Merge branch 'master' into cursor/oss-spec-cloud-fields-1c03

* fix(api-io): serialize MultiCombo multi_select as object config
* fix: remove dead code and redundant top-level keys from MultiCombo serialization
* fix: correct skip warning to mention comfy_entrypoint, remove nonexistent NODES_LIST
* fix: validate MultiCombo list values against options individually
* fix: gate multiselect validation on schema config, improve error message, add tests

---------

Co-authored-by: Ni-zav <ni-zav@users.noreply.github.com>
Co-authored-by: guill <jacob.e.segal@gmail.com>

.github/workflows/openapi-lint.yml

@@ -0,0 +1,31 @@
+name: OpenAPI Lint
+
+on:
+  pull_request:
+    paths:
+      - 'openapi.yaml'
+      - '.spectral.yaml'
+      - '.github/workflows/openapi-lint.yml'
+
+permissions:
+  contents: read
+
+jobs:
+  spectral:
+    name: Run Spectral
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+
+      - name: Install Spectral
+        run: npm install -g @stoplight/spectral-cli@6
+
+      - name: Lint openapi.yaml
+        run: spectral lint openapi.yaml --ruleset .spectral.yaml --fail-severity=error

.spectral.yaml

@@ -0,0 +1,91 @@
+extends:
+  - spectral:oas
+
+# Severity levels: error, warn, info, hint, off
+# Rules from the built-in "spectral:oas" ruleset are active by default.
+# Below we tune severity and add custom rules for our conventions.
+#
+# This ruleset mirrors Comfy-Org/cloud/.spectral.yaml so specs across the
+# organization are linted against a single consistent standard.
+
+rules:
+  # -----------------------------------------------------------------------
+  # Built-in rule severity overrides
+  # -----------------------------------------------------------------------
+  operation-operationId: error
+  operation-description: warn
+  operation-tag-defined: error
+  info-contact: off
+  info-description: warn
+  no-eval-in-markdown: error
+  no-$ref-siblings: error
+
+  # -----------------------------------------------------------------------
+  # Custom rules: naming conventions
+  # -----------------------------------------------------------------------
+
+  # Property names should be snake_case
+  property-name-snake-case:
+    description: Property names must be snake_case
+    severity: warn
+    given: "$.components.schemas.*.properties[*]~"
+    then:
+      function: pattern
+      functionOptions:
+        match: "^[a-z][a-z0-9]*(_[a-z0-9]+)*$"
+
+  # Operation IDs should be camelCase
+  operation-id-camel-case:
+    description: Operation IDs must be camelCase
+    severity: warn
+    given: "$.paths.*.*.operationId"
+    then:
+      function: pattern
+      functionOptions:
+        match: "^[a-z][a-zA-Z0-9]*$"
+
+  # -----------------------------------------------------------------------
+  # Custom rules: response conventions
+  # -----------------------------------------------------------------------
+
+  # Error responses (4xx, 5xx) should use a consistent shape
+  error-response-schema:
+    description: Error responses should reference a standard error schema
+    severity: hint
+    given: "$.paths.*.*.responses[?(@property >= '400' && @property < '600')].content['application/json'].schema"
+    then:
+      field: "$ref"
+      function: truthy
+
+  # All 2xx responses with JSON body should have a schema
+  response-schema-defined:
+    description: Success responses with JSON content should define a schema
+    severity: warn
+    given: "$.paths.*.*.responses[?(@property >= '200' && @property < '300')].content['application/json']"
+    then:
+      field: schema
+      function: truthy
+
+  # -----------------------------------------------------------------------
+  # Custom rules: best practices
+  # -----------------------------------------------------------------------
+
+  # Path parameters must have a description
+  path-param-description:
+    description: Path parameters should have a description
+    severity: warn
+    given:
+      - "$.paths.*.parameters[?(@.in == 'path')]"
+      - "$.paths.*.*.parameters[?(@.in == 'path')]"
+    then:
+      field: description
+      function: truthy
+
+  # Schemas should have a description
+  schema-description:
+    description: Component schemas should have a description
+    severity: hint
+    given: "$.components.schemas.*"
+    then:
+      field: description
+      function: truthy

README.md

@@ -133,7 +133,7 @@ Workflow examples can be found on the [Examples page](https://comfyanonymous.git
 ComfyUI follows a weekly release cycle targeting Monday but this regularly changes because of model releases or large changes to the codebase. There are three interconnected repositories:
 
 1. **[ComfyUI Core](https://github.com/comfyanonymous/ComfyUI)**
-   - Releases a new stable version (e.g., v0.7.0) roughly every week.
+   - Releases a new major stable version (e.g., v0.7.0) roughly every 2 weeks.
    - Starting from v0.4.0 patch versions will be used for fixes backported onto the current stable release.
    - Minor versions will be used for releases off the master branch.
    - Patch versions may still be used for releases on the master branch in cases where a backport would not make sense.

comfy/k_diffusion/sampling.py

@@ -1810,3 +1810,102 @@ def sample_sa_solver(model, x, sigmas, extra_args=None, callback=None, disable=F
 def sample_sa_solver_pece(model, x, sigmas, extra_args=None, callback=None, disable=False, tau_func=None, s_noise=1.0, noise_sampler=None, predictor_order=3, corrector_order=4, simple_order_2=False):
     """Stochastic Adams Solver with PECE (Predict–Evaluate–Correct–Evaluate) mode (NeurIPS 2023)."""
     return sample_sa_solver(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, tau_func=tau_func, s_noise=s_noise, noise_sampler=noise_sampler, predictor_order=predictor_order, corrector_order=corrector_order, use_pece=True, simple_order_2=simple_order_2)
+
+
+@torch.no_grad()
+def sample_ar_video(model, x, sigmas, extra_args=None, callback=None, disable=None,
+                    num_frame_per_block=1):
+    """
+    Autoregressive video sampler: block-by-block denoising with KV cache
+    and flow-match re-noising for Causal Forcing / Self-Forcing models.
+
+    Requires a Causal-WAN compatible model (diffusion_model must expose
+    init_kv_caches / init_crossattn_caches) and 5-D latents [B,C,T,H,W].
+
+    All AR-loop parameters are passed via the SamplerARVideo node, not read
+    from the checkpoint or transformer_options.
+    """
+    extra_args = {} if extra_args is None else extra_args
+    model_options = extra_args.get("model_options", {})
+    transformer_options = model_options.get("transformer_options", {})
+
+    if x.ndim != 5:
+        raise ValueError(
+            f"ar_video sampler requires 5-D video latents [B,C,T,H,W], got {x.ndim}-D tensor with shape {x.shape}. "
+            "This sampler is only compatible with autoregressive video models (e.g. Causal-WAN)."
+        )
+
+    inner_model = model.inner_model.inner_model
+    causal_model = inner_model.diffusion_model
+
+    if not (hasattr(causal_model, "init_kv_caches") and hasattr(causal_model, "init_crossattn_caches")):
+        raise TypeError(
+            "ar_video sampler requires a Causal-WAN compatible model whose diffusion_model "
+            "exposes init_kv_caches() and init_crossattn_caches(). The loaded checkpoint "
+            "does not support this interface — choose a different sampler."
+        )
+
+    seed = extra_args.get("seed", 0)
+
+    bs, c, lat_t, lat_h, lat_w = x.shape
+    frame_seq_len = -(-lat_h // 2) * -(-lat_w // 2) # ceiling division
+    num_blocks = -(-lat_t // num_frame_per_block)   # ceiling division
+    device = x.device
+    model_dtype = inner_model.get_dtype()
+
+    kv_caches = causal_model.init_kv_caches(bs, lat_t * frame_seq_len, device, model_dtype)
+    crossattn_caches = causal_model.init_crossattn_caches(bs, device, model_dtype)
+
+    output = torch.zeros_like(x)
+    s_in = x.new_ones([x.shape[0]])
+    current_start_frame = 0
+    num_sigma_steps = len(sigmas) - 1
+    total_real_steps = num_blocks * num_sigma_steps
+    step_count = 0
+
+    try:
+        for block_idx in trange(num_blocks, disable=disable):
+            bf = min(num_frame_per_block, lat_t - current_start_frame)
+            fs, fe = current_start_frame, current_start_frame + bf
+            noisy_input = x[:, :, fs:fe]
+
+            ar_state = {
+                "start_frame": current_start_frame,
+                "kv_caches": kv_caches,
+                "crossattn_caches": crossattn_caches,
+            }
+            transformer_options["ar_state"] = ar_state
+
+            for i in range(num_sigma_steps):
+                denoised = model(noisy_input, sigmas[i] * s_in, **extra_args)
+
+                if callback is not None:
+                    scaled_i = step_count * num_sigma_steps // total_real_steps
+                    callback({"x": noisy_input, "i": scaled_i, "sigma": sigmas[i],
+                              "sigma_hat": sigmas[i], "denoised": denoised})
+
+                if sigmas[i + 1] == 0:
+                    noisy_input = denoised
+                else:
+                    sigma_next = sigmas[i + 1]
+                    torch.manual_seed(seed + block_idx * 1000 + i)
+                    fresh_noise = torch.randn_like(denoised)
+                    noisy_input = (1.0 - sigma_next) * denoised + sigma_next * fresh_noise
+
+                    for cache in kv_caches:
+                        cache["end"] -= bf * frame_seq_len
+
+                step_count += 1
+
+            output[:, :, fs:fe] = noisy_input
+
+            for cache in kv_caches:
+                cache["end"] -= bf * frame_seq_len
+            zero_sigma = sigmas.new_zeros([1])
+            _ = model(noisy_input, zero_sigma * s_in, **extra_args)
+
+            current_start_frame += bf
+    finally:
+        transformer_options.pop("ar_state", None)
+
+    return output

comfy/ldm/wan/ar_model.py

@@ -0,0 +1,276 @@
+"""
+CausalWanModel: Wan 2.1 backbone with KV-cached causal self-attention for
+autoregressive (frame-by-frame) video generation via Causal Forcing.
+
+Weight-compatible with the standard WanModel -- same layer names, same shapes.
+The difference is purely in the forward pass: this model processes one temporal
+block at a time and maintains a KV cache across blocks.
+
+Reference: https://github.com/thu-ml/Causal-Forcing
+"""
+
+import torch
+import torch.nn as nn
+
+from comfy.ldm.modules.attention import optimized_attention
+from comfy.ldm.flux.math import apply_rope1
+from comfy.ldm.wan.model import (
+    sinusoidal_embedding_1d,
+    repeat_e,
+    WanModel,
+    WanAttentionBlock,
+)
+import comfy.ldm.common_dit
+import comfy.model_management
+
+
+class CausalWanSelfAttention(nn.Module):
+    """Self-attention with KV cache support for autoregressive inference."""
+
+    def __init__(self, dim, num_heads, window_size=(-1, -1), qk_norm=True,
+                 eps=1e-6, operation_settings={}):
+        assert dim % num_heads == 0
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.qk_norm = qk_norm
+        self.eps = eps
+
+        ops = operation_settings.get("operations")
+        device = operation_settings.get("device")
+        dtype = operation_settings.get("dtype")
+
+        self.q = ops.Linear(dim, dim, device=device, dtype=dtype)
+        self.k = ops.Linear(dim, dim, device=device, dtype=dtype)
+        self.v = ops.Linear(dim, dim, device=device, dtype=dtype)
+        self.o = ops.Linear(dim, dim, device=device, dtype=dtype)
+        self.norm_q = ops.RMSNorm(dim, eps=eps, elementwise_affine=True, device=device, dtype=dtype) if qk_norm else nn.Identity()
+        self.norm_k = ops.RMSNorm(dim, eps=eps, elementwise_affine=True, device=device, dtype=dtype) if qk_norm else nn.Identity()
+
+    def forward(self, x, freqs, kv_cache=None, transformer_options={}):
+        b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim
+
+        q = apply_rope1(self.norm_q(self.q(x)).view(b, s, n, d), freqs)
+        k = apply_rope1(self.norm_k(self.k(x)).view(b, s, n, d), freqs)
+        v = self.v(x).view(b, s, n, d)
+
+        if kv_cache is None:
+            x = optimized_attention(
+                q.view(b, s, n * d),
+                k.view(b, s, n * d),
+                v.view(b, s, n * d),
+                heads=self.num_heads,
+                transformer_options=transformer_options,
+            )
+        else:
+            end = kv_cache["end"]
+            new_end = end + s
+
+            # Roped K and plain V go into cache
+            kv_cache["k"][:, end:new_end] = k
+            kv_cache["v"][:, end:new_end] = v
+            kv_cache["end"] = new_end
+
+            x = optimized_attention(
+                q.view(b, s, n * d),
+                kv_cache["k"][:, :new_end].view(b, new_end, n * d),
+                kv_cache["v"][:, :new_end].view(b, new_end, n * d),
+                heads=self.num_heads,
+                transformer_options=transformer_options,
+            )
+
+        x = self.o(x)
+        return x
+
+
+class CausalWanAttentionBlock(WanAttentionBlock):
+    """Transformer block with KV-cached self-attention and cross-attention caching."""
+
+    def __init__(self, cross_attn_type, dim, ffn_dim, num_heads,
+                 window_size=(-1, -1), qk_norm=True, cross_attn_norm=False,
+                 eps=1e-6, operation_settings={}):
+        super().__init__(cross_attn_type, dim, ffn_dim, num_heads,
+                         window_size, qk_norm, cross_attn_norm, eps,
+                         operation_settings=operation_settings)
+        self.self_attn = CausalWanSelfAttention(
+            dim, num_heads, window_size, qk_norm, eps,
+            operation_settings=operation_settings)
+
+    def forward(self, x, e, freqs, context, context_img_len=257,
+                kv_cache=None, crossattn_cache=None, transformer_options={}):
+        if e.ndim < 4:
+            e = (comfy.model_management.cast_to(self.modulation, dtype=x.dtype, device=x.device) + e).chunk(6, dim=1)
+        else:
+            e = (comfy.model_management.cast_to(self.modulation, dtype=x.dtype, device=x.device).unsqueeze(0) + e).unbind(2)
+
+        # Self-attention with optional KV cache
+        x = x.contiguous()
+        y = self.self_attn(
+            torch.addcmul(repeat_e(e[0], x), self.norm1(x), 1 + repeat_e(e[1], x)),
+            freqs, kv_cache=kv_cache, transformer_options=transformer_options)
+        x = torch.addcmul(x, y, repeat_e(e[2], x))
+        del y
+
+        # Cross-attention with optional caching
+        if crossattn_cache is not None and crossattn_cache.get("is_init"):
+            q = self.cross_attn.norm_q(self.cross_attn.q(self.norm3(x)))
+            x_ca = optimized_attention(
+                q, crossattn_cache["k"], crossattn_cache["v"],
+                heads=self.num_heads, transformer_options=transformer_options)
+            x = x + self.cross_attn.o(x_ca)
+        else:
+            x = x + self.cross_attn(self.norm3(x), context, context_img_len=context_img_len, transformer_options=transformer_options)
+            if crossattn_cache is not None:
+                crossattn_cache["k"] = self.cross_attn.norm_k(self.cross_attn.k(context))
+                crossattn_cache["v"] = self.cross_attn.v(context)
+                crossattn_cache["is_init"] = True
+
+        # FFN
+        y = self.ffn(torch.addcmul(repeat_e(e[3], x), self.norm2(x), 1 + repeat_e(e[4], x)))
+        x = torch.addcmul(x, y, repeat_e(e[5], x))
+        return x
+
+
+class CausalWanModel(WanModel):
+    """
+    Wan 2.1 diffusion backbone with causal KV-cache support.
+
+    Same weight structure as WanModel -- loads identical state dicts.
+    Adds forward_block() for frame-by-frame autoregressive inference.
+    """
+
+    def __init__(self,
+                 model_type='t2v',
+                 patch_size=(1, 2, 2),
+                 text_len=512,
+                 in_dim=16,
+                 dim=2048,
+                 ffn_dim=8192,
+                 freq_dim=256,
+                 text_dim=4096,
+                 out_dim=16,
+                 num_heads=16,
+                 num_layers=32,
+                 window_size=(-1, -1),
+                 qk_norm=True,
+                 cross_attn_norm=True,
+                 eps=1e-6,
+                 image_model=None,
+                 device=None,
+                 dtype=None,
+                 operations=None):
+        super().__init__(
+            model_type=model_type, patch_size=patch_size, text_len=text_len,
+            in_dim=in_dim, dim=dim, ffn_dim=ffn_dim, freq_dim=freq_dim,
+            text_dim=text_dim, out_dim=out_dim, num_heads=num_heads,
+            num_layers=num_layers, window_size=window_size, qk_norm=qk_norm,
+            cross_attn_norm=cross_attn_norm, eps=eps, image_model=image_model,
+            wan_attn_block_class=CausalWanAttentionBlock,
+            device=device, dtype=dtype, operations=operations)
+
+    def forward_block(self, x, timestep, context, start_frame,
+                      kv_caches, crossattn_caches, clip_fea=None):
+        """
+        Forward one temporal block for autoregressive inference.
+
+        Args:
+            x: [B, C, block_frames, H, W] input latent for the current block
+            timestep: [B, block_frames] per-frame timesteps
+            context: [B, L, text_dim] raw text embeddings (pre-text_embedding)
+            start_frame: temporal frame index for RoPE offset
+            kv_caches: list of per-layer KV cache dicts
+            crossattn_caches: list of per-layer cross-attention cache dicts
+            clip_fea: optional CLIP features for I2V
+
+        Returns:
+            flow_pred: [B, C_out, block_frames, H, W] flow prediction
+        """
+        x = comfy.ldm.common_dit.pad_to_patch_size(x, self.patch_size)
+        bs, c, t, h, w = x.shape
+
+        x = self.patch_embedding(x.float()).to(x.dtype)
+        grid_sizes = x.shape[2:]
+        x = x.flatten(2).transpose(1, 2)
+
+        # Per-frame time embedding
+        e = self.time_embedding(
+            sinusoidal_embedding_1d(self.freq_dim, timestep.flatten()).to(dtype=x.dtype))
+        e = e.reshape(timestep.shape[0], -1, e.shape[-1])
+        e0 = self.time_projection(e).unflatten(2, (6, self.dim))
+
+        # Text embedding (reuses crossattn_cache after first block)
+        context = self.text_embedding(context)
+
+        context_img_len = None
+        if clip_fea is not None and self.img_emb is not None:
+            context_clip = self.img_emb(clip_fea)
+            context = torch.concat([context_clip, context], dim=1)
+            context_img_len = clip_fea.shape[-2]
+
+        # RoPE for current block's temporal position
+        freqs = self.rope_encode(t, h, w, t_start=start_frame, device=x.device, dtype=x.dtype)
+
+        # Transformer blocks
+        for i, block in enumerate(self.blocks):
+            x = block(x, e=e0, freqs=freqs, context=context,
+                      context_img_len=context_img_len,
+                      kv_cache=kv_caches[i],
+                      crossattn_cache=crossattn_caches[i])
+
+        # Head
+        x = self.head(x, e)
+
+        # Unpatchify
+        x = self.unpatchify(x, grid_sizes)
+        return x[:, :, :t, :h, :w]
+
+    def init_kv_caches(self, batch_size, max_seq_len, device, dtype):
+        """Create fresh KV caches for all layers."""
+        caches = []
+        for _ in range(self.num_layers):
+            caches.append({
+                "k": torch.zeros(batch_size, max_seq_len, self.num_heads, self.head_dim, device=device, dtype=dtype),
+                "v": torch.zeros(batch_size, max_seq_len, self.num_heads, self.head_dim, device=device, dtype=dtype),
+                "end": 0,
+            })
+        return caches
+
+    def init_crossattn_caches(self, batch_size, device, dtype):
+        """Create fresh cross-attention caches for all layers."""
+        caches = []
+        for _ in range(self.num_layers):
+            caches.append({"is_init": False})
+        return caches
+
+    def reset_kv_caches(self, kv_caches):
+        """Reset KV caches to empty (reuse allocated memory)."""
+        for cache in kv_caches:
+            cache["end"] = 0
+
+    def reset_crossattn_caches(self, crossattn_caches):
+        """Reset cross-attention caches."""
+        for cache in crossattn_caches:
+            cache["is_init"] = False
+
+    @property
+    def head_dim(self):
+        return self.dim // self.num_heads
+
+    def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, **kwargs):
+        ar_state = transformer_options.get("ar_state")
+        if ar_state is not None:
+            bs = x.shape[0]
+            block_frames = x.shape[2]
+            t_per_frame = timestep.unsqueeze(1).expand(bs, block_frames)
+            return self.forward_block(
+                x=x, timestep=t_per_frame, context=context,
+                start_frame=ar_state["start_frame"],
+                kv_caches=ar_state["kv_caches"],
+                crossattn_caches=ar_state["crossattn_caches"],
+                clip_fea=clip_fea,
+            )
+
+        return super().forward(x, timestep, context, clip_fea=clip_fea,
+                               time_dim_concat=time_dim_concat,
+                               transformer_options=transformer_options, **kwargs)

comfy/model_base.py

@@ -42,6 +42,7 @@ import comfy.ldm.cosmos.predict2
 import comfy.ldm.lumina.model
 import comfy.ldm.wan.model
 import comfy.ldm.wan.model_animate
+import comfy.ldm.wan.ar_model
 import comfy.ldm.hunyuan3d.model
 import comfy.ldm.hidream.model
 import comfy.ldm.chroma.model
@@ -1365,6 +1366,13 @@ class WAN21(BaseModel):
         return out
 
 
+class WAN21_CausalAR(WAN21):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super(WAN21, self).__init__(model_config, model_type, device=device,
+                                    unet_model=comfy.ldm.wan.ar_model.CausalWanModel)
+        self.image_to_video = False
+
+
 class WAN21_Vace(WAN21):
     def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
         super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.VaceWanModel)

comfy/supported_models.py

@@ -1167,6 +1167,25 @@ class WAN21_T2V(supported_models_base.BASE):
         t5_detect = comfy.text_encoders.sd3_clip.t5_xxl_detect(state_dict, "{}umt5xxl.transformer.".format(pref))
         return supported_models_base.ClipTarget(comfy.text_encoders.wan.WanT5Tokenizer, comfy.text_encoders.wan.te(**t5_detect))
 
+class WAN21_CausalAR_T2V(WAN21_T2V):
+    unet_config = {
+        "image_model": "wan2.1",
+        "model_type": "t2v",
+        "causal_ar": True,
+    }
+
+    sampling_settings = {
+        "shift": 5.0,
+    }
+
+    def __init__(self, unet_config):
+        super().__init__(unet_config)
+        self.unet_config.pop("causal_ar", None)
+
+    def get_model(self, state_dict, prefix="", device=None):
+        return model_base.WAN21_CausalAR(self, device=device)
+
+
 class WAN21_I2V(WAN21_T2V):
     unet_config = {
         "image_model": "wan2.1",
@@ -1929,6 +1948,7 @@ models = [
     ZImage,
     Lumina2,
     WAN22_T2V,
+    WAN21_CausalAR_T2V,
     WAN21_T2V,
     WAN21_I2V,
     WAN21_FunControl2V,

comfy_api/latest/_io.py

@@ -395,7 +395,6 @@ class Combo(ComfyTypeIO):
 @comfytype(io_type="COMBO")
 class MultiCombo(ComfyTypeI):
     '''Multiselect Combo input (dropdown for selecting potentially more than one value).'''
-    # TODO: something is wrong with the serialization, frontend does not recognize it as multiselect
     Type = list[str]
     class Input(Combo.Input):
         def __init__(self, id: str, options: list[str], display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
@@ -408,12 +407,14 @@ class MultiCombo(ComfyTypeI):
             self.default: list[str]
 
         def as_dict(self):
-            to_return = super().as_dict() | prune_dict({
-                "multi_select": self.multiselect,
-                "placeholder": self.placeholder,
-                "chip": self.chip,
+            # Frontend expects `multi_select` to be an object config (not a boolean).
+            # Keep top-level `multiselect` from Combo.Input for backwards compatibility.
+            return super().as_dict() | prune_dict({
+                "multi_select": prune_dict({
+                    "placeholder": self.placeholder,
+                    "chip": self.chip,
+                }),
             })
-            return to_return
 
 @comfytype(io_type="IMAGE")
 class Image(ComfyTypeIO):

comfy_api_nodes/apis/luma.py

@@ -1,15 +1,12 @@
 from __future__ import annotations
 
-
-import torch
-
 from enum import Enum
 from typing import Optional, Union
 
+import torch
 from pydantic import BaseModel, Field, confloat
 
 
-
 class LumaIO:
     LUMA_REF = "LUMA_REF"
     LUMA_CONCEPTS = "LUMA_CONCEPTS"
@@ -183,13 +180,13 @@ class LumaAssets(BaseModel):
 
 
 class LumaImageRef(BaseModel):
-    '''Used for image gen'''
+    """Used for image gen"""
     url: str = Field(..., description='The URL of the image reference')
     weight: confloat(ge=0.0, le=1.0) = Field(..., description='The weight of the image reference')
 
 
 class LumaImageReference(BaseModel):
-    '''Used for video gen'''
+    """Used for video gen"""
     type: Optional[str] = Field('image', description='Input type, defaults to image')
     url: str = Field(..., description='The URL of the image')
 
@@ -251,3 +248,32 @@ class LumaGeneration(BaseModel):
     assets: Optional[LumaAssets] = Field(None, description='The assets of the generation')
     model: str = Field(..., description='The model used for the generation')
     request: Union[LumaGenerationRequest, LumaImageGenerationRequest] = Field(..., description="The request used for the generation")
+
+
+class Luma2ImageRef(BaseModel):
+    url: str | None = None
+    data: str | None = None
+    media_type: str | None = None
+
+
+class Luma2GenerationRequest(BaseModel):
+    prompt: str = Field(..., min_length=1, max_length=6000)
+    model: str | None = None
+    type: str | None = None
+    aspect_ratio: str | None = None
+    style: str | None = None
+    output_format: str | None = None
+    web_search: bool | None = None
+    image_ref: list[Luma2ImageRef] | None = None
+    source: Luma2ImageRef | None = None
+
+
+class Luma2Generation(BaseModel):
+    id: str | None = None
+    type: str | None = None
+    state: str | None = None
+    model: str | None = None
+    created_at: str | None = None
+    output: list[LumaImageReference] | None = None
+    failure_reason: str | None = None
+    failure_code: str | None = None

comfy_api_nodes/apis/openai.py

@@ -56,14 +56,14 @@ class ModelResponseProperties(BaseModel):
     instructions: str | None = Field(None)
     max_output_tokens: int | None = Field(None)
     model: str | None = Field(None)
-    temperature: float | None = Field(1, description="Controls randomness in the response", ge=0.0, le=2.0)
+    temperature: float | None = Field(None, description="Controls randomness in the response", ge=0.0, le=2.0)
     top_p: float | None = Field(
-        1,
+        None,
         description="Controls diversity of the response via nucleus sampling",
         ge=0.0,
         le=1.0,
     )
-    truncation: str | None = Field("disabled", description="Allowed values: 'auto' or 'disabled'")
+    truncation: str | None = Field(None, description="Allowed values: 'auto' or 'disabled'")
 
 
 class ResponseProperties(BaseModel):

comfy_api_nodes/nodes_luma.py

@@ -1,10 +1,11 @@
-from typing import Optional
-
 import torch
 from typing_extensions import override
 
-from comfy_api.latest import IO, ComfyExtension
+from comfy_api.latest import IO, ComfyExtension, Input
 from comfy_api_nodes.apis.luma import (
+    Luma2Generation,
+    Luma2GenerationRequest,
+    Luma2ImageRef,
     LumaAspectRatio,
     LumaCharacterRef,
     LumaConceptChain,
@@ -30,6 +31,7 @@ from comfy_api_nodes.util import (
     download_url_to_video_output,
     poll_op,
     sync_op,
+    upload_image_to_comfyapi,
     upload_images_to_comfyapi,
     validate_string,
 )
@@ -212,9 +214,9 @@ class LumaImageGenerationNode(IO.ComfyNode):
         aspect_ratio: str,
         seed,
         style_image_weight: float,
-        image_luma_ref: Optional[LumaReferenceChain] = None,
-        style_image: Optional[torch.Tensor] = None,
-        character_image: Optional[torch.Tensor] = None,
+        image_luma_ref: LumaReferenceChain | None = None,
+        style_image: torch.Tensor | None = None,
+        character_image: torch.Tensor | None = None,
     ) -> IO.NodeOutput:
         validate_string(prompt, strip_whitespace=True, min_length=3)
         # handle image_luma_ref
@@ -434,7 +436,7 @@ class LumaTextToVideoGenerationNode(IO.ComfyNode):
         duration: str,
         loop: bool,
         seed,
-        luma_concepts: Optional[LumaConceptChain] = None,
+        luma_concepts: LumaConceptChain | None = None,
     ) -> IO.NodeOutput:
         validate_string(prompt, strip_whitespace=False, min_length=3)
         duration = duration if model != LumaVideoModel.ray_1_6 else None
@@ -533,7 +535,6 @@ class LumaImageToVideoGenerationNode(IO.ComfyNode):
             ],
             is_api_node=True,
             price_badge=PRICE_BADGE_VIDEO,
-
         )
 
     @classmethod
@@ -644,6 +645,293 @@ PRICE_BADGE_VIDEO = IO.PriceBadge(
 )
 
 
+def _luma2_uni1_common_inputs(max_image_refs: int) -> list:
+    return [
+        IO.Combo.Input(
+            "style",
+            options=["auto", "manga"],
+            default="auto",
+            tooltip="Style preset. 'auto' picks based on the prompt; "
+            "'manga' applies a manga/anime aesthetic and requires a portrait "
+            "aspect ratio (2:3, 9:16, 1:2, 1:3).",
+        ),
+        IO.Boolean.Input(
+            "web_search",
+            default=False,
+            tooltip="Search the web for visual references before generating.",
+        ),
+        IO.Autogrow.Input(
+            "image_ref",
+            template=IO.Autogrow.TemplateNames(
+                IO.Image.Input("image"),
+                names=[f"image_{i}" for i in range(1, max_image_refs + 1)],
+                min=0,
+            ),
+            optional=True,
+            tooltip=f"Up to {max_image_refs} reference images for style/content guidance.",
+        ),
+    ]
+
+
+async def _luma2_upload_image_refs(
+    cls: type[IO.ComfyNode],
+    refs: dict | None,
+    max_count: int,
+) -> list[Luma2ImageRef] | None:
+    if not refs:
+        return None
+    out: list[Luma2ImageRef] = []
+    for key in refs:
+        url = await upload_image_to_comfyapi(cls, refs[key])
+        out.append(Luma2ImageRef(url=url))
+    if len(out) > max_count:
+        raise ValueError(f"Maximum {max_count} reference images are allowed.")
+    return out or None
+
+
+async def _luma2_submit_and_poll(
+    cls: type[IO.ComfyNode],
+    request: Luma2GenerationRequest,
+) -> Input.Image:
+    initial = await sync_op(
+        cls,
+        ApiEndpoint(path="/proxy/luma_2/generations", method="POST"),
+        response_model=Luma2Generation,
+        data=request,
+    )
+    if not initial.id:
+        raise RuntimeError("Luma 2 API did not return a generation id.")
+    final = await poll_op(
+        cls,
+        ApiEndpoint(path=f"/proxy/luma_2/generations/{initial.id}", method="GET"),
+        response_model=Luma2Generation,
+        status_extractor=lambda r: r.state,
+        progress_extractor=lambda r: None,
+    )
+    if not final.output:
+        msg = final.failure_reason or "no output returned"
+        raise RuntimeError(f"Luma 2 generation failed: {msg}")
+    url = final.output[0].url
+    if not url:
+        raise RuntimeError("Luma 2 generation completed without an output URL.")
+    return await download_url_to_image_tensor(url)
+
+
+class LumaImageNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="LumaImageNode2",
+            display_name="Luma UNI-1 Image",
+            category="api node/image/Luma",
+            description="Generate images from text using the Luma UNI-1 model.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Text description of the desired image. 1–6000 characters.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "uni-1",
+                            [
+                                IO.Combo.Input(
+                                    "aspect_ratio",
+                                    options=[
+                                        "auto",
+                                        "3:1",
+                                        "2:1",
+                                        "16:9",
+                                        "3:2",
+                                        "1:1",
+                                        "2:3",
+                                        "9:16",
+                                        "1:2",
+                                        "1:3",
+                                    ],
+                                    default="auto",
+                                    tooltip="Output image aspect ratio. 'auto' lets "
+                                    "the model pick based on the prompt.",
+                                ),
+                                *_luma2_uni1_common_inputs(max_image_refs=9),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            "uni-1-max",
+                            [
+                                IO.Combo.Input(
+                                    "aspect_ratio",
+                                    options=[
+                                        "auto",
+                                        "3:1",
+                                        "2:1",
+                                        "16:9",
+                                        "3:2",
+                                        "1:1",
+                                        "2:3",
+                                        "9:16",
+                                        "1:2",
+                                        "1:3",
+                                    ],
+                                    default="auto",
+                                    tooltip="Output image aspect ratio. 'auto' lets "
+                                    "the model pick based on the prompt.",
+                                ),
+                                *_luma2_uni1_common_inputs(max_image_refs=9),
+                            ],
+                        ),
+                    ],
+                    tooltip="Model to use for generation.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    control_after_generate=True,
+                    tooltip="Seed controls whether the node should re-run; "
+                    "results are non-deterministic regardless of seed.",
+                ),
+            ],
+            outputs=[IO.Image.Output()],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+            price_badge=IO.PriceBadge(
+                depends_on=IO.PriceBadgeDepends(widgets=["model"], input_groups=["model.image_ref"]),
+                expr="""
+                (
+                  $m := widgets.model;
+                  $refs := $lookup(inputGroups, "model.image_ref");
+                  $base := $m = "uni-1-max" ? 0.1 : 0.0404;
+                  {"type":"usd","usd": $round($base + 0.003 * $refs, 4)}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        model: dict,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=6000)
+        aspect_ratio = model["aspect_ratio"]
+        style = model["style"]
+        allowed_manga_ratios = {"2:3", "9:16", "1:2", "1:3"}
+        if style == "manga" and aspect_ratio != "auto" and aspect_ratio not in allowed_manga_ratios:
+            raise ValueError(
+                f"'manga' style requires a portrait aspect ratio "
+                f"({', '.join(sorted(allowed_manga_ratios))}) or 'auto'; got '{aspect_ratio}'."
+            )
+        request = Luma2GenerationRequest(
+            prompt=prompt,
+            model=model["model"],
+            type="image",
+            aspect_ratio=aspect_ratio if aspect_ratio != "auto" else None,
+            style=style if style != "auto" else None,
+            output_format="png",
+            web_search=model["web_search"],
+            image_ref=await _luma2_upload_image_refs(cls, model.get("image_ref"), max_count=9),
+        )
+        return IO.NodeOutput(await _luma2_submit_and_poll(cls, request))
+
+
+class LumaImageEditNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="LumaImageEditNode2",
+            display_name="Luma UNI-1 Image Edit",
+            category="api node/image/Luma",
+            description="Edit an existing image with a text prompt using the Luma UNI-1 model.",
+            inputs=[
+                IO.Image.Input(
+                    "source",
+                    tooltip="Source image to edit.",
+                ),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Description of the desired edit. 1–6000 characters.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "uni-1",
+                            _luma2_uni1_common_inputs(max_image_refs=8),
+                        ),
+                        IO.DynamicCombo.Option(
+                            "uni-1-max",
+                            _luma2_uni1_common_inputs(max_image_refs=8),
+                        ),
+                    ],
+                    tooltip="Model to use for editing.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    control_after_generate=True,
+                    tooltip="Seed controls whether the node should re-run; "
+                    "results are non-deterministic regardless of seed.",
+                ),
+            ],
+            outputs=[IO.Image.Output()],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+            price_badge=IO.PriceBadge(
+                depends_on=IO.PriceBadgeDepends(widgets=["model"], input_groups=["model.image_ref"]),
+                expr="""
+                (
+                  $m := widgets.model;
+                  $refs := $lookup(inputGroups, "model.image_ref");
+                  $base := $m = "uni-1-max" ? 0.103 : 0.0434;
+                  {"type":"usd","usd": $round($base + 0.003 * $refs, 4)}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        source: Input.Image,
+        prompt: str,
+        model: dict,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, min_length=1, max_length=6000)
+        request = Luma2GenerationRequest(
+            prompt=prompt,
+            model=model["model"],
+            type="image_edit",
+            source=Luma2ImageRef(url=await upload_image_to_comfyapi(cls, source)),
+            style=model["style"] if model["style"] != "auto" else None,
+            output_format="png",
+            web_search=model["web_search"],
+            image_ref=await _luma2_upload_image_refs(cls, model.get("image_ref"), max_count=8),
+        )
+        return IO.NodeOutput(await _luma2_submit_and_poll(cls, request))
+
+
 class LumaExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -654,6 +942,8 @@ class LumaExtension(ComfyExtension):
             LumaImageToVideoGenerationNode,
             LumaReferenceNode,
             LumaConceptsNode,
+            LumaImageNode,
+            LumaImageEditNode,
         ]
 
 

comfy_api_nodes/nodes_openai.py

@@ -39,16 +39,18 @@ STARTING_POINT_ID_PATTERN = r"<starting_point_id:(.*)>"
 
 
 class SupportedOpenAIModel(str, Enum):
-    o4_mini = "o4-mini"
-    o1 = "o1"
-    o3 = "o3"
-    o1_pro = "o1-pro"
-    gpt_4_1 = "gpt-4.1"
-    gpt_4_1_mini = "gpt-4.1-mini"
-    gpt_4_1_nano = "gpt-4.1-nano"
+    gpt_5_5_pro = "gpt-5.5-pro"
+    gpt_5_5 = "gpt-5.5"
     gpt_5 = "gpt-5"
     gpt_5_mini = "gpt-5-mini"
     gpt_5_nano = "gpt-5-nano"
+    gpt_4_1 = "gpt-4.1"
+    gpt_4_1_mini = "gpt-4.1-mini"
+    gpt_4_1_nano = "gpt-4.1-nano"
+    o4_mini = "o4-mini"
+    o3 = "o3"
+    o1_pro = "o1-pro"
+    o1 = "o1"
 
 
 async def validate_and_cast_response(response, timeout: int = None) -> torch.Tensor:
@@ -739,6 +741,16 @@ class OpenAIChatNode(IO.ComfyNode):
                     "usd": [0.002, 0.008],
                     "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
                   }
+                  : $contains($m, "gpt-5.5-pro") ? {
+                    "type": "list_usd",
+                    "usd": [0.03, 0.18],
+                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
+                  }
+                  : $contains($m, "gpt-5.5") ? {
+                    "type": "list_usd",
+                    "usd": [0.005, 0.03],
+                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
+                  }
                   : $contains($m, "gpt-5-nano") ? {
                     "type": "list_usd",
                     "usd": [0.00005, 0.0004],

comfy_extras/nodes_ar_video.py

@@ -0,0 +1,84 @@
+"""
+ComfyUI nodes for autoregressive video generation (Causal Forcing, Self-Forcing, etc.).
+  - EmptyARVideoLatent: create 5D [B, C, T, H, W] video latent tensors
+  - SamplerARVideo: SAMPLER for the block-by-block autoregressive denoising loop
+"""
+
+import torch
+from typing_extensions import override
+
+import comfy.model_management
+import comfy.samplers
+from comfy_api.latest import ComfyExtension, io
+
+
+class EmptyARVideoLatent(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="EmptyARVideoLatent",
+            category="latent/video",
+            inputs=[
+                io.Int.Input("width", default=832, min=16, max=8192, step=16),
+                io.Int.Input("height", default=480, min=16, max=8192, step=16),
+                io.Int.Input("length", default=81, min=1, max=1024, step=4),
+                io.Int.Input("batch_size", default=1, min=1, max=64),
+            ],
+            outputs=[
+                io.Latent.Output(display_name="LATENT"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, width, height, length, batch_size) -> io.NodeOutput:
+        lat_t = ((length - 1) // 4) + 1
+        latent = torch.zeros(
+            [batch_size, 16, lat_t, height // 8, width // 8],
+            device=comfy.model_management.intermediate_device(),
+        )
+        return io.NodeOutput({"samples": latent})
+
+
+class SamplerARVideo(io.ComfyNode):
+    """Sampler for autoregressive video models (Causal Forcing, Self-Forcing).
+
+    All AR-loop parameters are owned by this node so they live in the workflow.
+    Add new widgets here as the AR sampler grows new options.
+    """
+
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="SamplerARVideo",
+            display_name="Sampler AR Video",
+            category="sampling/custom_sampling/samplers",
+            inputs=[
+                io.Int.Input(
+                    "num_frame_per_block",
+                    default=1, min=1, max=64,
+                    tooltip="Frames per autoregressive block. 1 = framewise, "
+                            "3 = chunkwise. Must match the checkpoint's training mode.",
+                ),
+            ],
+            outputs=[io.Sampler.Output()],
+        )
+
+    @classmethod
+    def execute(cls, num_frame_per_block) -> io.NodeOutput:
+        extra_options = {
+            "num_frame_per_block": num_frame_per_block,
+        }
+        return io.NodeOutput(comfy.samplers.ksampler("ar_video", extra_options))
+
+
+class ARVideoExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[io.ComfyNode]]:
+        return [
+            EmptyARVideoLatent,
+            SamplerARVideo,
+        ]
+
+
+async def comfy_entrypoint() -> ARVideoExtension:
+    return ARVideoExtension()

comfy_extras/nodes_images.py

@@ -716,7 +716,7 @@ class SplitImageToTileList(IO.ComfyNode):
     def get_grid_coords(width, height, tile_width, tile_height, overlap):
         coords = []
         stride_x = round(max(tile_width * 0.25, tile_width - overlap))
-        stride_y = round(max(tile_width * 0.25, tile_height - overlap))
+        stride_y = round(max(tile_height * 0.25, tile_height - overlap))
 
         y = 0
         while y < height:

comfy_extras/nodes_lt_audio.py

@@ -147,7 +147,6 @@ class LTXVEmptyLatentAudio(io.ComfyNode):
 
         z_channels = audio_vae.latent_channels
         audio_freq = audio_vae.first_stage_model.latent_frequency_bins
-        sampling_rate = int(audio_vae.first_stage_model.sample_rate)
 
         num_audio_latents = audio_vae.first_stage_model.num_of_latents_from_frames(frames_number, frame_rate)
 
@@ -159,7 +158,6 @@ class LTXVEmptyLatentAudio(io.ComfyNode):
         return io.NodeOutput(
             {
                 "samples": audio_latents,
-                "sample_rate": sampling_rate,
                 "type": "audio",
             }
         )

comfy_extras/nodes_primitive.py

@@ -9,7 +9,8 @@ class String(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PrimitiveString",
-            display_name="String",
+            search_aliases=["text", "string", "text box", "prompt"],
+            display_name="Text String",
             category="utils/primitive",
             inputs=[
                 io.String.Input("value"),
@@ -27,7 +28,8 @@ class StringMultiline(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="PrimitiveStringMultiline",
-            display_name="String (Multiline)",
+            search_aliases=["text", "string", "text multiline", "string multiline", "text box", "prompt"],
+            display_name="Text String (Multiline)",
             category="utils/primitive",
             essentials_category="Basics",
             inputs=[

comfy_extras/nodes_string.py

@@ -10,9 +10,9 @@ class StringConcatenate(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringConcatenate",
-            display_name="Text Concatenate",
-            category="utils/string",
-            search_aliases=["Concatenate", "text concat", "join text", "merge text", "combine strings", "concat", "concatenate", "append text", "combine text", "string"],
+            search_aliases=["concatenate", "text concat", "join text", "merge text", "combine strings", "string concat", "append text", "combine text"],
+            display_name="Concatenate Text",
+            category="text",
             inputs=[
                 io.String.Input("string_a", multiline=True),
                 io.String.Input("string_b", multiline=True),
@@ -33,9 +33,9 @@ class StringSubstring(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringSubstring",
-            search_aliases=["Substring", "extract text", "text portion"],
-            display_name="Text Substring",
-            category="utils/string",
+            search_aliases=["substring", "extract text", "text portion"],
+            display_name="Substring",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.Int.Input("start"),
@@ -58,7 +58,7 @@ class StringLength(io.ComfyNode):
             node_id="StringLength",
             search_aliases=["character count", "text size", "string length"],
             display_name="Text Length",
-            category="utils/string",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
             ],
@@ -77,9 +77,9 @@ class CaseConverter(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="CaseConverter",
-            search_aliases=["Case Converter", "text case", "uppercase", "lowercase", "capitalize"],
-            display_name="Text Case Converter",
-            category="utils/string",
+            search_aliases=["case converter", "text case", "uppercase", "lowercase", "capitalize"],
+            display_name="Convert Text Case",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.Combo.Input("mode", options=["UPPERCASE", "lowercase", "Capitalize", "Title Case"]),
@@ -110,9 +110,9 @@ class StringTrim(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringTrim",
-            search_aliases=["Trim", "clean whitespace", "remove whitespace", "strip"],
-            display_name="Text Trim",
-            category="utils/string",
+            search_aliases=["trim", "clean whitespace", "remove whitespace", "remove spaces","strip"],
+            display_name="Trim Text",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.Combo.Input("mode", options=["Both", "Left", "Right"]),
@@ -141,9 +141,9 @@ class StringReplace(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringReplace",
-            search_aliases=["Replace", "find and replace", "substitute", "swap text"],
-            display_name="Text Replace",
-            category="utils/string",
+            search_aliases=["replace", "find and replace", "substitute", "swap text"],
+            display_name="Replace Text",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.String.Input("find", multiline=True),
@@ -164,9 +164,9 @@ class StringContains(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringContains",
-            search_aliases=["Contains", "text includes", "string includes"],
-            display_name="Text Contains",
-            category="utils/string",
+            search_aliases=["contains", "text includes", "string includes"],
+            display_name="Contains Text",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.String.Input("substring", multiline=True),
@@ -192,9 +192,9 @@ class StringCompare(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="StringCompare",
-            search_aliases=["Compare", "text match", "string equals", "starts with", "ends with"],
-            display_name="Text Compare",
-            category="utils/string",
+            search_aliases=["compare", "text match", "string equals", "starts with", "ends with"],
+            display_name="Compare Text",
+            category="text",
             inputs=[
                 io.String.Input("string_a", multiline=True),
                 io.String.Input("string_b", multiline=True),
@@ -228,9 +228,9 @@ class RegexMatch(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="RegexMatch",
-            search_aliases=["Regex Match", "regex", "pattern match", "text contains", "string match"],
-            display_name="Text Match",
-            category="utils/string",
+            search_aliases=["regex match", "regex", "pattern match", "text contains", "string match"],
+            display_name="Match Text",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.String.Input("regex_pattern", multiline=True),
@@ -269,9 +269,9 @@ class RegexExtract(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="RegexExtract",
-            search_aliases=["Regex Extract", "regex", "pattern extract", "text parser", "parse text"],
-            display_name="Text Extract Substring",
-            category="utils/string",
+            search_aliases=["regex extract", "regex", "pattern extract", "text parser", "parse text"],
+            display_name="Extract Text",
+            category="text",
             inputs=[
                 io.String.Input("string", multiline=True),
                 io.String.Input("regex_pattern", multiline=True),
@@ -344,9 +344,9 @@ class RegexReplace(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="RegexReplace",
-            search_aliases=["Regex Replace", "regex", "pattern replace", "regex replace", "substitution"],
-            display_name="Text Replace (Regex)",
-            category="utils/string",
+            search_aliases=["regex replace", "regex", "pattern replace", "substitution"],
+            display_name="Replace Text (Regex)",
+            category="text",
             description="Find and replace text using regex patterns.",
             inputs=[
                 io.String.Input("string", multiline=True),
@@ -381,8 +381,8 @@ class JsonExtractString(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="JsonExtractString",
-            display_name="Extract String from JSON",
-            category="utils/string",
+            display_name="Extract Text from JSON",
+            category="text",
             search_aliases=["json", "extract json", "parse json", "json value", "read json"],
             inputs=[
                 io.String.Input("json_string", multiline=True),

execution.py

@@ -1019,7 +1019,12 @@ async def validate_inputs(prompt_id, prompt, item, validated, visiting=None):
                         combo_options = extra_info.get("options", [])
                     else:
                         combo_options = input_type
-                    if val not in combo_options:
+                    is_multiselect = extra_info.get("multiselect", False)
+                    if is_multiselect and isinstance(val, list):
+                        invalid_vals = [v for v in val if v not in combo_options]
+                    else:
+                        invalid_vals = [val] if val not in combo_options else []
+                    if invalid_vals:
                         input_config = info
                         list_info = ""
 
@@ -1034,7 +1039,7 @@ async def validate_inputs(prompt_id, prompt, item, validated, visiting=None):
                         error = {
                             "type": "value_not_in_list",
                             "message": "Value not in list",
-                            "details": f"{x}: '{val}' not in {list_info}",
+                            "details": f"{x}: {', '.join(repr(v) for v in invalid_vals)} not in {list_info}",
                             "extra_info": {
                                 "input_name": x,
                                 "input_config": input_config,

nodes.py

@@ -2262,7 +2262,7 @@ async def load_custom_node(module_path: str, ignore=set(), module_parent="custom
                 logging.warning(f"Error while calling comfy_entrypoint in {module_path}: {e}")
                 return False
         else:
-            logging.warning(f"Skip {module_path} module for custom nodes due to the lack of NODE_CLASS_MAPPINGS or NODES_LIST (need one).")
+            logging.warning(f"Skip {module_path} module for custom nodes due to the lack of NODE_CLASS_MAPPINGS or comfy_entrypoint (need one).")
             return False
     except Exception as e:
         logging.warning(traceback.format_exc())
@@ -2412,6 +2412,7 @@ async def init_builtin_extra_nodes():
         "nodes_nop.py",
         "nodes_kandinsky5.py",
         "nodes_wanmove.py",
+        "nodes_ar_video.py",
         "nodes_image_compare.py",
         "nodes_zimage.py",
         "nodes_glsl.py",

openapi.yaml

@@ -631,7 +631,7 @@ paths:
       operationId: getFeatures
       tags: [system]
       summary: Get enabled feature flags
-      description: Returns a dictionary of feature flag names to their enabled state.
+      description: Returns a dictionary of feature flag names to their enabled state. Cloud deployments may include additional typed fields alongside the boolean flags.
       responses:
         "200":
           description: Feature flags
@@ -641,6 +641,43 @@ paths:
                 type: object
                 additionalProperties:
                   type: boolean
+                properties:
+                  max_upload_size:
+                    type: integer
+                    format: int64
+                    minimum: 0
+                    description: "Maximum file upload size in bytes."
+                  free_tier_credits:
+                    type: integer
+                    format: int32
+                    minimum: 0
+                    nullable: true
+                    x-runtime: [cloud]
+                    description: "[cloud-only] Credits available to free-tier users. Local ComfyUI returns null."
+                  posthog_api_host:
+                    type: string
+                    format: uri
+                    nullable: true
+                    x-runtime: [cloud]
+                    description: "[cloud-only] PostHog analytics proxy URL for frontend telemetry. Local ComfyUI returns null."
+                  max_concurrent_jobs:
+                    type: integer
+                    format: int32
+                    minimum: 0
+                    nullable: true
+                    x-runtime: [cloud]
+                    description: "[cloud-only] Maximum concurrent jobs the authenticated user can run. Local ComfyUI returns null."
+                  workflow_templates_version:
+                    type: string
+                    nullable: true
+                    x-runtime: [cloud]
+                    description: "[cloud-only] Version identifier for the workflow templates bundle. Local ComfyUI returns null."
+                  workflow_templates_source:
+                    type: string
+                    nullable: true
+                    enum: [dynamic_config_override, workflow_templates_version_json]
+                    x-runtime: [cloud]
+                    description: "[cloud-only] How the templates version was resolved. Local ComfyUI returns null."
 
   # ---------------------------------------------------------------------------
   # Node / Object Info
@@ -1497,6 +1534,24 @@ paths:
             type: string
             enum: [asc, desc]
           description: Sort direction
+        - name: job_ids
+          in: query
+          schema:
+            type: string
+          x-runtime: [cloud]
+          description: "[cloud-only] Comma-separated UUIDs to filter assets by associated job."
+        - name: include_public
+          in: query
+          schema:
+            type: boolean
+          x-runtime: [cloud]
+          description: "[cloud-only] Include workspace-public assets in addition to the caller's own."
+        - name: asset_hash
+          in: query
+          schema:
+            type: string
+          x-runtime: [cloud]
+          description: "[cloud-only] Filter by exact content hash."
       responses:
         "200":
           description: Asset list
@@ -1542,6 +1597,49 @@ paths:
                   type: string
                   format: uuid
                   description: ID of an existing asset to use as the preview image
+                id:
+                  type: string
+                  format: uuid
+                  nullable: true
+                  x-runtime: [cloud]
+                  description: "[cloud-only] Client-supplied asset ID for idempotent creation. If an asset with this ID already exists, the existing asset is returned."
+          application/json:
+            schema:
+              type: object
+              x-runtime: [cloud]
+              description: "[cloud-only] URL-based asset upload. Caller supplies a URL instead of a file body; the server fetches the content."
+              required:
+                - url
+              properties:
+                url:
+                  type: string
+                  format: uri
+                  description: "[cloud-only] URL of the file to import as an asset"
+                name:
+                  type: string
+                  description: Display name for the asset
+                tags:
+                  type: string
+                  description: Comma-separated tags
+                user_metadata:
+                  type: string
+                  description: JSON-encoded user metadata
+                hash:
+                  type: string
+                  description: "Blake3 hash of the file content (e.g. blake3:abc123...)"
+                mime_type:
+                  type: string
+                  description: MIME type of the file (overrides auto-detected type)
+                preview_id:
+                  type: string
+                  format: uuid
+                  description: ID of an existing asset to use as the preview image
+                id:
+                  type: string
+                  format: uuid
+                  nullable: true
+                  x-runtime: [cloud]
+                  description: "[cloud-only] Client-supplied asset ID for idempotent creation. If an asset with this ID already exists, the existing asset is returned."
       responses:
         "201":
           description: Asset created
@@ -1580,6 +1678,11 @@ paths:
                 user_metadata:
                   type: object
                   additionalProperties: true
+                mime_type:
+                  type: string
+                  nullable: true
+                  x-runtime: [cloud]
+                  description: "[cloud-only] MIME type of the content, so the type is preserved without re-inspecting content. Ignored by local ComfyUI."
       responses:
         "201":
           description: Asset created from hash
@@ -1644,6 +1747,11 @@ paths:
                   type: string
                   format: uuid
                   description: ID of the asset to use as the preview
+                mime_type:
+                  type: string
+                  nullable: true
+                  x-runtime: [cloud]
+                  description: "[cloud-only] MIME type override when auto-detection was wrong. Ignored by local ComfyUI."
       responses:
         "200":
           description: Asset updated
@@ -2004,21 +2112,13 @@ components:
           format: uuid
           nullable: true
           x-runtime: [cloud]
-          description: |
-            UUID identifying a hosted-cloud workflow entity to associate with this
-            job. Local ComfyUI doesn't track workflow entities and returns `null`
-            (or omits the field). The `x-runtime: [cloud]` extension marks this
-            as populated only by the hosted-cloud runtime; absence of the tag
-            means a field is populated by all runtimes.
+          description: "[cloud-only] Cloud workflow entity ID for tracking and gallery association. Ignored by local ComfyUI."
         workflow_version_id:
           type: string
           format: uuid
           nullable: true
           x-runtime: [cloud]
-          description: |
-            UUID identifying a hosted-cloud workflow version to associate with
-            this job. Local ComfyUI returns `null` (or omits the field). See
-            `workflow_id` above for `x-runtime` semantics.
+          description: "[cloud-only] Cloud workflow version ID for pinning execution to a specific version. Ignored by local ComfyUI."
 
     PromptResponse:
       type: object

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.42.15
-comfyui-workflow-templates==0.9.68
+comfyui-workflow-templates==0.9.69
 comfyui-embedded-docs==0.4.4
 torch
 torchsde

server.py

@@ -560,7 +560,7 @@ class PromptServer():
                             buffer.seek(0)
 
                             return web.Response(body=buffer.read(), content_type=f'image/{image_format}',
-                                                headers={"Content-Disposition": f"filename=\"{filename}\""})
+                                                headers={"Content-Disposition": f"attachment; filename=\"{filename}\""})
 
                     if 'channel' not in request.rel_url.query:
                         channel = 'rgba'
@@ -580,7 +580,7 @@ class PromptServer():
                             buffer.seek(0)
 
                             return web.Response(body=buffer.read(), content_type='image/png',
-                                                headers={"Content-Disposition": f"filename=\"{filename}\""})
+                                                headers={"Content-Disposition": f"attachment; filename=\"{filename}\""})
 
                     elif channel == 'a':
                         with Image.open(file) as img:
@@ -597,7 +597,7 @@ class PromptServer():
                             alpha_buffer.seek(0)
 
                             return web.Response(body=alpha_buffer.read(), content_type='image/png',
-                                                headers={"Content-Disposition": f"filename=\"{filename}\""})
+                                                headers={"Content-Disposition": f"attachment; filename=\"{filename}\""})
                     else:
                         # Use the content type from asset resolution if available,
                         # otherwise guess from the filename.
@@ -614,7 +614,7 @@ class PromptServer():
                         return web.FileResponse(
                             file,
                             headers={
-                                "Content-Disposition": f"filename=\"{filename}\"",
+                                "Content-Disposition": f"attachment; filename=\"{filename}\"",
                                 "Content-Type": content_type
                             }
                         )

tests-unit/comfy_api_test/multicombo_serialization_test.py

@@ -0,0 +1,78 @@
+from comfy_api.latest._io import Combo, MultiCombo
+
+
+def test_multicombo_serializes_multi_select_as_object():
+    multi_combo = MultiCombo.Input(
+        id="providers",
+        options=["a", "b", "c"],
+        default=["a"],
+    )
+
+    serialized = multi_combo.as_dict()
+
+    assert serialized["multiselect"] is True
+    assert "multi_select" in serialized
+    assert serialized["multi_select"] == {}
+
+
+def test_multicombo_serializes_multi_select_with_placeholder_and_chip():
+    multi_combo = MultiCombo.Input(
+        id="providers",
+        options=["a", "b", "c"],
+        default=["a"],
+        placeholder="Select providers",
+        chip=True,
+    )
+
+    serialized = multi_combo.as_dict()
+
+    assert serialized["multiselect"] is True
+    assert serialized["multi_select"] == {
+        "placeholder": "Select providers",
+        "chip": True,
+    }
+
+
+def test_combo_does_not_serialize_multiselect():
+    """Regular Combo should not have multiselect in its serialized output."""
+    combo = Combo.Input(
+        id="choice",
+        options=["a", "b", "c"],
+    )
+
+    serialized = combo.as_dict()
+
+    # Combo sets multiselect=False, but prune_dict keeps False (not None),
+    # so it should be present but False
+    assert serialized.get("multiselect") is False
+    assert "multi_select" not in serialized
+
+
+def _validate_combo_values(val, combo_options, is_multiselect):
+    """Reproduce the validation logic from execution.py for testing."""
+    if is_multiselect and isinstance(val, list):
+        return [v for v in val if v not in combo_options]
+    else:
+        return [val] if val not in combo_options else []
+
+
+def test_multicombo_validation_accepts_valid_list():
+    options = ["a", "b", "c"]
+    assert _validate_combo_values(["a", "b"], options, True) == []
+
+
+def test_multicombo_validation_rejects_invalid_values():
+    options = ["a", "b", "c"]
+    assert _validate_combo_values(["a", "x"], options, True) == ["x"]
+
+
+def test_multicombo_validation_accepts_empty_list():
+    options = ["a", "b", "c"]
+    assert _validate_combo_values([], options, True) == []
+
+
+def test_combo_validation_rejects_list_even_with_valid_items():
+    """A regular Combo should not accept a list value."""
+    options = ["a", "b", "c"]
+    invalid = _validate_combo_values(["a", "b"], options, False)
+    assert len(invalid) > 0