Don't instantiate nodes during validation

Addresses review feedback: the V1 executability check fell back to
constructing the node (class_def()) when the FUNCTION method wasn't found on
the class. That runs __init__ during validation, so a constructor's side
effects or failure could be misreported as invalid_node_definition for an
otherwise valid node.

Inspect only the class. No core/extra node defines its FUNCTION method on the
instance, so this loses no real coverage while removing the side-effect risk.

Replace the instance-fallback test with one asserting a node with a raising
__init__ but a valid class-level method still passes validation (i.e. it is
never instantiated).

comfy/ops.py

@@ -1089,6 +1089,19 @@ def _load_quantized_module(module, super_load, state_dict, prefix, local_metadat
             if ts is None or bs is None:
                 raise ValueError(f"Missing NVFP4 scales for layer {layer_name}")
             scales = {"scale": ts, "block_scale": bs}
+        elif module.quant_format == "int8_tensorwise":
+            scale = pop_scale("weight_scale")
+            if scale is None:
+                raise ValueError(f"Missing INT8 weight scale for layer {layer_name}")
+            scales = {"scale": scale}
+            params_conf = layer_conf.get("params", {})
+            if not isinstance(params_conf, dict):
+                params_conf = {}
+            if layer_conf.get("convrot", params_conf.get("convrot", False)):
+                scales["convrot"] = True
+                scales["convrot_groupsize"] = int(
+                    layer_conf.get("convrot_groupsize", params_conf.get("convrot_groupsize", 256))
+                )
         else:
             raise ValueError(f"Unsupported quantization format: {module.quant_format}")
 
@@ -1131,6 +1144,10 @@ def _quantized_weight_state_dict(module, sd, prefix, extra_quant_conf=None, extr
         quant_conf = {"format": module.quant_format}
         if getattr(module, '_full_precision_mm_config', False):
             quant_conf["full_precision_matrix_mult"] = True
+        params = getattr(module.weight, "_params", None)
+        if module.quant_format == "int8_tensorwise" and getattr(params, "convrot", False):
+            quant_conf["convrot"] = True
+            quant_conf["convrot_groupsize"] = getattr(params, "convrot_groupsize", 256)
         if extra_quant_conf:
             quant_conf.update(extra_quant_conf)
         sd[f"{prefix}comfy_quant"] = torch.tensor(list(json.dumps(quant_conf).encode("utf-8")), dtype=torch.uint8)
@@ -1183,8 +1200,33 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
             def _forward(self, input, weight, bias):
                 return torch.nn.functional.linear(input, weight, bias)
 
-            def forward_comfy_cast_weights(self, input, compute_dtype=None, want_requant=False):
-                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype, want_requant=want_requant)
+            def forward_comfy_cast_weights(
+                self,
+                input,
+                compute_dtype=None,
+                want_requant=False,
+                weight_only_quant=False,
+            ):
+                if weight_only_quant:
+                    weight, bias, offload_stream = cast_bias_weight(
+                        self,
+                        input=None,
+                        dtype=self.weight.dtype,
+                        device=input.device,
+                        bias_dtype=input.dtype,
+                        offloadable=True,
+                        compute_dtype=compute_dtype,
+                        want_requant=want_requant,
+                    )
+                    weight = weight.to(dtype=input.dtype)
+                else:
+                    weight, bias, offload_stream = cast_bias_weight(
+                        self,
+                        input,
+                        offloadable=True,
+                        compute_dtype=compute_dtype,
+                        want_requant=want_requant,
+                    )
                 x = self._forward(input, weight, bias)
                 uncast_bias_weight(self, weight, bias, offload_stream)
                 return x
@@ -1203,9 +1245,10 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                     not getattr(self, 'comfy_force_cast_weights', False) and
                     len(self.weight_function) == 0 and len(self.bias_function) == 0
                 )
+                quantize_input = QUANT_ALGOS.get(getattr(self, 'quant_format', None), {}).get("quantize_input", True)
 
                 # Training path: quantized forward with compute_dtype backward via autograd function
-                if (input.requires_grad and _use_quantized):
+                if (input.requires_grad and _use_quantized and quantize_input):
 
                     weight, bias, offload_stream = cast_bias_weight(
                         self,
@@ -1227,7 +1270,7 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                     return output
 
                 # Inference path (unchanged)
-                if _use_quantized:
+                if _use_quantized and quantize_input:
 
                     # Reshape 3D tensors to 2D for quantization (needed for NVFP4 and others)
                     input_reshaped = input.reshape(-1, input_shape[2]) if input.ndim == 3 else input
@@ -1241,7 +1284,13 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                             scale = comfy.model_management.cast_to_device(scale, input.device, None)
                         input = QuantizedTensor.from_float(input_reshaped, self.layout_type, scale=scale)
 
-                output = self.forward_comfy_cast_weights(input, compute_dtype, want_requant=isinstance(input, QuantizedTensor))
+                weight_only_quant = _use_quantized and not quantize_input and isinstance(self.weight, QuantizedTensor)
+                output = self.forward_comfy_cast_weights(
+                    input,
+                    compute_dtype,
+                    want_requant=isinstance(input, QuantizedTensor),
+                    weight_only_quant=weight_only_quant,
+                )
 
                 # Reshape output back to 3D if input was 3D
                 if reshaped_3d:

comfy/quant_ops.py

@@ -10,6 +10,7 @@ try:
         QuantizedLayout,
         TensorCoreFP8Layout as _CKFp8Layout,
         TensorCoreNVFP4Layout as _CKNvfp4Layout,
+        TensorWiseINT8Layout as _CKTensorWiseINT8Layout,
         register_layout_op,
         register_layout_class,
         get_layout_class,
@@ -47,6 +48,9 @@ except ImportError as e:
     class _CKNvfp4Layout:
         pass
 
+    class _CKTensorWiseINT8Layout:
+        pass
+
     def register_layout_class(name, cls):
         pass
 
@@ -174,6 +178,7 @@ class TensorCoreFP8E5M2Layout(_TensorCoreFP8LayoutBase):
 
 # Backward compatibility alias - default to E4M3
 TensorCoreFP8Layout = TensorCoreFP8E4M3Layout
+TensorWiseINT8Layout = _CKTensorWiseINT8Layout
 
 
 # ==============================================================================
@@ -184,6 +189,7 @@ register_layout_class("TensorCoreFP8Layout", TensorCoreFP8Layout)
 register_layout_class("TensorCoreFP8E4M3Layout", TensorCoreFP8E4M3Layout)
 register_layout_class("TensorCoreFP8E5M2Layout", TensorCoreFP8E5M2Layout)
 register_layout_class("TensorCoreNVFP4Layout", TensorCoreNVFP4Layout)
+register_layout_class("TensorWiseINT8Layout", _CKTensorWiseINT8Layout)
 if _CK_MXFP8_AVAILABLE:
     register_layout_class("TensorCoreMXFP8Layout", TensorCoreMXFP8Layout)
 
@@ -214,6 +220,13 @@ if _CK_MXFP8_AVAILABLE:
         "group_size": 32,
     }
 
+QUANT_ALGOS["int8_tensorwise"] = {
+    "storage_t": torch.int8,
+    "parameters": {"weight_scale"},
+    "comfy_tensor_layout": "TensorWiseINT8Layout",
+    "quantize_input": False,
+}
+
 
 # ==============================================================================
 # Re-exports for backward compatibility
@@ -226,6 +239,7 @@ __all__ = [
     "TensorCoreFP8E4M3Layout",
     "TensorCoreFP8E5M2Layout",
     "TensorCoreNVFP4Layout",
+    "TensorWiseINT8Layout",
     "QUANT_ALGOS",
     "register_layout_op",
 ]

comfy_api_nodes/apis/bytedance.py

@@ -177,6 +177,10 @@ SEEDANCE2_PRICE_PER_1K_TOKENS = {
     ("dreamina-seedance-2-0-fast-260128", True, "480p"): 0.0033,
     ("dreamina-seedance-2-0-fast-260128", False, "720p"): 0.0056,
     ("dreamina-seedance-2-0-fast-260128", True, "720p"): 0.0033,
+    ("dreamina-seedance-2-0-mini", False, "480p"): 0.0035,
+    ("dreamina-seedance-2-0-mini", True, "480p"): 0.0021,
+    ("dreamina-seedance-2-0-mini", False, "720p"): 0.0035,
+    ("dreamina-seedance-2-0-mini", True, "720p"): 0.0021,
 }
 
 
@@ -278,6 +282,10 @@ SEEDANCE2_REF_VIDEO_PIXEL_LIMITS = {
         "480p": {"min": 409_600, "max": 927_408},
         "720p": {"min": 409_600, "max": 927_408},
     },
+    "dreamina-seedance-2-0-mini": {
+        "480p": {"min": 409_600, "max": 927_408},
+        "720p": {"min": 409_600, "max": 927_408},
+    },
 }
 
 # The time in this dictionary are given for 10 seconds duration.

comfy_api_nodes/nodes_bytedance.py

@@ -89,6 +89,7 @@ BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT = "/proxy/byteplus-seedance2/api/v3/cont
 SEEDANCE_MODELS = {
     "Seedance 2.0": "dreamina-seedance-2-0-260128",
     "Seedance 2.0 Fast": "dreamina-seedance-2-0-fast-260128",
+    "Seedance 2.0 Mini": "dreamina-seedance-2-0-mini",
 }
 
 DEPRECATED_MODELS = {"seedance-1-0-lite-t2v-250428", "seedance-1-0-lite-i2v-250428"}
@@ -1623,8 +1624,10 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
                     options=[
                         IO.DynamicCombo.Option("Seedance 2.0", _seedance2_text_inputs(["480p", "720p", "1080p", "4k"])),
                         IO.DynamicCombo.Option("Seedance 2.0 Fast", _seedance2_text_inputs(["480p", "720p"])),
+                        IO.DynamicCombo.Option("Seedance 2.0 Mini", _seedance2_text_inputs(["480p", "720p"])),
                     ],
-                    tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
+                    tooltip="Seedance 2.0 for maximum quality; Fast for speed optimization; "
+                    "Mini for the fastest, lowest-cost generation.",
                 ),
                 IO.Int.Input(
                     "seed",
@@ -1666,6 +1669,7 @@ class ByteDance2TextToVideoNode(IO.ComfyNode):
                   $dur := $lookup(widgets, "model.duration");
                   $pricePer1K := $res = "4k"    ? 0.00572 :
                                  $res = "1080p" ? 0.011011 :
+                                 $contains($m, "mini") ? 0.005005 :
                                  $contains($m, "fast") ? 0.008008 : 0.01001;
                   $rate := $res = "4k"    ? $rate4k :
                            $res = "1080p" ? $rate1080 :
@@ -1734,8 +1738,13 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
                             "Seedance 2.0 Fast",
                             _seedance2_text_inputs(["480p", "720p"], default_ratio="adaptive"),
                         ),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0 Mini",
+                            _seedance2_text_inputs(["480p", "720p"], default_ratio="adaptive"),
+                        ),
                     ],
-                    tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
+                    tooltip="Seedance 2.0 for maximum quality; Fast for speed optimization; "
+                    "Mini for the fastest, lowest-cost generation.",
                 ),
                 IO.Image.Input(
                     "first_frame",
@@ -1801,6 +1810,7 @@ class ByteDance2FirstLastFrameNode(IO.ComfyNode):
                   $dur := $lookup(widgets, "model.duration");
                   $pricePer1K := $res = "4k"    ? 0.00572 :
                                  $res = "1080p" ? 0.011011 :
+                                 $contains($m, "mini") ? 0.005005 :
                                  $contains($m, "fast") ? 0.008008 : 0.01001;
                   $rate := $res = "4k"    ? $rate4k :
                            $res = "1080p" ? $rate1080 :
@@ -2024,8 +2034,13 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
                             "Seedance 2.0 Fast",
                             _seedance2_reference_inputs(["480p", "720p"], default_ratio="adaptive"),
                         ),
+                        IO.DynamicCombo.Option(
+                            "Seedance 2.0 Mini",
+                            _seedance2_reference_inputs(["480p", "720p"], default_ratio="adaptive"),
+                        ),
                     ],
-                    tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.",
+                    tooltip="Seedance 2.0 for maximum quality; Fast for speed optimization; "
+                    "Mini for the fastest, lowest-cost generation.",
                 ),
                 IO.Int.Input(
                     "seed",
@@ -2071,9 +2086,11 @@ class ByteDance2ReferenceNode(IO.ComfyNode):
                   $dur := $lookup(widgets, "model.duration");
                   $noVideoPricePer1K := $res = "4k"    ? 0.00572 :
                                         $res = "1080p" ? 0.011011 :
+                                        $contains($m, "mini") ? 0.005005 :
                                         $contains($m, "fast") ? 0.008008 : 0.01001;
                   $videoPricePer1K := $res = "4k"    ? 0.003432 :
                                       $res = "1080p" ? 0.006721 :
+                                      $contains($m, "mini") ? 0.003003 :
                                       $contains($m, "fast") ? 0.004719 : 0.006149;
                   $rate := $res = "4k"    ? $rate4k :
                            $res = "1080p" ? $rate1080 :

execution.py

@@ -1113,6 +1113,32 @@ def full_type_name(klass):
         return klass.__qualname__
     return module + '.' + klass.__qualname__
 
+def node_not_executable_reason(class_def, class_type):
+    """Return a human-readable reason the node cannot be executed, or None if it's fine.
+
+    Catches a node whose declared entry point doesn't resolve to a real method
+    (e.g. a V1 ``FUNCTION = "invert"`` where the method is misspelled, or a V3 node
+    missing its ``execute`` override). Running this during validation surfaces the
+    problem before execution starts, instead of after upstream nodes have run.
+
+    Only the class is inspected; the node is never instantiated here, so a node's
+    ``__init__`` side effects cannot run (or fail) during validation.
+    """
+    try:
+        if issubclass(class_def, _ComfyNodeInternal):
+            # V3: validates that execute()/define_schema() overrides exist.
+            class_def.VALIDATE_CLASS()
+            return None
+        # V1: FUNCTION names the method to call; it must exist on the class.
+        function_name = getattr(class_def, "FUNCTION", None)
+        if function_name is None:
+            return f"'{class_type}' does not define FUNCTION"
+        if not callable(getattr(class_def, function_name, None)):
+            return f"'{class_type}' has no method '{function_name}' (declared in FUNCTION)"
+        return None
+    except Exception as ex:
+        return str(ex)
+
 async def validate_prompt(prompt_id, prompt, partial_execution_list: Union[list[str], None]):
     outputs = set()
     for x in prompt:
@@ -1148,6 +1174,35 @@ async def validate_prompt(prompt_id, prompt, partial_execution_list: Union[list[
             }
             return (False, error, [], {})
 
+        # Make sure the node is actually executable (its FUNCTION/execute entry
+        # point resolves to a real method) before we touch any schema-derived
+        # attributes below or start execution. Catches code typos up front and
+        # attributes the error to the offending node.
+        not_executable = node_not_executable_reason(class_, class_type)
+        if not_executable is not None:
+            node_title = prompt[x].get('_meta', {}).get('title', class_type)
+            error = {
+                "type": "invalid_node_definition",
+                "message": "Node is not executable",
+                "details": f"{not_executable} (Node ID '#{x}')",
+                "extra_info": {
+                    "node_id": x,
+                    "class_type": class_type,
+                    "node_title": node_title,
+                }
+            }
+            node_errors = {x: {
+                "errors": [{
+                    "type": "invalid_node_definition",
+                    "message": "Node is not executable",
+                    "details": not_executable,
+                    "extra_info": {},
+                }],
+                "dependent_outputs": [],
+                "class_type": class_type,
+            }}
+            return (False, error, [], node_errors)
+
         if hasattr(class_, 'OUTPUT_NODE') and class_.OUTPUT_NODE is True:
             if partial_execution_list is None or x in partial_execution_list:
                 outputs.add(x)

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.45.19
-comfyui-workflow-templates==0.10.2
+comfyui-workflow-templates==0.10.7
 comfyui-embedded-docs==0.5.5
 torch
 torchsde
@@ -22,7 +22,7 @@ alembic
 SQLAlchemy>=2.0.0
 filelock
 av>=16.0.0
-comfy-kitchen==0.2.10
+comfy-kitchen==0.2.12
 comfy-aimdo==0.4.10
 requests
 simpleeval>=1.0.0

tests-unit/comfy_quant/test_mixed_precision.py

@@ -228,6 +228,62 @@ class TestMixedPrecisionOps(unittest.TestCase):
         with self.assertRaises(KeyError):
             model.load_state_dict(state_dict, strict=False)
 
+    def test_int8_convrot_metadata_loads_into_params(self):
+        """ConvRot metadata must reach TensorWiseINT8Layout params."""
+        torch.manual_seed(123)
+        layer_quant_config = {
+            "layer": {
+                "format": "int8_tensorwise",
+                "convrot": True,
+                "convrot_groupsize": 256,
+            }
+        }
+        weight = torch.randn(16, 256, dtype=torch.bfloat16)
+        bias = torch.randn(16, dtype=torch.bfloat16)
+        q_weight = QuantizedTensor.from_float(
+            weight,
+            "TensorWiseINT8Layout",
+            per_channel=True,
+            convrot=True,
+            convrot_groupsize=256,
+        )
+        state_dict = {
+            "layer.weight": q_weight._qdata,
+            "layer.bias": bias,
+            "layer.weight_scale": q_weight._params.scale,
+        }
+
+        state_dict, _ = comfy.utils.convert_old_quants(
+            state_dict,
+            metadata={"_quantization_metadata": json.dumps({"layers": layer_quant_config})},
+        )
+        model = torch.nn.Module()
+        model.layer = ops.mixed_precision_ops({}).Linear(256, 16, device="cpu", dtype=torch.bfloat16)
+        model.load_state_dict(state_dict, strict=False)
+
+        self.assertIsInstance(model.layer.weight, QuantizedTensor)
+        self.assertEqual(model.layer.weight._layout_cls, "TensorWiseINT8Layout")
+        self.assertTrue(model.layer.weight._params.convrot)
+        self.assertEqual(model.layer.weight._params.convrot_groupsize, 256)
+
+        input_tensor = torch.randn(4, 256, dtype=torch.bfloat16)
+        loaded_out = model.layer(input_tensor)
+        ref_out = torch.nn.functional.linear(input_tensor, q_weight, bias)
+        self.assertTrue(torch.equal(loaded_out, ref_out))
+
+        fp16_input = input_tensor.to(torch.float16)
+        loaded_fp16_out = model.layer(fp16_input)
+        ref_fp16_out = torch.nn.functional.linear(
+            fp16_input,
+            q_weight.to(dtype=torch.float16),
+            bias.to(dtype=torch.float16),
+        )
+        self.assertTrue(torch.equal(loaded_fp16_out, ref_fp16_out))
+
+        saved = model.state_dict()
+        saved_conf = json.loads(saved["layer.comfy_quant"].numpy().tobytes())
+        self.assertTrue(saved_conf["convrot"])
+        self.assertEqual(saved_conf["convrot_groupsize"], 256)
+
 if __name__ == "__main__":
     unittest.main()
-

tests-unit/execution_test/validate_node_executable_test.py

@@ -0,0 +1,137 @@
+"""Tests for pre-execution validation that a node is actually executable.
+
+validate_prompt rejects a node whose declared entry point does not resolve to a
+real method (a V1 FUNCTION typo, or a V3 node missing its execute override) before
+any node runs, attributing the error to the offending node.
+"""
+import asyncio
+
+import nodes
+from comfy_api.latest import io
+from execution import node_not_executable_reason, validate_prompt
+
+
+class _GoodV1Node:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {}}
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "run"
+    OUTPUT_NODE = True
+    CATEGORY = "Test"
+
+    def run(self):
+        return (None,)
+
+
+class _TypoV1Node:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {}}
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "invert"  # method below is misspelled
+    OUTPUT_NODE = True
+    CATEGORY = "Test"
+
+    def invvert(self):
+        return (None,)
+
+
+class _SideEffectInitV1Node:
+    """Valid class-level method, but a constructor that must never run in validation."""
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {}}
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "run"
+    OUTPUT_NODE = True
+    CATEGORY = "Test"
+
+    def __init__(self):
+        raise RuntimeError("__init__ must not run during validation")
+
+    def run(self):
+        return (None,)
+
+
+def _v3_schema(node_id):
+    return io.Schema(
+        node_id=node_id,
+        display_name=node_id,
+        category="Test",
+        inputs=[],
+        outputs=[io.Image.Output()],
+        is_output_node=True,
+    )
+
+
+class _GoodV3Node(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return _v3_schema("GoodV3Node")
+
+    @classmethod
+    def execute(cls):
+        return io.NodeOutput(None)
+
+
+class _TypoV3Node(io.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return _v3_schema("TypoV3Node")
+
+    @classmethod
+    def exicute(cls):  # typo: should be "execute"
+        return io.NodeOutput(None)
+
+
+def _register(class_type, class_def):
+    nodes.NODE_CLASS_MAPPINGS[class_type] = class_def
+
+
+def _validate(class_type):
+    prompt = {"1": {"class_type": class_type, "inputs": {}}}
+    return asyncio.run(validate_prompt("pid", prompt, None))
+
+
+def test_good_node_passes():
+    _register("GoodV1Node", _GoodV1Node)
+    assert node_not_executable_reason(_GoodV1Node, "GoodV1Node") is None
+    valid, _, _, _ = _validate("GoodV1Node")
+    assert valid is True
+
+
+def test_typo_node_rejected_with_node_error():
+    _register("TypoV1Node", _TypoV1Node)
+    valid, error, _, node_errors = _validate("TypoV1Node")
+    assert valid is False
+    assert error["type"] == "invalid_node_definition"
+    assert node_errors["1"]["class_type"] == "TypoV1Node"
+    assert node_errors["1"]["errors"][0]["type"] == "invalid_node_definition"
+    assert "invert" in node_errors["1"]["errors"][0]["details"]
+
+
+def test_validation_does_not_instantiate_node():
+    """A valid node is not constructed during validation, so __init__ never runs."""
+    _register("SideEffectInitV1Node", _SideEffectInitV1Node)
+    assert node_not_executable_reason(_SideEffectInitV1Node, "SideEffectInitV1Node") is None
+    valid, _, _, _ = _validate("SideEffectInitV1Node")
+    assert valid is True
+
+
+def test_good_v3_node_passes():
+    _register("GoodV3Node", _GoodV3Node)
+    assert node_not_executable_reason(_GoodV3Node, "GoodV3Node") is None
+    valid, _, _, _ = _validate("GoodV3Node")
+    assert valid is True
+
+
+def test_typo_v3_node_rejected_with_node_error():
+    _register("TypoV3Node", _TypoV3Node)
+    valid, error, _, node_errors = _validate("TypoV3Node")
+    assert valid is False
+    assert error["type"] == "invalid_node_definition"
+    assert node_errors["1"]["errors"][0]["type"] == "invalid_node_definition"