green channel curve
uniform sampler2D u_curve3; // Blue channel curve
in vec2 v_texCoord;
layout(location = 0) out vec4 fragColor0;
// GIMP-compatible curve lookup with manual linear interpolation.
// Matches gimp_curve_map_value_inline() from gimpcurve-map.c:
// index = value * (n_samples - 1)
// f = fract(index)
// result = (1-f) * samples[floor] + f * samples[ceil]
//
// Uses texelFetch (NEAREST) to avoid GPU half-texel offset issues
// that occur with texture() + GL_LINEAR on small 256x1 LUTs.
float applyCurve(sampler2D curve, float value) {
value = clamp(value, 0.0, 1.0);
float pos = value * 255.0;
int lo = int(floor(pos));
int hi = min(lo + 1, 255);
float f = pos - float(lo);
float a = texelFetch(curve, ivec2(lo, 0), 0).r;
float b = texelFetch(curve, ivec2(hi, 0), 0).r;
return a + f * (b - a);
}
void main() {
vec4 color = texture(u_image0, v_texCoord);
// GIMP order: per-channel curves first, then RGB master curve.
// See gimp_curve_map_pixels() default case in gimpcurve-map.c:
// dest = colors_curve( channel_curve( src ) )
float tmp_r = applyCurve(u_curve1, color.r);
float tmp_g = applyCurve(u_curve2, color.g);
float tmp_b = applyCurve(u_curve3, color.b);
color.r = applyCurve(u_curve0, tmp_r);
color.g = applyCurve(u_curve0, tmp_g);
color.b = applyCurve(u_curve0, tmp_b);
fragColor0 = vec4(color.rgb, color.a);
}

View File

@ -2,6 +2,7 @@
precision mediump float;
uniform sampler2D u_image0;
uniform vec2 u_resolution;
uniform int u_int0; // Blend mode
uniform int u_int1; // Color tint
uniform float u_float0; // Intensity
@ -74,7 +75,7 @@ void main() {
float t0 = threshold - 0.15;
float t1 = threshold + 0.15;
vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));
vec2 texelSize = 1.0 / u_resolution;
float radius2 = radius * radius;
float sampleScale = clamp(radius * 0.75, 0.35, 1.0);

View File

@ -12,6 +12,7 @@ const int RADIAL_SAMPLES = 12;
const float RADIAL_STRENGTH = 0.0003;
uniform sampler2D u_image0;
uniform vec2 u_resolution;
uniform int u_int0; // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL)
uniform float u_float0; // Blur radius/amount
uniform int u_pass; // Pass index (0 = horizontal, 1 = vertical)
@ -24,7 +25,7 @@ float gaussian(float x, float sigma) {
}
void main() {
vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));
vec2 texelSize = 1.0 / u_resolution;
float radius = max(u_float0, 0.0);
// Radial (angular) blur - single pass, doesn't use separable

View File

@ -2,13 +2,14 @@
precision highp float;
uniform sampler2D u_image0;
uniform vec2 u_resolution;
uniform float u_float0; // strength [0.0 2.0] typical: 0.31.0
in vec2 v_texCoord;
layout(location = 0) out vec4 fragColor0;
void main() {
vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));
vec2 texel = 1.0 / u_resolution;
// Sample center and neighbors
vec4 center = texture(u_image0, v_texCoord);

View File

@ -2,6 +2,7 @@
precision highp float;
uniform sampler2D u_image0;
uniform vec2 u_resolution;
uniform float u_float0; // amount [0.0 - 3.0] typical: 0.5-1.5
uniform float u_float1; // radius [0.5 - 10.0] blur radius in pixels
uniform float u_float2; // threshold [0.0 - 0.1] min difference to sharpen
@ -18,7 +19,7 @@ float getLuminance(vec3 color) {
}
void main() {
vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));
vec2 texel = 1.0 / u_resolution;
float radius = max(u_float1, 0.5);
float amount = u_float0;
float threshold = u_float2;

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

View File

@ -1,615 +0,0 @@
{
"revision": 0,
"last_node_id": 10,
"last_link_id": 0,
"nodes": [
{
"id": 10,
"type": "d5c462c8-1372-4af8-84f2-547c83470d04",
"pos": [
3610,
-2630
],
"size": [
270,
420
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"label": "image",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": null
}
],
"outputs": [
{
"label": "IMAGE",
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": []
}
],
"properties": {
"proxyWidgets": [
[
"4",
"curve"
],
[
"5",
"curve"
],
[
"6",
"curve"
],
[
"7",
"curve"
]
]
},
"widgets_values": [],
"title": "Color Curves"
}
],
"links": [],
"version": 0.4,
"definitions": {
"subgraphs": [
{
"id": "d5c462c8-1372-4af8-84f2-547c83470d04",
"version": 1,
"state": {
"lastGroupId": 0,
"lastNodeId": 9,
"lastLinkId": 38,
"lastRerouteId": 0
},
"revision": 0,
"config": {},
"name": "Color Curves",
"inputNode": {
"id": -10,
"bounding": [
2660,
-4500,
120,
60
]
},
"outputNode": {
"id": -20,
"bounding": [
4270,
-4500,
120,
60
]
},
"inputs": [
{
"id": "abc345b7-f55e-4f32-a11d-3aa4c2b0936b",
"name": "images.image0",
"type": "IMAGE",
"linkIds": [
29,
34
],
"localized_name": "images.image0",
"label": "image",
"pos": [
2760,
-4480
]
}
],
"outputs": [
{
"id": "eb0ec079-46da-4408-8263-9ef85569d33d",
"name": "IMAGE0",
"type": "IMAGE",
"linkIds": [
28
],
"localized_name": "IMAGE0",
"label": "IMAGE",
"pos": [
4290,
-4480
]
}
],
"widgets": [],
"nodes": [
{
"id": 4,
"type": "CurveEditor",
"pos": [
3060,
-4500
],
"size": [
270,
200
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"label": "curve",
"localized_name": "curve",
"name": "curve",
"type": "CURVE",
"widget": {
"name": "curve"
},
"link": null
},
{
"label": "histogram",
"localized_name": "histogram",
"name": "histogram",
"type": "HISTOGRAM",
"shape": 7,
"link": 35
}
],
"outputs": [
{
"localized_name": "CURVE",
"name": "CURVE",
"type": "CURVE",
"links": [
30
]
}
],
"title": "RGB Master",
"properties": {
"Node name for S&R": "CurveEditor"
},
"widgets_values": []
},
{
"id": 5,
"type": "CurveEditor",
"pos": [
3060,
-4250
],
"size": [
270,
200
],
"flags": {},
"order": 1,
"mode": 0,
"inputs": [
{
"label": "curve",
"localized_name": "curve",
"name": "curve",
"type": "CURVE",
"widget": {
"name": "curve"
},
"link": null
},
{
"label": "histogram",
"localized_name": "histogram",
"name": "histogram",
"type": "HISTOGRAM",
"shape": 7,
"link": 36
}
],
"outputs": [
{
"localized_name": "CURVE",
"name": "CURVE",
"type": "CURVE",
"links": [
31
]
}
],
"title": "Red",
"properties": {
"Node name for S&R": "CurveEditor"
},
"widgets_values": []
},
{
"id": 6,
"type": "CurveEditor",
"pos": [
3060,
-4000
],
"size": [
270,
200
],
"flags": {},
"order": 2,
"mode": 0,
"inputs": [
{
"label": "curve",
"localized_name": "curve",
"name": "curve",
"type": "CURVE",
"widget": {
"name": "curve"
},
"link": null
},
{
"label": "histogram",
"localized_name": "histogram",
"name": "histogram",
"type": "HISTOGRAM",
"shape": 7,
"link": 37
}
],
"outputs": [
{
"localized_name": "CURVE",
"name": "CURVE",
"type": "CURVE",
"links": [
32
]
}
],
"title": "Green",
"properties": {
"Node name for S&R": "CurveEditor"
},
"widgets_values": []
},
{
"id": 7,
"type": "CurveEditor",
"pos": [
3060,
-3750
],
"size": [
270,
200
],
"flags": {},
"order": 3,
"mode": 0,
"inputs": [
{
"label": "curve",
"localized_name": "curve",
"name": "curve",
"type": "CURVE",
"widget": {
"name": "curve"
},
"link": null
},
{
"label": "histogram",
"localized_name": "histogram",
"name": "histogram",
"type": "HISTOGRAM",
"shape": 7,
"link": 38
}
],
"outputs": [
{
"localized_name": "CURVE",
"name": "CURVE",
"type": "CURVE",
"links": [
33
]
}
],
"title": "Blue",
"properties": {
"Node name for S&R": "CurveEditor"
},
"widgets_values": []
},
{
"id": 8,
"type": "GLSLShader",
"pos": [
3590,
-4500
],
"size": [
420,
500
],
"flags": {},
"order": 4,
"mode": 0,
"inputs": [
{
"label": "image0",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": 29
},
{
"label": "image1",
"localized_name": "images.image1",
"name": "images.image1",
"shape": 7,
"type": "IMAGE",
"link": null
},
{
"label": "u_curve0",
"localized_name": "curves.u_curve0",
"name": "curves.u_curve0",
"shape": 7,
"type": "CURVE",
"link": 30
},
{
"label": "u_curve1",
"localized_name": "curves.u_curve1",
"name": "curves.u_curve1",
"shape": 7,
"type": "CURVE",
"link": 31
},
{
"label": "u_curve2",
"localized_name": "curves.u_curve2",
"name": "curves.u_curve2",
"shape": 7,
"type": "CURVE",
"link": 32
},
{
"label": "u_curve3",
"localized_name": "curves.u_curve3",
"name": "curves.u_curve3",
"shape": 7,
"type": "CURVE",
"link": 33
},
{
"localized_name": "fragment_shader",
"name": "fragment_shader",
"type": "STRING",
"widget": {
"name": "fragment_shader"
},
"link": null
},
{
"localized_name": "size_mode",
"name": "size_mode",
"type": "COMFY_DYNAMICCOMBO_V3",
"widget": {
"name": "size_mode"
},
"link": null
}
],
"outputs": [
{
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": [
28
]
},
{
"localized_name": "IMAGE1",
"name": "IMAGE1",
"type": "IMAGE",
"links": null
},
{
"localized_name": "IMAGE2",
"name": "IMAGE2",
"type": "IMAGE",
"links": null
},
{
"localized_name": "IMAGE3",
"name": "IMAGE3",
"type": "IMAGE",
"links": null
}
],
"properties": {
"Node name for S&R": "GLSLShader"
},
"widgets_values": [
"#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform sampler2D u_curve0; // RGB master curve (256x1 LUT)\nuniform sampler2D u_curve1; // Red channel curve\nuniform sampler2D u_curve2; // Green channel curve\nuniform sampler2D u_curve3; // Blue channel curve\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\n// GIMP-compatible curve lookup with manual linear interpolation.\n// Matches gimp_curve_map_value_inline() from gimpcurve-map.c:\n// index = value * (n_samples - 1)\n// f = fract(index)\n// result = (1-f) * samples[floor] + f * samples[ceil]\n//\n// Uses texelFetch (NEAREST) to avoid GPU half-texel offset issues\n// that occur with texture() + GL_LINEAR on small 256x1 LUTs.\nfloat applyCurve(sampler2D curve, float value) {\n value = clamp(value, 0.0, 1.0);\n\n float pos = value * 255.0;\n int lo = int(floor(pos));\n int hi = min(lo + 1, 255);\n float f = pos - float(lo);\n\n float a = texelFetch(curve, ivec2(lo, 0), 0).r;\n float b = texelFetch(curve, ivec2(hi, 0), 0).r;\n\n return a + f * (b - a);\n}\n\nvoid main() {\n vec4 color = texture(u_image0, v_texCoord);\n\n // GIMP order: per-channel curves first, then RGB master curve.\n // See gimp_curve_map_pixels() default case in gimpcurve-map.c:\n // dest = colors_curve( channel_curve( src ) )\n float tmp_r = applyCurve(u_curve1, color.r);\n float tmp_g = applyCurve(u_curve2, color.g);\n float tmp_b = applyCurve(u_curve3, color.b);\n color.r = applyCurve(u_curve0, tmp_r);\n color.g = applyCurve(u_curve0, tmp_g);\n color.b = applyCurve(u_curve0, tmp_b);\n\n fragColor0 = vec4(color.rgb, color.a);\n}\n",
"from_input"
]
},
{
"id": 9,
"type": "ImageHistogram",
"pos": [
2800,
-4300
],
"size": [
210,
150
],
"flags": {},
"order": 5,
"mode": 0,
"inputs": [
{
"label": "image",
"localized_name": "image",
"name": "image",
"type": "IMAGE",
"link": 34
}
],
"outputs": [
{
"localized_name": "HISTOGRAM",
"name": "rgb",
"type": "HISTOGRAM",
"links": [
35
]
},
{
"localized_name": "HISTOGRAM",
"name": "luminance",
"type": "HISTOGRAM",
"links": []
},
{
"localized_name": "HISTOGRAM",
"name": "red",
"type": "HISTOGRAM",
"links": [
36
]
},
{
"localized_name": "HISTOGRAM",
"name": "green",
"type": "HISTOGRAM",
"links": [
37
]
},
{
"localized_name": "HISTOGRAM",
"name": "blue",
"type": "HISTOGRAM",
"links": [
38
]
}
],
"properties": {
"Node name for S&R": "ImageHistogram"
},
"widgets_values": []
}
],
"groups": [],
"links": [
{
"id": 29,
"origin_id": -10,
"origin_slot": 0,
"target_id": 8,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 28,
"origin_id": 8,
"origin_slot": 0,
"target_id": -20,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 30,
"origin_id": 4,
"origin_slot": 0,
"target_id": 8,
"target_slot": 2,
"type": "CURVE"
},
{
"id": 31,
"origin_id": 5,
"origin_slot": 0,
"target_id": 8,
"target_slot": 3,
"type": "CURVE"
},
{
"id": 32,
"origin_id": 6,
"origin_slot": 0,
"target_id": 8,
"target_slot": 4,
"type": "CURVE"
},
{
"id": 33,
"origin_id": 7,
"origin_slot": 0,
"target_id": 8,
"target_slot": 5,
"type": "CURVE"
},
{
"id": 34,
"origin_id": -10,
"origin_slot": 0,
"target_id": 9,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 35,
"origin_id": 9,
"origin_slot": 0,
"target_id": 4,
"target_slot": 1,
"type": "HISTOGRAM"
},
{
"id": 36,
"origin_id": 9,
"origin_slot": 2,
"target_id": 5,
"target_slot": 1,
"type": "HISTOGRAM"
},
{
"id": 37,
"origin_id": 9,
"origin_slot": 3,
"target_id": 6,
"target_slot": 1,
"type": "HISTOGRAM"
},
{
"id": 38,
"origin_id": 9,
"origin_slot": 4,
"target_id": 7,
"target_slot": 1,
"type": "HISTOGRAM"
}
],
"extra": {
"workflowRendererVersion": "LG"
},
"category": "Image Tools/Color adjust"
}
]
}
}

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

View File

@ -1,322 +1 @@
{
"revision": 0,
"last_node_id": 29,
"last_link_id": 0,
"nodes": [
{
"id": 29,
"type": "4c9d6ea4-b912-40e5-8766-6793a9758c53",
"pos": [
1970,
-230
],
"size": [
180,
86
],
"flags": {},
"order": 5,
"mode": 0,
"inputs": [
{
"label": "image",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": null
}
],
"outputs": [
{
"label": "R",
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": []
},
{
"label": "G",
"localized_name": "IMAGE1",
"name": "IMAGE1",
"type": "IMAGE",
"links": []
},
{
"label": "B",
"localized_name": "IMAGE2",
"name": "IMAGE2",
"type": "IMAGE",
"links": []
},
{
"label": "A",
"localized_name": "IMAGE3",
"name": "IMAGE3",
"type": "IMAGE",
"links": []
}
],
"title": "Image Channels",
"properties": {
"proxyWidgets": []
},
"widgets_values": []
}
],
"links": [],
"version": 0.4,
"definitions": {
"subgraphs": [
{
"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53",
"version": 1,
"state": {
"lastGroupId": 0,
"lastNodeId": 28,
"lastLinkId": 39,
"lastRerouteId": 0
},
"revision": 0,
"config": {},
"name": "Image Channels",
"inputNode": {
"id": -10,
"bounding": [
1820,
-185,
120,
60
]
},
"outputNode": {
"id": -20,
"bounding": [
2460,
-215,
120,
120
]
},
"inputs": [
{
"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe",
"name": "images.image0",
"type": "IMAGE",
"linkIds": [
39
],
"localized_name": "images.image0",
"label": "image",
"pos": [
1920,
-165
]
}
],
"outputs": [
{
"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b",
"name": "IMAGE0",
"type": "IMAGE",
"linkIds": [
26
],
"localized_name": "IMAGE0",
"label": "R",
"pos": [
2480,
-195
]
},
{
"id": "fb44a77e-0522-43e9-9527-82e7465b3596",
"name": "IMAGE1",
"type": "IMAGE",
"linkIds": [
27
],
"localized_name": "IMAGE1",
"label": "G",
"pos": [
2480,
-175
]
},
{
"id": "81460ee6-0131-402a-874f-6bf3001fc4ff",
"name": "IMAGE2",
"type": "IMAGE",
"linkIds": [
28
],
"localized_name": "IMAGE2",
"label": "B",
"pos": [
2480,
-155
]
},
{
"id": "ae690246-80d4-4951-b1d9-9306d8a77417",
"name": "IMAGE3",
"type": "IMAGE",
"linkIds": [
29
],
"localized_name": "IMAGE3",
"label": "A",
"pos": [
2480,
-135
]
}
],
"widgets": [],
"nodes": [
{
"id": 23,
"type": "GLSLShader",
"pos": [
2000,
-330
],
"size": [
400,
172
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"label": "image",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": 39
},
{
"localized_name": "fragment_shader",
"name": "fragment_shader",
"type": "STRING",
"widget": {
"name": "fragment_shader"
},
"link": null
},
{
"localized_name": "size_mode",
"name": "size_mode",
"type": "COMFY_DYNAMICCOMBO_V3",
"widget": {
"name": "size_mode"
},
"link": null
},
{
"label": "image1",
"localized_name": "images.image1",
"name": "images.image1",
"shape": 7,
"type": "IMAGE",
"link": null
}
],
"outputs": [
{
"label": "R",
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": [
26
]
},
{
"label": "G",
"localized_name": "IMAGE1",
"name": "IMAGE1",
"type": "IMAGE",
"links": [
27
]
},
{
"label": "B",
"localized_name": "IMAGE2",
"name": "IMAGE2",
"type": "IMAGE",
"links": [
28
]
},
{
"label": "A",
"localized_name": "IMAGE3",
"name": "IMAGE3",
"type": "IMAGE",
"links": [
29
]
}
],
"properties": {
"Node name for S&R": "GLSLShader"
},
"widgets_values": [
"#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n vec4 color = texture(u_image0, v_texCoord);\n // Output each channel as grayscale to separate render targets\n fragColor0 = vec4(vec3(color.r), 1.0); // Red channel\n fragColor1 = vec4(vec3(color.g), 1.0); // Green channel\n fragColor2 = vec4(vec3(color.b), 1.0); // Blue channel\n fragColor3 = vec4(vec3(color.a), 1.0); // Alpha channel\n}\n",
"from_input"
]
}
],
"groups": [],
"links": [
{
"id": 39,
"origin_id": -10,
"origin_slot": 0,
"target_id": 23,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 26,
"origin_id": 23,
"origin_slot": 0,
"target_id": -20,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 27,
"origin_id": 23,
"origin_slot": 1,
"target_id": -20,
"target_slot": 1,
"type": "IMAGE"
},
{
"id": 28,
"origin_id": 23,
"origin_slot": 2,
"target_id": -20,
"target_slot": 2,
"type": "IMAGE"
},
{
"id": 29,
"origin_id": 23,
"origin_slot": 3,
"target_id": -20,
"target_slot": 3,
"type": "IMAGE"
}
],
"extra": {
"workflowRendererVersion": "LG"
},
"category": "Image Tools/Color adjust"
}
]
}
}
{"revision": 0, "last_node_id": 29, "last_link_id": 0, "nodes": [{"id": 29, "type": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "pos": [1970, -230], "size": [180, 86], "flags": {}, "order": 5, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": []}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": []}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": []}], "title": "Image Channels", "properties": {"proxyWidgets": []}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 28, "lastLinkId": 39, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Image Channels", "inputNode": {"id": -10, "bounding": [1820, -185, 120, 60]}, "outputNode": {"id": -20, "bounding": [2460, -215, 120, 120]}, "inputs": [{"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe", "name": "images.image0", "type": "IMAGE", "linkIds": [39], "localized_name": "images.image0", "label": "image", "pos": [1920, -165]}], "outputs": [{"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b", "name": "IMAGE0", "type": "IMAGE", "linkIds": [26], "localized_name": "IMAGE0", "label": "R", "pos": [2480, -195]}, {"id": "fb44a77e-0522-43e9-9527-82e7465b3596", "name": "IMAGE1", "type": "IMAGE", "linkIds": [27], "localized_name": "IMAGE1", "label": "G", "pos": [2480, -175]}, {"id": "81460ee6-0131-402a-874f-6bf3001fc4ff", "name": "IMAGE2", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE2", "label": "B", "pos": [2480, -155]}, {"id": "ae690246-80d4-4951-b1d9-9306d8a77417", "name": "IMAGE3", "type": "IMAGE", "linkIds": [29], "localized_name": "IMAGE3", "label": "A", "pos": [2480, -135]}], "widgets": [], "nodes": [{"id": 23, "type": "GLSLShader", "pos": [2000, -330], "size": [400, 172], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 39}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [26]}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": [27]}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": [28]}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": [29]}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n vec4 color = texture(u_image0, v_texCoord);\n // Output each channel as grayscale to separate render targets\n fragColor0 = vec4(vec3(color.r), 1.0); // Red channel\n fragColor1 = vec4(vec3(color.g), 1.0); // Green channel\n fragColor2 = vec4(vec3(color.b), 1.0); // Blue channel\n fragColor3 = vec4(vec3(color.a), 1.0); // Alpha channel\n}\n", "from_input"]}], "groups": [], "links": [{"id": 39, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 26, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}, {"id": 27, "origin_id": 23, "origin_slot": 1, "target_id": -20, "target_slot": 1, "type": "IMAGE"}, {"id": 28, "origin_id": 23, "origin_slot": 2, "target_id": -20, "target_slot": 2, "type": "IMAGE"}, {"id": 29, "origin_id": 23, "origin_slot": 3, "target_id": -20, "target_slot": 3, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}}

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

View File

@ -1,278 +1 @@
{
"revision": 0,
"last_node_id": 15,
"last_link_id": 0,
"nodes": [
{
"id": 15,
"type": "24d8bbfd-39d4-4774-bff0-3de40cc7a471",
"pos": [
-1490,
2040
],
"size": [
400,
260
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"name": "prompt",
"type": "STRING",
"widget": {
"name": "prompt"
},
"link": null
},
{
"label": "reference images",
"name": "images",
"type": "IMAGE",
"link": null
}
],
"outputs": [
{
"name": "STRING",
"type": "STRING",
"links": null
}
],
"title": "Prompt Enhance",
"properties": {
"proxyWidgets": [
[
"-1",
"prompt"
]
],
"cnr_id": "comfy-core",
"ver": "0.14.1"
},
"widgets_values": [
""
]
}
],
"links": [],
"version": 0.4,
"definitions": {
"subgraphs": [
{
"id": "24d8bbfd-39d4-4774-bff0-3de40cc7a471",
"version": 1,
"state": {
"lastGroupId": 0,
"lastNodeId": 15,
"lastLinkId": 14,
"lastRerouteId": 0
},
"revision": 0,
"config": {},
"name": "Prompt Enhance",
"inputNode": {
"id": -10,
"bounding": [
-2170,
2110,
138.876953125,
80
]
},
"outputNode": {
"id": -20,
"bounding": [
-640,
2110,
120,
60
]
},
"inputs": [
{
"id": "aeab7216-00e0-4528-a09b-bba50845c5a6",
"name": "prompt",
"type": "STRING",
"linkIds": [
11
],
"pos": [
-2051.123046875,
2130
]
},
{
"id": "7b73fd36-aa31-4771-9066-f6c83879994b",
"name": "images",
"type": "IMAGE",
"linkIds": [
14
],
"label": "reference images",
"pos": [
-2051.123046875,
2150
]
}
],
"outputs": [
{
"id": "c7b0d930-68a1-48d1-b496-0519e5837064",
"name": "STRING",
"type": "STRING",
"linkIds": [
13
],
"pos": [
-620,
2130
]
}
],
"widgets": [],
"nodes": [
{
"id": 11,
"type": "GeminiNode",
"pos": [
-1560,
1990
],
"size": [
470,
470
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"localized_name": "images",
"name": "images",
"shape": 7,
"type": "IMAGE",
"link": 14
},
{
"localized_name": "audio",
"name": "audio",
"shape": 7,
"type": "AUDIO",
"link": null
},
{
"localized_name": "video",
"name": "video",
"shape": 7,
"type": "VIDEO",
"link": null
},
{
"localized_name": "files",
"name": "files",
"shape": 7,
"type": "GEMINI_INPUT_FILES",
"link": null
},
{
"localized_name": "prompt",
"name": "prompt",
"type": "STRING",
"widget": {
"name": "prompt"
},
"link": 11
},
{
"localized_name": "model",
"name": "model",
"type": "COMBO",
"widget": {
"name": "model"
},
"link": null
},
{
"localized_name": "seed",
"name": "seed",
"type": "INT",
"widget": {
"name": "seed"
},
"link": null
},
{
"localized_name": "system_prompt",
"name": "system_prompt",
"shape": 7,
"type": "STRING",
"widget": {
"name": "system_prompt"
},
"link": null
}
],
"outputs": [
{
"localized_name": "STRING",
"name": "STRING",
"type": "STRING",
"links": [
13
]
}
],
"properties": {
"cnr_id": "comfy-core",
"ver": "0.14.1",
"Node name for S&R": "GeminiNode"
},
"widgets_values": [
"",
"gemini-3-pro-preview",
42,
"randomize",
"You are an expert in prompt writing.\nBased on the input, rewrite the user's input into a detailed prompt.\nincluding camera settings, lighting, composition, and style.\nReturn the prompt only"
],
"color": "#432",
"bgcolor": "#653"
}
],
"groups": [],
"links": [
{
"id": 11,
"origin_id": -10,
"origin_slot": 0,
"target_id": 11,
"target_slot": 4,
"type": "STRING"
},
{
"id": 13,
"origin_id": 11,
"origin_slot": 0,
"target_id": -20,
"target_slot": 0,
"type": "STRING"
},
{
"id": 14,
"origin_id": -10,
"origin_slot": 1,
"target_id": 11,
"target_slot": 0,
"type": "IMAGE"
}
],
"extra": {
"workflowRendererVersion": "LG"
},
"category": "Text generation/Prompt enhance"
}
]
},
"extra": {}
}
{"revision": 0, "last_node_id": 15, "last_link_id": 0, "nodes": [{"id": 15, "type": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "pos": [-1490, 2040], "size": [400, 260], "flags": {}, "order": 0, "mode": 0, "inputs": [{"name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": null}, {"label": "reference images", "name": "images", "type": "IMAGE", "link": null}], "outputs": [{"name": "STRING", "type": "STRING", "links": null}], "title": "Prompt Enhance", "properties": {"proxyWidgets": [["-1", "prompt"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": [""]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 15, "lastLinkId": 14, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Prompt Enhance", "inputNode": {"id": -10, "bounding": [-2170, 2110, 138.876953125, 80]}, "outputNode": {"id": -20, "bounding": [-640, 2110, 120, 60]}, "inputs": [{"id": "aeab7216-00e0-4528-a09b-bba50845c5a6", "name": "prompt", "type": "STRING", "linkIds": [11], "pos": [-2051.123046875, 2130]}, {"id": "7b73fd36-aa31-4771-9066-f6c83879994b", "name": "images", "type": "IMAGE", "linkIds": [14], "label": "reference images", "pos": [-2051.123046875, 2150]}], "outputs": [{"id": "c7b0d930-68a1-48d1-b496-0519e5837064", "name": "STRING", "type": "STRING", "linkIds": [13], "pos": [-620, 2130]}], "widgets": [], "nodes": [{"id": 11, "type": "GeminiNode", "pos": [-1560, 1990], "size": [470, 470], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "shape": 7, "type": "IMAGE", "link": 14}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": null}, {"localized_name": "video", "name": "video", "shape": 7, "type": "VIDEO", "link": null}, {"localized_name": "files", "name": "files", "shape": 7, "type": "GEMINI_INPUT_FILES", "link": null}, {"localized_name": "prompt", "name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": 11}, {"localized_name": "model", "name": "model", "type": "COMBO", "widget": {"name": "model"}, "link": null}, {"localized_name": "seed", "name": "seed", "type": "INT", "widget": {"name": "seed"}, "link": null}, {"localized_name": "system_prompt", "name": "system_prompt", "shape": 7, "type": "STRING", "widget": {"name": "system_prompt"}, "link": null}], "outputs": [{"localized_name": "STRING", "name": "STRING", "type": "STRING", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.14.1", "Node name for S&R": "GeminiNode"}, "widgets_values": ["", "gemini-3-pro-preview", 42, "randomize", "You are an expert in prompt writing.\nBased on the input, rewrite the user's input into a detailed prompt.\nincluding camera settings, lighting, composition, and style.\nReturn the prompt only"], "color": "#432", "bgcolor": "#653"}], "groups": [], "links": [{"id": 11, "origin_id": -10, "origin_slot": 0, "target_id": 11, "target_slot": 4, "type": "STRING"}, {"id": 13, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "STRING"}, {"id": 14, "origin_id": -10, "origin_slot": 1, "target_id": 11, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Text generation/Prompt enhance"}]}, "extra": {}}

View File

@ -1,309 +1 @@
{
"revision": 0,
"last_node_id": 25,
"last_link_id": 0,
"nodes": [
{
"id": 25,
"type": "621ba4e2-22a8-482d-a369-023753198b7b",
"pos": [
4610,
-790
],
"size": [
230,
58
],
"flags": {},
"order": 4,
"mode": 0,
"inputs": [
{
"label": "image",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": null
}
],
"outputs": [
{
"label": "IMAGE",
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": []
}
],
"title": "Sharpen",
"properties": {
"proxyWidgets": [
[
"24",
"value"
]
]
},
"widgets_values": []
}
],
"links": [],
"version": 0.4,
"definitions": {
"subgraphs": [
{
"id": "621ba4e2-22a8-482d-a369-023753198b7b",
"version": 1,
"state": {
"lastGroupId": 0,
"lastNodeId": 24,
"lastLinkId": 36,
"lastRerouteId": 0
},
"revision": 0,
"config": {},
"name": "Sharpen",
"inputNode": {
"id": -10,
"bounding": [
4090,
-825,
120,
60
]
},
"outputNode": {
"id": -20,
"bounding": [
5150,
-825,
120,
60
]
},
"inputs": [
{
"id": "37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7",
"name": "images.image0",
"type": "IMAGE",
"linkIds": [
34
],
"localized_name": "images.image0",
"label": "image",
"pos": [
4190,
-805
]
}
],
"outputs": [
{
"id": "e9182b3f-635c-4cd4-a152-4b4be17ae4b9",
"name": "IMAGE0",
"type": "IMAGE",
"linkIds": [
35
],
"localized_name": "IMAGE0",
"label": "IMAGE",
"pos": [
5170,
-805
]
}
],
"widgets": [],
"nodes": [
{
"id": 24,
"type": "PrimitiveFloat",
"pos": [
4280,
-1240
],
"size": [
270,
58
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"label": "strength",
"localized_name": "value",
"name": "value",
"type": "FLOAT",
"widget": {
"name": "value"
},
"link": null
}
],
"outputs": [
{
"localized_name": "FLOAT",
"name": "FLOAT",
"type": "FLOAT",
"links": [
36
]
}
],
"properties": {
"Node name for S&R": "PrimitiveFloat",
"min": 0,
"max": 3,
"precision": 2,
"step": 0.05
},
"widgets_values": [
0.5
]
},
{
"id": 23,
"type": "GLSLShader",
"pos": [
4570,
-1240
],
"size": [
370,
192
],
"flags": {},
"order": 1,
"mode": 0,
"inputs": [
{
"label": "image0",
"localized_name": "images.image0",
"name": "images.image0",
"type": "IMAGE",
"link": 34
},
{
"label": "image1",
"localized_name": "images.image1",
"name": "images.image1",
"shape": 7,
"type": "IMAGE",
"link": null
},
{
"label": "u_float0",
"localized_name": "floats.u_float0",
"name": "floats.u_float0",
"shape": 7,
"type": "FLOAT",
"link": 36
},
{
"label": "u_float1",
"localized_name": "floats.u_float1",
"name": "floats.u_float1",
"shape": 7,
"type": "FLOAT",
"link": null
},
{
"label": "u_int0",
"localized_name": "ints.u_int0",
"name": "ints.u_int0",
"shape": 7,
"type": "INT",
"link": null
},
{
"localized_name": "fragment_shader",
"name": "fragment_shader",
"type": "STRING",
"widget": {
"name": "fragment_shader"
},
"link": null
},
{
"localized_name": "size_mode",
"name": "size_mode",
"type": "COMFY_DYNAMICCOMBO_V3",
"widget": {
"name": "size_mode"
},
"link": null
}
],
"outputs": [
{
"localized_name": "IMAGE0",
"name": "IMAGE0",
"type": "IMAGE",
"links": [
35
]
},
{
"localized_name": "IMAGE1",
"name": "IMAGE1",
"type": "IMAGE",
"links": null
},
{
"localized_name": "IMAGE2",
"name": "IMAGE2",
"type": "IMAGE",
"links": null
},
{
"localized_name": "IMAGE3",
"name": "IMAGE3",
"type": "IMAGE",
"links": null
}
],
"properties": {
"Node name for S&R": "GLSLShader"
},
"widgets_values": [
"#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0; // strength [0.0 2.0] typical: 0.31.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));\n \n // Sample center and neighbors\n vec4 center = texture(u_image0, v_texCoord);\n vec4 top = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0, texel.y));\n vec4 left = texture(u_image0, v_texCoord + vec2(-texel.x, 0.0));\n vec4 right = texture(u_image0, v_texCoord + vec2( texel.x, 0.0));\n \n // Edge enhancement (Laplacian)\n vec4 edges = center * 4.0 - top - bottom - left - right;\n \n // Add edges back scaled by strength\n vec4 sharpened = center + edges * u_float0;\n \n fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}",
"from_input"
]
}
],
"groups": [],
"links": [
{
"id": 36,
"origin_id": 24,
"origin_slot": 0,
"target_id": 23,
"target_slot": 2,
"type": "FLOAT"
},
{
"id": 34,
"origin_id": -10,
"origin_slot": 0,
"target_id": 23,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 35,
"origin_id": 23,
"origin_slot": 0,
"target_id": -20,
"target_slot": 0,
"type": "IMAGE"
}
],
"extra": {
"workflowRendererVersion": "LG"
},
"category": "Image Tools/Sharpen"
}
]
}
}
{"revision": 0, "last_node_id": 25, "last_link_id": 0, "nodes": [{"id": 25, "type": "621ba4e2-22a8-482d-a369-023753198b7b", "pos": [4610, -790], "size": [230, 58], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "title": "Sharpen", "properties": {"proxyWidgets": [["24", "value"]]}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "621ba4e2-22a8-482d-a369-023753198b7b", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 24, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Sharpen", "inputNode": {"id": -10, "bounding": [4090, -825, 120, 60]}, "outputNode": {"id": -20, "bounding": [5150, -825, 120, 60]}, "inputs": [{"id": "37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7", "name": "images.image0", "type": "IMAGE", "linkIds": [34], "localized_name": "images.image0", "label": "image", "pos": [4190, -805]}], "outputs": [{"id": "e9182b3f-635c-4cd4-a152-4b4be17ae4b9", "name": "IMAGE0", "type": "IMAGE", "linkIds": [35], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [5170, -805]}], "widgets": [], "nodes": [{"id": 24, "type": "PrimitiveFloat", "pos": [4280, -1240], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "strength", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [36]}], "properties": {"Node name for S&R": "PrimitiveFloat", "min": 0, "max": 3, "precision": 2, "step": 0.05}, "widgets_values": [0.5]}, {"id": 23, "type": "GLSLShader", "pos": [4570, -1240], "size": [370, 192], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 34}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 36}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [35]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0; // strength [0.0 2.0] typical: 0.31.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n vec2 texel = 1.0 / u_resolution;\n \n // Sample center and neighbors\n vec4 center = texture(u_image0, v_texCoord);\n vec4 top = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0, texel.y));\n vec4 left = texture(u_image0, v_texCoord + vec2(-texel.x, 0.0));\n vec4 right = texture(u_image0, v_texCoord + vec2( texel.x, 0.0));\n \n // Edge enhancement (Laplacian)\n vec4 edges = center * 4.0 - top - bottom - left - right;\n \n // Add edges back scaled by strength\n vec4 sharpened = center + edges * u_float0;\n \n fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}", "from_input"]}], "groups": [], "links": [{"id": 36, "origin_id": 24, "origin_slot": 0, "target_id": 23, "target_slot": 2, "type": "FLOAT"}, {"id": 34, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 35, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Sharpen"}]}}

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

File diff suppressed because one or more lines are too long

View File

@ -1,420 +1 @@
{
"revision": 0,
"last_node_id": 13,
"last_link_id": 0,
"nodes": [
{
"id": 13,
"type": "cf95b747-3e17-46cb-8097-cac60ff9b2e1",
"pos": [
1120,
330
],
"size": [
240,
58
],
"flags": {},
"order": 3,
"mode": 0,
"inputs": [
{
"localized_name": "video",
"name": "video",
"type": "VIDEO",
"link": null
},
{
"name": "model_name",
"type": "COMBO",
"widget": {
"name": "model_name"
},
"link": null
}
],
"outputs": [
{
"localized_name": "VIDEO",
"name": "VIDEO",
"type": "VIDEO",
"links": []
}
],
"title": "Video Upscale(GAN x4)",
"properties": {
"proxyWidgets": [
[
"-1",
"model_name"
]
],
"cnr_id": "comfy-core",
"ver": "0.14.1"
},
"widgets_values": [
"RealESRGAN_x4plus.safetensors"
]
}
],
"links": [],
"version": 0.4,
"definitions": {
"subgraphs": [
{
"id": "cf95b747-3e17-46cb-8097-cac60ff9b2e1",
"version": 1,
"state": {
"lastGroupId": 0,
"lastNodeId": 13,
"lastLinkId": 19,
"lastRerouteId": 0
},
"revision": 0,
"config": {},
"name": "Video Upscale(GAN x4)",
"inputNode": {
"id": -10,
"bounding": [
550,
460,
120,
80
]
},
"outputNode": {
"id": -20,
"bounding": [
1490,
460,
120,
60
]
},
"inputs": [
{
"id": "666d633e-93e7-42dc-8d11-2b7b99b0f2a6",
"name": "video",
"type": "VIDEO",
"linkIds": [
10
],
"localized_name": "video",
"pos": [
650,
480
]
},
{
"id": "2e23a087-caa8-4d65-99e6-662761aa905a",
"name": "model_name",
"type": "COMBO",
"linkIds": [
19
],
"pos": [
650,
500
]
}
],
"outputs": [
{
"id": "0c1768ea-3ec2-412f-9af6-8e0fa36dae70",
"name": "VIDEO",
"type": "VIDEO",
"linkIds": [
15
],
"localized_name": "VIDEO",
"pos": [
1510,
480
]
}
],
"widgets": [],
"nodes": [
{
"id": 2,
"type": "ImageUpscaleWithModel",
"pos": [
1110,
450
],
"size": [
320,
46
],
"flags": {},
"order": 1,
"mode": 0,
"inputs": [
{
"localized_name": "upscale_model",
"name": "upscale_model",
"type": "UPSCALE_MODEL",
"link": 1
},
{
"localized_name": "image",
"name": "image",
"type": "IMAGE",
"link": 14
}
],
"outputs": [
{
"localized_name": "IMAGE",
"name": "IMAGE",
"type": "IMAGE",
"links": [
13
]
}
],
"properties": {
"cnr_id": "comfy-core",
"ver": "0.10.0",
"Node name for S&R": "ImageUpscaleWithModel"
}
},
{
"id": 11,
"type": "CreateVideo",
"pos": [
1110,
550
],
"size": [
320,
78
],
"flags": {},
"order": 3,
"mode": 0,
"inputs": [
{
"localized_name": "images",
"name": "images",
"type": "IMAGE",
"link": 13
},
{
"localized_name": "audio",
"name": "audio",
"shape": 7,
"type": "AUDIO",
"link": 16
},
{
"localized_name": "fps",
"name": "fps",
"type": "FLOAT",
"widget": {
"name": "fps"
},
"link": 12
}
],
"outputs": [
{
"localized_name": "VIDEO",
"name": "VIDEO",
"type": "VIDEO",
"links": [
15
]
}
],
"properties": {
"cnr_id": "comfy-core",
"ver": "0.10.0",
"Node name for S&R": "CreateVideo"
},
"widgets_values": [
30
]
},
{
"id": 10,
"type": "GetVideoComponents",
"pos": [
1110,
330
],
"size": [
320,
70
],
"flags": {},
"order": 2,
"mode": 0,
"inputs": [
{
"localized_name": "video",
"name": "video",
"type": "VIDEO",
"link": 10
}
],
"outputs": [
{
"localized_name": "images",
"name": "images",
"type": "IMAGE",
"links": [
14
]
},
{
"localized_name": "audio",
"name": "audio",
"type": "AUDIO",
"links": [
16
]
},
{
"localized_name": "fps",
"name": "fps",
"type": "FLOAT",
"links": [
12
]
}
],
"properties": {
"cnr_id": "comfy-core",
"ver": "0.10.0",
"Node name for S&R": "GetVideoComponents"
}
},
{
"id": 1,
"type": "UpscaleModelLoader",
"pos": [
750,
450
],
"size": [
280,
60
],
"flags": {},
"order": 0,
"mode": 0,
"inputs": [
{
"localized_name": "model_name",
"name": "model_name",
"type": "COMBO",
"widget": {
"name": "model_name"
},
"link": 19
}
],
"outputs": [
{
"localized_name": "UPSCALE_MODEL",
"name": "UPSCALE_MODEL",
"type": "UPSCALE_MODEL",
"links": [
1
]
}
],
"properties": {
"cnr_id": "comfy-core",
"ver": "0.10.0",
"Node name for S&R": "UpscaleModelLoader",
"models": [
{
"name": "RealESRGAN_x4plus.safetensors",
"url": "https://huggingface.co/Comfy-Org/Real-ESRGAN_repackaged/resolve/main/RealESRGAN_x4plus.safetensors",
"directory": "upscale_models"
}
]
},
"widgets_values": [
"RealESRGAN_x4plus.safetensors"
]
}
],
"groups": [],
"links": [
{
"id": 1,
"origin_id": 1,
"origin_slot": 0,
"target_id": 2,
"target_slot": 0,
"type": "UPSCALE_MODEL"
},
{
"id": 14,
"origin_id": 10,
"origin_slot": 0,
"target_id": 2,
"target_slot": 1,
"type": "IMAGE"
},
{
"id": 13,
"origin_id": 2,
"origin_slot": 0,
"target_id": 11,
"target_slot": 0,
"type": "IMAGE"
},
{
"id": 16,
"origin_id": 10,
"origin_slot": 1,
"target_id": 11,
"target_slot": 1,
"type": "AUDIO"
},
{
"id": 12,
"origin_id": 10,
"origin_slot": 2,
"target_id": 11,
"target_slot": 2,
"type": "FLOAT"
},
{
"id": 10,
"origin_id": -10,
"origin_slot": 0,
"target_id": 10,
"target_slot": 0,
"type": "VIDEO"
},
{
"id": 15,
"origin_id": 11,
"origin_slot": 0,
"target_id": -20,
"target_slot": 0,
"type": "VIDEO"
},
{
"id": 19,
"origin_id": -10,
"origin_slot": 1,
"target_id": 1,
"target_slot": 0,
"type": "COMBO"
}
],
"extra": {
"workflowRendererVersion": "LG"
},
"category": "Video generation and editing/Enhance video"
}
]
},
"extra": {}
}
{"revision": 0, "last_node_id": 13, "last_link_id": 0, "nodes": [{"id": 13, "type": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "pos": [1120, 330], "size": [240, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": null}, {"name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": null}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": []}], "title": "Video Upscale(GAN x4)", "properties": {"proxyWidgets": [["-1", "model_name"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 13, "lastLinkId": 19, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Video Upscale(GAN x4)", "inputNode": {"id": -10, "bounding": [550, 460, 120, 80]}, "outputNode": {"id": -20, "bounding": [1490, 460, 120, 60]}, "inputs": [{"id": "666d633e-93e7-42dc-8d11-2b7b99b0f2a6", "name": "video", "type": "VIDEO", "linkIds": [10], "localized_name": "video", "pos": [650, 480]}, {"id": "2e23a087-caa8-4d65-99e6-662761aa905a", "name": "model_name", "type": "COMBO", "linkIds": [19], "pos": [650, 500]}], "outputs": [{"id": "0c1768ea-3ec2-412f-9af6-8e0fa36dae70", "name": "VIDEO", "type": "VIDEO", "linkIds": [15], "localized_name": "VIDEO", "pos": [1510, 480]}], "widgets": [], "nodes": [{"id": 2, "type": "ImageUpscaleWithModel", "pos": [1110, 450], "size": [320, 46], "flags": {}, "order": 1, "mode": 0, "inputs": [{"localized_name": "upscale_model", "name": "upscale_model", "type": "UPSCALE_MODEL", "link": 1}, {"localized_name": "image", "name": "image", "type": "IMAGE", "link": 14}], "outputs": [{"localized_name": "IMAGE", "name": "IMAGE", "type": "IMAGE", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "ImageUpscaleWithModel"}}, {"id": 11, "type": "CreateVideo", "pos": [1110, 550], "size": [320, 78], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "link": 13}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": 16}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "widget": {"name": "fps"}, "link": 12}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": [15]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "CreateVideo"}, "widgets_values": [30]}, {"id": 10, "type": "GetVideoComponents", "pos": [1110, 330], "size": [320, 70], "flags": {}, "order": 2, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": 10}], "outputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "links": [14]}, {"localized_name": "audio", "name": "audio", "type": "AUDIO", "links": [16]}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "links": [12]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "GetVideoComponents"}}, {"id": 1, "type": "UpscaleModelLoader", "pos": [750, 450], "size": [280, 60], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "model_name", "name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": 19}], "outputs": [{"localized_name": "UPSCALE_MODEL", "name": "UPSCALE_MODEL", "type": "UPSCALE_MODEL", "links": [1]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "UpscaleModelLoader", "models": [{"name": "RealESRGAN_x4plus.safetensors", "url": "https://huggingface.co/Comfy-Org/Real-ESRGAN_repackaged/resolve/main/RealESRGAN_x4plus.safetensors", "directory": "upscale_models"}]}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "groups": [], "links": [{"id": 1, "origin_id": 1, "origin_slot": 0, "target_id": 2, "target_slot": 0, "type": "UPSCALE_MODEL"}, {"id": 14, "origin_id": 10, "origin_slot": 0, "target_id": 2, "target_slot": 1, "type": "IMAGE"}, {"id": 13, "origin_id": 2, "origin_slot": 0, "target_id": 11, "target_slot": 0, "type": "IMAGE"}, {"id": 16, "origin_id": 10, "origin_slot": 1, "target_id": 11, "target_slot": 1, "type": "AUDIO"}, {"id": 12, "origin_id": 10, "origin_slot": 2, "target_id": 11, "target_slot": 2, "type": "FLOAT"}, {"id": 10, "origin_id": -10, "origin_slot": 0, "target_id": 10, "target_slot": 0, "type": "VIDEO"}, {"id": 15, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "VIDEO"}, {"id": 19, "origin_id": -10, "origin_slot": 1, "target_id": 1, "target_slot": 0, "type": "COMBO"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Video generation and editing/Enhance video"}]}, "extra": {}}

View File

@ -83,13 +83,12 @@ fpte_group.add_argument("--fp16-text-enc", action="store_true", help="Store text
fpte_group.add_argument("--fp32-text-enc", action="store_true", help="Store text encoder weights in fp32.")
fpte_group.add_argument("--bf16-text-enc", action="store_true", help="Store text encoder weights in bf16.")
parser.add_argument("--fp16-intermediates", action="store_true", help="Experimental: Use fp16 for intermediate tensors between nodes instead of fp32.")
parser.add_argument("--force-channels-last", action="store_true", help="Force channels last format when inferencing the models.")
parser.add_argument("--directml", type=int, nargs="?", metavar="DIRECTML_DEVICE", const=-1, help="Use torch-directml.")
parser.add_argument("--oneapi-device-selector", type=str, default=None, metavar="SELECTOR_STRING", help="Sets the oneAPI device(s) this instance will use.")
parser.add_argument("--disable-ipex-optimize", action="store_true", help="Disables ipex.optimize default when loading models with Intel's Extension for Pytorch.")
parser.add_argument("--supports-fp8-compute", action="store_true", help="ComfyUI will act like if the device supports fp8 compute.")
class LatentPreviewMethod(enum.Enum):
@ -109,13 +108,11 @@ parser.add_argument("--preview-method", type=LatentPreviewMethod, default=Latent
parser.add_argument("--preview-size", type=int, default=512, help="Sets the maximum preview size for sampler nodes.")
CACHE_RAM_AUTO_GB = -1.0
cache_group = parser.add_mutually_exclusive_group()
cache_group.add_argument("--cache-classic", action="store_true", help="Use the old style (aggressive) caching.")
cache_group.add_argument("--cache-lru", type=int, default=0, help="Use LRU caching with a maximum of N node results cached. May use more RAM/VRAM.")
cache_group.add_argument("--cache-none", action="store_true", help="Reduced RAM/VRAM usage at the expense of executing every node for each run.")
cache_group.add_argument("--cache-ram", nargs='?', const=CACHE_RAM_AUTO_GB, type=float, default=0, help="Use RAM pressure caching with the specified headroom threshold. If available RAM drops below the threshold the cache removes large items to free RAM. Default (when no value is provided): 25%% of system RAM (min 4GB, max 32GB).")
cache_group.add_argument("--cache-ram", nargs='?', const=4.0, type=float, default=0, help="Use RAM pressure caching with the specified headroom threshold. If available RAM drops below the threhold the cache remove large items to free RAM. Default 4GB")
attn_group = parser.add_mutually_exclusive_group()
attn_group.add_argument("--use-split-cross-attention", action="store_true", help="Use the split cross attention optimization. Ignored when xformers is used.")
@ -150,7 +147,6 @@ parser.add_argument("--reserve-vram", type=float, default=None, help="Set the am
parser.add_argument("--async-offload", nargs='?', const=2, type=int, default=None, metavar="NUM_STREAMS", help="Use async weight offloading. An optional argument controls the amount of offload streams. Default is 2. Enabled by default on Nvidia.")
parser.add_argument("--disable-async-offload", action="store_true", help="Disable async weight offloading.")
parser.add_argument("--disable-dynamic-vram", action="store_true", help="Disable dynamic VRAM and use estimate based model loading.")
parser.add_argument("--enable-dynamic-vram", action="store_true", help="Enable dynamic VRAM on systems where it's not enabled by default.")
parser.add_argument("--force-non-blocking", action="store_true", help="Force ComfyUI to use non-blocking operations for all applicable tensors. This may improve performance on some non-Nvidia systems but can cause issues with some workflows.")
@ -264,6 +260,4 @@ else:
args.fast = set(args.fast)
def enables_dynamic_vram():
if args.enable_dynamic_vram:
return True
return not args.disable_dynamic_vram and not args.highvram and not args.gpu_only and not args.novram and not args.cpu

View File

@ -93,50 +93,6 @@ class IndexListCallbacks:
return {}
def slice_cond(cond_value, window: IndexListContextWindow, x_in: torch.Tensor, device, temporal_dim: int, temporal_scale: int=1, temporal_offset: int=0, retain_index_list: list[int]=[]):
if not (hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor)):
return None
cond_tensor = cond_value.cond
if temporal_dim >= cond_tensor.ndim:
return None
cond_size = cond_tensor.size(temporal_dim)
if temporal_scale == 1:
expected_size = x_in.size(window.dim) - temporal_offset
if cond_size != expected_size:
return None
if temporal_offset == 0 and temporal_scale == 1:
sliced = window.get_tensor(cond_tensor, device, dim=temporal_dim, retain_index_list=retain_index_list)
return cond_value._copy_with(sliced)
# skip leading latent positions that have no corresponding conditioning (e.g. reference frames)
if temporal_offset > 0:
indices = [i - temporal_offset for i in window.index_list[temporal_offset:]]
indices = [i for i in indices if 0 <= i]
else:
indices = list(window.index_list)
if not indices:
return None
if temporal_scale > 1:
scaled = []
for i in indices:
for k in range(temporal_scale):
si = i * temporal_scale + k
if si < cond_size:
scaled.append(si)
indices = scaled
if not indices:
return None
idx = tuple([slice(None)] * temporal_dim + [indices])
sliced = cond_tensor[idx].to(device)
return cond_value._copy_with(sliced)
@dataclass
class ContextSchedule:
name: str
@ -221,17 +177,10 @@ class IndexListContextHandler(ContextHandlerABC):
new_cond_item[cond_key] = result
handled = True
break
if not handled and self._model is not None:
result = self._model.resize_cond_for_context_window(
cond_key, cond_value, window, x_in, device,
retain_index_list=self.cond_retain_index_list)
if result is not None:
new_cond_item[cond_key] = result
handled = True
if handled:
continue
if isinstance(cond_value, torch.Tensor):
if (self.dim < cond_value.ndim and cond_value.size(self.dim) == x_in.size(self.dim)) or \
if (self.dim < cond_value.ndim and cond_value(self.dim) == x_in.size(self.dim)) or \
(cond_value.ndim < self.dim and cond_value.size(0) == x_in.size(self.dim)):
new_cond_item[cond_key] = window.get_tensor(cond_value, device)
# Handle audio_embed (temporal dim is 1)
@ -275,7 +224,6 @@ class IndexListContextHandler(ContextHandlerABC):
return context_windows
def execute(self, calc_cond_batch: Callable, model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
self._model = model
self.set_step(timestep, model_options)
context_windows = self.get_context_windows(model, x_in, model_options)
enumerated_context_windows = list(enumerate(context_windows))

View File

@ -209,39 +209,3 @@ def stochastic_round_quantize_nvfp4_by_block(x, per_tensor_scale, pad_16x, seed=
output_block[i:i + slice_size].copy_(block)
return output_fp4, to_blocked(output_block, flatten=False)
def stochastic_round_quantize_mxfp8_by_block(x, pad_32x, seed=0):
def roundup(x_val, multiple):
return ((x_val + multiple - 1) // multiple) * multiple
if pad_32x:
rows, cols = x.shape
padded_rows = roundup(rows, 32)
padded_cols = roundup(cols, 32)
if padded_rows != rows or padded_cols != cols:
x = torch.nn.functional.pad(x, (0, padded_cols - cols, 0, padded_rows - rows))
F8_E4M3_MAX = 448.0
E8M0_BIAS = 127
BLOCK_SIZE = 32
rows, cols = x.shape
x_blocked = x.reshape(rows, -1, BLOCK_SIZE)
max_abs = torch.amax(torch.abs(x_blocked), dim=-1)
# E8M0 block scales (power-of-2 exponents)
scale_needed = torch.clamp(max_abs.float() / F8_E4M3_MAX, min=2**(-127))
exp_biased = torch.clamp(torch.ceil(torch.log2(scale_needed)).to(torch.int32) + E8M0_BIAS, 0, 254)
block_scales_e8m0 = exp_biased.to(torch.uint8)
zero_mask = (max_abs == 0)
block_scales_f32 = (block_scales_e8m0.to(torch.int32) << 23).view(torch.float32)
block_scales_f32 = torch.where(zero_mask, torch.ones_like(block_scales_f32), block_scales_f32)
# Scale per-block then stochastic round
data_scaled = (x_blocked.float() / block_scales_f32.unsqueeze(-1)).reshape(rows, cols)
output_fp8 = stochastic_rounding(data_scaled, torch.float8_e4m3fn, seed=seed)
block_scales_e8m0 = torch.where(zero_mask, torch.zeros_like(block_scales_e8m0), block_scales_e8m0)
return output_fp8, to_blocked(block_scales_e8m0, flatten=False).view(torch.float8_e8m0fnu)

View File

@ -224,7 +224,6 @@ class Flux2(LatentFormat):
self.latent_rgb_factors_bias = [-0.0329, -0.0718, -0.0851]
self.latent_rgb_factors_reshape = lambda t: t.reshape(t.shape[0], 32, 2, 2, t.shape[-2], t.shape[-1]).permute(0, 1, 4, 2, 5, 3).reshape(t.shape[0], 32, t.shape[-2] * 2, t.shape[-1] * 2)
self.taesd_decoder_name = "taef2_decoder"
def process_in(self, latent):
return latent
@ -784,10 +783,3 @@ class ZImagePixelSpace(ChromaRadiance):
No VAE encoding/decoding — the model operates directly on RGB pixels.
"""
pass
class CogVideoX(LatentFormat):
latent_channels = 16
latent_dimensions = 3
def __init__(self):
self.scale_factor = 1.15258426

View File

@ -611,7 +611,6 @@ class AceStepDiTModel(nn.Module):
intermediate_size,
patch_size,
audio_acoustic_hidden_dim,
condition_dim=None,
layer_types=None,
sliding_window=128,
rms_norm_eps=1e-6,
@ -641,7 +640,7 @@ class AceStepDiTModel(nn.Module):
self.time_embed = TimestepEmbedding(256, hidden_size, dtype=dtype, device=device, operations=operations)
self.time_embed_r = TimestepEmbedding(256, hidden_size, dtype=dtype, device=device, operations=operations)
self.condition_embedder = Linear(condition_dim, hidden_size, dtype=dtype, device=device)
self.condition_embedder = Linear(hidden_size, hidden_size, dtype=dtype, device=device)
if layer_types is None:
layer_types = ["full_attention"] * num_layers
@ -1036,9 +1035,6 @@ class AceStepConditionGenerationModel(nn.Module):
fsq_dim=2048,
fsq_levels=[8, 8, 8, 5, 5, 5],
fsq_input_num_quantizers=1,
encoder_hidden_size=2048,
encoder_intermediate_size=6144,
encoder_num_heads=16,
audio_model=None,
dtype=None,
device=None,
@ -1058,24 +1054,24 @@ class AceStepConditionGenerationModel(nn.Module):
self.decoder = AceStepDiTModel(
in_channels, hidden_size, num_dit_layers, num_heads, num_kv_heads, head_dim,
intermediate_size, patch_size, audio_acoustic_hidden_dim, condition_dim=encoder_hidden_size,
intermediate_size, patch_size, audio_acoustic_hidden_dim,
layer_types=layer_types, sliding_window=sliding_window, rms_norm_eps=rms_norm_eps,
dtype=dtype, device=device, operations=operations
)
self.encoder = AceStepConditionEncoder(
text_hidden_dim, timbre_hidden_dim, encoder_hidden_size, num_lyric_layers, num_timbre_layers,
encoder_num_heads, num_kv_heads, head_dim, encoder_intermediate_size, rms_norm_eps,
text_hidden_dim, timbre_hidden_dim, hidden_size, num_lyric_layers, num_timbre_layers,
num_heads, num_kv_heads, head_dim, intermediate_size, rms_norm_eps,
dtype=dtype, device=device, operations=operations
)
self.tokenizer = AceStepAudioTokenizer(
audio_acoustic_hidden_dim, encoder_hidden_size, pool_window_size, fsq_dim=fsq_dim, fsq_levels=fsq_levels, fsq_input_num_quantizers=fsq_input_num_quantizers, num_layers=num_tokenizer_layers, head_dim=head_dim, rms_norm_eps=rms_norm_eps,
audio_acoustic_hidden_dim, hidden_size, pool_window_size, fsq_dim=fsq_dim, fsq_levels=fsq_levels, fsq_input_num_quantizers=fsq_input_num_quantizers, num_layers=num_tokenizer_layers, head_dim=head_dim, rms_norm_eps=rms_norm_eps,
dtype=dtype, device=device, operations=operations
)
self.detokenizer = AudioTokenDetokenizer(
encoder_hidden_size, pool_window_size, audio_acoustic_hidden_dim, num_layers=2, head_dim=head_dim,
hidden_size, pool_window_size, audio_acoustic_hidden_dim, num_layers=2, head_dim=head_dim,
dtype=dtype, device=device, operations=operations
)
self.null_condition_emb = nn.Parameter(torch.empty(1, 1, encoder_hidden_size, dtype=dtype, device=device))
self.null_condition_emb = nn.Parameter(torch.empty(1, 1, hidden_size, dtype=dtype, device=device))
def prepare_condition(
self,

View File

@ -136,7 +136,16 @@ class ResBlock(nn.Module):
ops.Linear(c_hidden, c),
)
self.gammas = nn.Parameter(torch.zeros(6), requires_grad=False)
self.gammas = nn.Parameter(torch.zeros(6), requires_grad=True)
# Init weights
def _basic_init(module):
if isinstance(module, nn.Linear) or isinstance(module, nn.Conv2d):
torch.nn.init.xavier_uniform_(module.weight)
if module.bias is not None:
nn.init.constant_(module.bias, 0)
self.apply(_basic_init)
def _norm(self, x, norm):
return norm(x.permute(0, 2, 3, 1)).permute(0, 3, 1, 2)

View File

@ -1,573 +0,0 @@
# CogVideoX 3D Transformer - ported to ComfyUI native ops
# Architecture reference: diffusers CogVideoXTransformer3DModel
# Style reference: comfy/ldm/wan/model.py
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from comfy.ldm.modules.attention import optimized_attention
import comfy.patcher_extension
import comfy.ldm.common_dit
def _get_1d_rotary_pos_embed(dim, pos, theta=10000.0):
"""Returns (cos, sin) each with shape [seq_len, dim].
Frequencies are computed at dim//2 resolution then repeat_interleaved
to full dim, matching CogVideoX's interleaved (real, imag) pair format.
"""
freqs = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float32, device=pos.device) / dim))
angles = torch.outer(pos.float(), freqs.float())
cos = angles.cos().repeat_interleave(2, dim=-1).float()
sin = angles.sin().repeat_interleave(2, dim=-1).float()
return (cos, sin)
def apply_rotary_emb(x, freqs_cos_sin):
"""Apply CogVideoX rotary embedding to query or key tensor.
x: [B, heads, seq_len, head_dim]
freqs_cos_sin: (cos, sin) each [seq_len, head_dim//2]
Uses interleaved pair rotation (same as diffusers CogVideoX/Flux).
head_dim is reshaped to (-1, 2) pairs, rotated, then flattened back.
"""
cos, sin = freqs_cos_sin
cos = cos[None, None, :, :].to(x.device)
sin = sin[None, None, :, :].to(x.device)
# Interleaved pairs: [B, H, S, D] -> [B, H, S, D//2, 2] -> (real, imag)
x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1)
x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(3)
return (x.float() * cos + x_rotated.float() * sin).to(x.dtype)
def get_timestep_embedding(timesteps, dim, flip_sin_to_cos=True, downscale_freq_shift=0, scale=1, max_period=10000):
half = dim // 2
freqs = torch.exp(-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32, device=timesteps.device) / half)
args = timesteps[:, None].float() * freqs[None] * scale
embedding = torch.cat([torch.sin(args), torch.cos(args)], dim=-1)
if flip_sin_to_cos:
embedding = torch.cat([embedding[:, half:], embedding[:, :half]], dim=-1)
if dim % 2:
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
return embedding
def get_3d_sincos_pos_embed(embed_dim, spatial_size, temporal_size, spatial_interpolation_scale=1.0, temporal_interpolation_scale=1.0, device=None):
if isinstance(spatial_size, int):
spatial_size = (spatial_size, spatial_size)
grid_w = torch.arange(spatial_size[0], dtype=torch.float32, device=device) / spatial_interpolation_scale
grid_h = torch.arange(spatial_size[1], dtype=torch.float32, device=device) / spatial_interpolation_scale
grid_t = torch.arange(temporal_size, dtype=torch.float32, device=device) / temporal_interpolation_scale
grid_t, grid_h, grid_w = torch.meshgrid(grid_t, grid_h, grid_w, indexing="ij")
embed_dim_spatial = 2 * (embed_dim // 3)
embed_dim_temporal = embed_dim // 3
pos_embed_spatial = _get_2d_sincos_pos_embed(embed_dim_spatial, grid_h, grid_w, device=device)
pos_embed_temporal = _get_1d_sincos_pos_embed(embed_dim_temporal, grid_t[:, 0, 0], device=device)
T, H, W = grid_t.shape
pos_embed_temporal = pos_embed_temporal.unsqueeze(1).unsqueeze(1).expand(-1, H, W, -1)
pos_embed = torch.cat([pos_embed_temporal, pos_embed_spatial], dim=-1)
return pos_embed
def _get_2d_sincos_pos_embed(embed_dim, grid_h, grid_w, device=None):
T, H, W = grid_h.shape
half_dim = embed_dim // 2
pos_h = _get_1d_sincos_pos_embed(half_dim, grid_h.reshape(-1), device=device).reshape(T, H, W, half_dim)
pos_w = _get_1d_sincos_pos_embed(half_dim, grid_w.reshape(-1), device=device).reshape(T, H, W, half_dim)
return torch.cat([pos_h, pos_w], dim=-1)
def _get_1d_sincos_pos_embed(embed_dim, pos, device=None):
half = embed_dim // 2
freqs = torch.exp(-math.log(10000.0) * torch.arange(start=0, end=half, dtype=torch.float32, device=device) / half)
args = pos.float().reshape(-1)[:, None] * freqs[None]
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
if embed_dim % 2:
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
return embedding
class CogVideoXPatchEmbed(nn.Module):
def __init__(self, patch_size=2, patch_size_t=None, in_channels=16, dim=1920,
text_dim=4096, bias=True, sample_width=90, sample_height=60,
sample_frames=49, temporal_compression_ratio=4,
max_text_seq_length=226, spatial_interpolation_scale=1.875,
temporal_interpolation_scale=1.0, use_positional_embeddings=True,
use_learned_positional_embeddings=True,
device=None, dtype=None, operations=None):
super().__init__()
self.patch_size = patch_size
self.patch_size_t = patch_size_t
self.dim = dim
self.sample_height = sample_height
self.sample_width = sample_width
self.sample_frames = sample_frames
self.temporal_compression_ratio = temporal_compression_ratio
self.max_text_seq_length = max_text_seq_length
self.spatial_interpolation_scale = spatial_interpolation_scale
self.temporal_interpolation_scale = temporal_interpolation_scale
self.use_positional_embeddings = use_positional_embeddings
self.use_learned_positional_embeddings = use_learned_positional_embeddings
if patch_size_t is None:
self.proj = operations.Conv2d(in_channels, dim, kernel_size=patch_size, stride=patch_size, bias=bias, device=device, dtype=dtype)
else:
self.proj = operations.Linear(in_channels * patch_size * patch_size * patch_size_t, dim, device=device, dtype=dtype)
self.text_proj = operations.Linear(text_dim, dim, device=device, dtype=dtype)
if use_positional_embeddings or use_learned_positional_embeddings:
persistent = use_learned_positional_embeddings
pos_embedding = self._get_positional_embeddings(sample_height, sample_width, sample_frames)
self.register_buffer("pos_embedding", pos_embedding, persistent=persistent)
def _get_positional_embeddings(self, sample_height, sample_width, sample_frames, device=None):
post_patch_height = sample_height // self.patch_size
post_patch_width = sample_width // self.patch_size
post_time_compression_frames = (sample_frames - 1) // self.temporal_compression_ratio + 1
if self.patch_size_t is not None:
post_time_compression_frames = post_time_compression_frames // self.patch_size_t
num_patches = post_patch_height * post_patch_width * post_time_compression_frames
pos_embedding = get_3d_sincos_pos_embed(
self.dim,
(post_patch_width, post_patch_height),
post_time_compression_frames,
self.spatial_interpolation_scale,
self.temporal_interpolation_scale,
device=device,
)
pos_embedding = pos_embedding.reshape(-1, self.dim)
joint_pos_embedding = pos_embedding.new_zeros(
1, self.max_text_seq_length + num_patches, self.dim, requires_grad=False
)
joint_pos_embedding.data[:, self.max_text_seq_length:].copy_(pos_embedding)
return joint_pos_embedding
def forward(self, text_embeds, image_embeds):
input_dtype = text_embeds.dtype
text_embeds = self.text_proj(text_embeds.to(self.text_proj.weight.dtype)).to(input_dtype)
batch_size, num_frames, channels, height, width = image_embeds.shape
proj_dtype = self.proj.weight.dtype
if self.patch_size_t is None:
image_embeds = image_embeds.reshape(-1, channels, height, width)
image_embeds = self.proj(image_embeds.to(proj_dtype)).to(input_dtype)
image_embeds = image_embeds.view(batch_size, num_frames, *image_embeds.shape[1:])
image_embeds = image_embeds.flatten(3).transpose(2, 3)
image_embeds = image_embeds.flatten(1, 2)
else:
p = self.patch_size
p_t = self.patch_size_t
image_embeds = image_embeds.permute(0, 1, 3, 4, 2)
image_embeds = image_embeds.reshape(
batch_size, num_frames // p_t, p_t, height // p, p, width // p, p, channels
)
image_embeds = image_embeds.permute(0, 1, 3, 5, 7, 2, 4, 6).flatten(4, 7).flatten(1, 3)
image_embeds = self.proj(image_embeds.to(proj_dtype)).to(input_dtype)
embeds = torch.cat([text_embeds, image_embeds], dim=1).contiguous()
if self.use_positional_embeddings or self.use_learned_positional_embeddings:
text_seq_length = text_embeds.shape[1]
num_image_patches = image_embeds.shape[1]
if self.use_learned_positional_embeddings:
image_pos = self.pos_embedding[
:, self.max_text_seq_length:self.max_text_seq_length + num_image_patches
].to(device=embeds.device, dtype=embeds.dtype)
else:
image_pos = get_3d_sincos_pos_embed(
self.dim,
(width // self.patch_size, height // self.patch_size),
num_image_patches // ((height // self.patch_size) * (width // self.patch_size)),
self.spatial_interpolation_scale,
self.temporal_interpolation_scale,
device=embeds.device,
).reshape(1, num_image_patches, self.dim).to(dtype=embeds.dtype)
# Build joint: zeros for text + sincos for image
joint_pos = torch.zeros(1, text_seq_length + num_image_patches, self.dim, device=embeds.device, dtype=embeds.dtype)
joint_pos[:, text_seq_length:] = image_pos
embeds = embeds + joint_pos
return embeds
class CogVideoXLayerNormZero(nn.Module):
def __init__(self, time_dim, dim, elementwise_affine=True, eps=1e-5, bias=True,
device=None, dtype=None, operations=None):
super().__init__()
self.silu = nn.SiLU()
self.linear = operations.Linear(time_dim, 6 * dim, bias=bias, device=device, dtype=dtype)
self.norm = operations.LayerNorm(dim, eps=eps, elementwise_affine=elementwise_affine, device=device, dtype=dtype)
def forward(self, hidden_states, encoder_hidden_states, temb):
shift, scale, gate, enc_shift, enc_scale, enc_gate = self.linear(self.silu(temb)).chunk(6, dim=1)
hidden_states = self.norm(hidden_states) * (1 + scale)[:, None, :] + shift[:, None, :]
encoder_hidden_states = self.norm(encoder_hidden_states) * (1 + enc_scale)[:, None, :] + enc_shift[:, None, :]
return hidden_states, encoder_hidden_states, gate[:, None, :], enc_gate[:, None, :]
class CogVideoXAdaLayerNorm(nn.Module):
def __init__(self, time_dim, dim, elementwise_affine=True, eps=1e-5,
device=None, dtype=None, operations=None):
super().__init__()
self.silu = nn.SiLU()
self.linear = operations.Linear(time_dim, 2 * dim, device=device, dtype=dtype)
self.norm = operations.LayerNorm(dim, eps=eps, elementwise_affine=elementwise_affine, device=device, dtype=dtype)
def forward(self, x, temb):
temb = self.linear(self.silu(temb))
shift, scale = temb.chunk(2, dim=1)
x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]
return x
class CogVideoXBlock(nn.Module):
def __init__(self, dim, num_heads, head_dim, time_dim,
eps=1e-5, ff_inner_dim=None, ff_bias=True,
device=None, dtype=None, operations=None):
super().__init__()
self.dim = dim
self.num_heads = num_heads
self.head_dim = head_dim
self.norm1 = CogVideoXLayerNormZero(time_dim, dim, eps=eps, device=device, dtype=dtype, operations=operations)
# Self-attention (joint text + latent)
self.q = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype)
self.k = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype)
self.v = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype)
self.norm_q = operations.LayerNorm(head_dim, eps=1e-6, elementwise_affine=True, device=device, dtype=dtype)
self.norm_k = operations.LayerNorm(head_dim, eps=1e-6, elementwise_affine=True, device=device, dtype=dtype)
self.attn_out = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype)
self.norm2 = CogVideoXLayerNormZero(time_dim, dim, eps=eps, device=device, dtype=dtype, operations=operations)
# Feed-forward (GELU approximate)
inner_dim = ff_inner_dim or dim * 4
self.ff_proj = operations.Linear(dim, inner_dim, bias=ff_bias, device=device, dtype=dtype)
self.ff_out = operations.Linear(inner_dim, dim, bias=ff_bias, device=device, dtype=dtype)
def forward(self, hidden_states, encoder_hidden_states, temb, image_rotary_emb=None, transformer_options=None):
if transformer_options is None:
transformer_options = {}
text_seq_length = encoder_hidden_states.size(1)
# Norm & modulate
norm_hidden, norm_encoder, gate_msa, enc_gate_msa = self.norm1(hidden_states, encoder_hidden_states, temb)
# Joint self-attention
qkv_input = torch.cat([norm_encoder, norm_hidden], dim=1)
b, s, _ = qkv_input.shape
n, d = self.num_heads, self.head_dim
q = self.q(qkv_input).view(b, s, n, d)
k = self.k(qkv_input).view(b, s, n, d)
v = self.v(qkv_input)
q = self.norm_q(q).view(b, s, n, d)
k = self.norm_k(k).view(b, s, n, d)
# Apply rotary embeddings to image tokens only (diffusers format: [B, heads, seq, head_dim])
if image_rotary_emb is not None:
q_img = q[:, text_seq_length:].transpose(1, 2) # [B, heads, img_seq, head_dim]
k_img = k[:, text_seq_length:].transpose(1, 2)
q_img = apply_rotary_emb(q_img, image_rotary_emb)
k_img = apply_rotary_emb(k_img, image_rotary_emb)
q = torch.cat([q[:, :text_seq_length], q_img.transpose(1, 2)], dim=1)
k = torch.cat([k[:, :text_seq_length], k_img.transpose(1, 2)], dim=1)
attn_out = optimized_attention(
q.reshape(b, s, n * d),
k.reshape(b, s, n * d),
v,
heads=self.num_heads,
transformer_options=transformer_options,
)
attn_out = self.attn_out(attn_out)
attn_encoder, attn_hidden = attn_out.split([text_seq_length, s - text_seq_length], dim=1)
hidden_states = hidden_states + gate_msa * attn_hidden
encoder_hidden_states = encoder_hidden_states + enc_gate_msa * attn_encoder
# Norm & modulate for FF
norm_hidden, norm_encoder, gate_ff, enc_gate_ff = self.norm2(hidden_states, encoder_hidden_states, temb)
# Feed-forward (GELU on concatenated text + latent)
ff_input = torch.cat([norm_encoder, norm_hidden], dim=1)
ff_output = self.ff_out(F.gelu(self.ff_proj(ff_input), approximate="tanh"))
hidden_states = hidden_states + gate_ff * ff_output[:, text_seq_length:]
encoder_hidden_states = encoder_hidden_states + enc_gate_ff * ff_output[:, :text_seq_length]
return hidden_states, encoder_hidden_states
class CogVideoXTransformer3DModel(nn.Module):
def __init__(self,
num_attention_heads=30,
attention_head_dim=64,
in_channels=16,
out_channels=16,
flip_sin_to_cos=True,
freq_shift=0,
time_embed_dim=512,
ofs_embed_dim=None,
text_embed_dim=4096,
num_layers=30,
dropout=0.0,
attention_bias=True,
sample_width=90,
sample_height=60,
sample_frames=49,
patch_size=2,
patch_size_t=None,
temporal_compression_ratio=4,
max_text_seq_length=226,
spatial_interpolation_scale=1.875,
temporal_interpolation_scale=1.0,
use_rotary_positional_embeddings=False,
use_learned_positional_embeddings=False,
patch_bias=True,
image_model=None,
device=None,
dtype=None,
operations=None,
):
super().__init__()
self.dtype = dtype
dim = num_attention_heads * attention_head_dim
self.dim = dim
self.num_attention_heads = num_attention_heads
self.attention_head_dim = attention_head_dim
self.in_channels = in_channels
self.out_channels = out_channels
self.patch_size = patch_size
self.patch_size_t = patch_size_t
self.max_text_seq_length = max_text_seq_length
self.use_rotary_positional_embeddings = use_rotary_positional_embeddings
# 1. Patch embedding
self.patch_embed = CogVideoXPatchEmbed(
patch_size=patch_size,
patch_size_t=patch_size_t,
in_channels=in_channels,
dim=dim,
text_dim=text_embed_dim,
bias=patch_bias,
sample_width=sample_width,
sample_height=sample_height,
sample_frames=sample_frames,
temporal_compression_ratio=temporal_compression_ratio,
max_text_seq_length=max_text_seq_length,
spatial_interpolation_scale=spatial_interpolation_scale,
temporal_interpolation_scale=temporal_interpolation_scale,
use_positional_embeddings=not use_rotary_positional_embeddings,
use_learned_positional_embeddings=use_learned_positional_embeddings,
device=device, dtype=torch.float32, operations=operations,
)
# 2. Time embedding
self.time_proj_dim = dim
self.time_proj_flip = flip_sin_to_cos
self.time_proj_shift = freq_shift
self.time_embedding_linear_1 = operations.Linear(dim, time_embed_dim, device=device, dtype=dtype)
self.time_embedding_act = nn.SiLU()
self.time_embedding_linear_2 = operations.Linear(time_embed_dim, time_embed_dim, device=device, dtype=dtype)
# Optional OFS embedding (CogVideoX 1.5 I2V)
self.ofs_proj_dim = ofs_embed_dim
if ofs_embed_dim:
self.ofs_embedding_linear_1 = operations.Linear(ofs_embed_dim, ofs_embed_dim, device=device, dtype=dtype)
self.ofs_embedding_act = nn.SiLU()
self.ofs_embedding_linear_2 = operations.Linear(ofs_embed_dim, ofs_embed_dim, device=device, dtype=dtype)
else:
self.ofs_embedding_linear_1 = None
# 3. Transformer blocks
self.blocks = nn.ModuleList([
CogVideoXBlock(
dim=dim,
num_heads=num_attention_heads,
head_dim=attention_head_dim,
time_dim=time_embed_dim,
eps=1e-5,
device=device, dtype=dtype, operations=operations,
)
for _ in range(num_layers)
])
self.norm_final = operations.LayerNorm(dim, eps=1e-5, elementwise_affine=True, device=device, dtype=dtype)
# 4. Output
self.norm_out = CogVideoXAdaLayerNorm(
time_dim=time_embed_dim, dim=dim, eps=1e-5,
device=device, dtype=dtype, operations=operations,
)
if patch_size_t is None:
output_dim = patch_size * patch_size * out_channels
else:
output_dim = patch_size * patch_size * patch_size_t * out_channels
self.proj_out = operations.Linear(dim, output_dim, device=device, dtype=dtype)
self.spatial_interpolation_scale = spatial_interpolation_scale
self.temporal_interpolation_scale = temporal_interpolation_scale
self.temporal_compression_ratio = temporal_compression_ratio
def forward(self, x, timestep, context, ofs=None, transformer_options=None, **kwargs):
if transformer_options is None:
transformer_options = {}
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
self._forward,
self,
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
).execute(x, timestep, context, ofs, transformer_options, **kwargs)
def _forward(self, x, timestep, context, ofs=None, transformer_options=None, **kwargs):
if transformer_options is None:
transformer_options = {}
# ComfyUI passes [B, C, T, H, W]
batch_size, channels, t, h, w = x.shape
# Pad to patch size (temporal + spatial), same pattern as WAN
p_t = self.patch_size_t if self.patch_size_t is not None else 1
x = comfy.ldm.common_dit.pad_to_patch_size(x, (p_t, self.patch_size, self.patch_size))
# CogVideoX expects [B, T, C, H, W]
x = x.permute(0, 2, 1, 3, 4)
batch_size, num_frames, channels, height, width = x.shape
# Time embedding
t_emb = get_timestep_embedding(timestep, self.time_proj_dim, self.time_proj_flip, self.time_proj_shift)
t_emb = t_emb.to(dtype=x.dtype)
emb = self.time_embedding_linear_2(self.time_embedding_act(self.time_embedding_linear_1(t_emb)))
if self.ofs_embedding_linear_1 is not None and ofs is not None:
ofs_emb = get_timestep_embedding(ofs, self.ofs_proj_dim, self.time_proj_flip, self.time_proj_shift)
ofs_emb = ofs_emb.to(dtype=x.dtype)
ofs_emb = self.ofs_embedding_linear_2(self.ofs_embedding_act(self.ofs_embedding_linear_1(ofs_emb)))
emb = emb + ofs_emb
# Patch embedding
hidden_states = self.patch_embed(context, x)
text_seq_length = context.shape[1]
encoder_hidden_states = hidden_states[:, :text_seq_length]
hidden_states = hidden_states[:, text_seq_length:]
# Rotary embeddings (if used)
image_rotary_emb = None
if self.use_rotary_positional_embeddings:
post_patch_height = height // self.patch_size
post_patch_width = width // self.patch_size
if self.patch_size_t is None:
post_time = num_frames
else:
post_time = num_frames // self.patch_size_t
image_rotary_emb = self._get_rotary_emb(post_patch_height, post_patch_width, post_time, device=x.device)
# Transformer blocks
for i, block in enumerate(self.blocks):
hidden_states, encoder_hidden_states = block(
hidden_states=hidden_states,
encoder_hidden_states=encoder_hidden_states,
temb=emb,
image_rotary_emb=image_rotary_emb,
transformer_options=transformer_options,
)
hidden_states = self.norm_final(hidden_states)
# Output projection
hidden_states = self.norm_out(hidden_states, temb=emb)
hidden_states = self.proj_out(hidden_states)
# Unpatchify
p = self.patch_size
p_t = self.patch_size_t
if p_t is None:
output = hidden_states.reshape(batch_size, num_frames, height // p, width // p, -1, p, p)
output = output.permute(0, 1, 4, 2, 5, 3, 6).flatten(5, 6).flatten(3, 4)
else:
output = hidden_states.reshape(
batch_size, (num_frames + p_t - 1) // p_t, height // p, width // p, -1, p_t, p, p
)
output = output.permute(0, 1, 5, 4, 2, 6, 3, 7).flatten(6, 7).flatten(4, 5).flatten(1, 2)
# Back to ComfyUI format [B, C, T, H, W] and crop padding
output = output.permute(0, 2, 1, 3, 4)[:, :, :t, :h, :w]
return output
def _get_rotary_emb(self, h, w, t, device):
"""Compute CogVideoX 3D rotary positional embeddings.
For CogVideoX 1.5 (patch_size_t != None): uses "slice" mode — grid positions
are integer arange computed at max_size, then sliced to actual size.
For CogVideoX 1.0 (patch_size_t == None): uses "linspace" mode with crop coords
scaled by spatial_interpolation_scale.
"""
d = self.attention_head_dim
dim_t = d // 4
dim_h = d // 8 * 3
dim_w = d // 8 * 3
if self.patch_size_t is not None:
# CogVideoX 1.5: "slice" mode — positions are simple integer indices
# Compute at max(sample_size, actual_size) then slice to actual
base_h = self.patch_embed.sample_height // self.patch_size
base_w = self.patch_embed.sample_width // self.patch_size
max_h = max(base_h, h)
max_w = max(base_w, w)
grid_h = torch.arange(max_h, device=device, dtype=torch.float32)
grid_w = torch.arange(max_w, device=device, dtype=torch.float32)
grid_t = torch.arange(t, device=device, dtype=torch.float32)
else:
# CogVideoX 1.0: "linspace" mode with interpolation scale
grid_h = torch.linspace(0, h - 1, h, device=device, dtype=torch.float32) * self.spatial_interpolation_scale
grid_w = torch.linspace(0, w - 1, w, device=device, dtype=torch.float32) * self.spatial_interpolation_scale
grid_t = torch.arange(t, device=device, dtype=torch.float32)
freqs_t = _get_1d_rotary_pos_embed(dim_t, grid_t)
freqs_h = _get_1d_rotary_pos_embed(dim_h, grid_h)
freqs_w = _get_1d_rotary_pos_embed(dim_w, grid_w)
t_cos, t_sin = freqs_t
h_cos, h_sin = freqs_h
w_cos, w_sin = freqs_w
# Slice to actual size (for "slice" mode where grids may be larger)
t_cos, t_sin = t_cos[:t], t_sin[:t]
h_cos, h_sin = h_cos[:h], h_sin[:h]
w_cos, w_sin = w_cos[:w], w_sin[:w]
# Broadcast and concatenate into [T*H*W, head_dim]
t_cos = t_cos[:, None, None, :].expand(-1, h, w, -1)
t_sin = t_sin[:, None, None, :].expand(-1, h, w, -1)
h_cos = h_cos[None, :, None, :].expand(t, -1, w, -1)
h_sin = h_sin[None, :, None, :].expand(t, -1, w, -1)
w_cos = w_cos[None, None, :, :].expand(t, h, -1, -1)
w_sin = w_sin[None, None, :, :].expand(t, h, -1, -1)
cos = torch.cat([t_cos, h_cos, w_cos], dim=-1).reshape(t * h * w, -1)
sin = torch.cat([t_sin, h_sin, w_sin], dim=-1).reshape(t * h * w, -1)
return (cos, sin)

Some files were not shown because too many files have changed in this diff Show More