Avoid TorchCodec for tensor-native inference

README.md

@@ -27,6 +27,8 @@ python ComfyUI-SAM-Audio/install.py
 
 Then restart ComfyUI. If installed through ComfyUI Manager, the installer should run automatically.
 
+The node uses ComfyUI's decoded media tensors and does not install or load TorchCodec.
+
 ## Model download
 
 Select a model in **SAM-Audio Model Loader** and queue the workflow. The loader first tries Meta's official repository. If Hugging Face denies gated access, it automatically downloads the matching pinned and checksum-verified checkpoint from the [public mirror collection](https://huggingface.co/collections/mrfakename/sam-audio).

pyproject.toml

@@ -13,7 +13,6 @@ dependencies = [
     "regex",
     "safetensors",
     "timm",
-    "torchcodec",
     "torchdiffeq",
     "tqdm",
     "transformers>=4.54.0",

requirements.txt

@@ -6,7 +6,6 @@ huggingface-hub
 regex
 safetensors
 timm
-torchcodec
 torchdiffeq
 tqdm
 transformers>=4.54.0

sam_audio_comfy/runtime.py

@@ -13,6 +13,7 @@ import torch
 
 from .attention import validate_attention_backend
 from .audiotools_compat import audiotools_import_compatibility
+from .torchcodec_compat import torchcodec_import_compatibility
 from .xformers_compat import xformers_import_compatibility
 
 LOGGER = logging.getLogger(__name__)
@@ -305,14 +306,17 @@ def import_sam_audio() -> tuple[Any, Any]:
         LOGGER.info(
             "Loading SAM-Audio through isolated inference-only compatibility layers"
         )
-        with xformers_import_compatibility(), audiotools_import_compatibility():
+        with (
+            xformers_import_compatibility(),
+            audiotools_import_compatibility(),
+            torchcodec_import_compatibility(),
+        ):
             module = importlib.import_module("sam_audio")
         return module.SAMAudio, module.SAMAudioProcessor
     except Exception as error:
         raise RuntimeError(
             "SAM-Audio could not be imported. Install this node's requirements with "
-            "ComfyUI's Python interpreter, then restart ComfyUI. This node does not "
-            "require or install xFormers."
+            "ComfyUI's Python interpreter, then restart ComfyUI."
         ) from error
 
 

sam_audio_comfy/torchcodec_compat.py

@@ -0,0 +1,64 @@
+"""Import-only TorchCodec surface for tensor-native ComfyUI inference."""
+
+from __future__ import annotations
+
+import contextlib
+import sys
+import types
+from collections.abc import Iterator
+
+
+class _TensorInputsOnly:
+    def __init__(self, *_args, **_kwargs):
+        raise RuntimeError(
+            "TorchCodec file decoding is disabled in ComfyUI-SAM-Audio. Pass decoded "
+            "ComfyUI AUDIO, IMAGE, and MASK tensors to the node instead."
+        )
+
+
+class AudioDecoder(_TensorInputsOnly):
+    """Placeholder for an upstream file-I/O path unused by the ComfyUI node."""
+
+
+class VideoDecoder(_TensorInputsOnly):
+    """Placeholder for an upstream file-I/O path unused by the ComfyUI node."""
+
+
+class AudioEncoder(_TensorInputsOnly):
+    """Placeholder for an optional upstream ranker disabled by this integration."""
+
+
+def _compatibility_modules() -> dict[str, types.ModuleType]:
+    torchcodec = types.ModuleType("torchcodec")
+    torchcodec.__path__ = []
+    decoders = types.ModuleType("torchcodec.decoders")
+    encoders = types.ModuleType("torchcodec.encoders")
+    decoders.AudioDecoder = AudioDecoder
+    decoders.VideoDecoder = VideoDecoder
+    encoders.AudioEncoder = AudioEncoder
+    torchcodec.decoders = decoders
+    torchcodec.encoders = encoders
+    return {
+        "torchcodec": torchcodec,
+        "torchcodec.decoders": decoders,
+        "torchcodec.encoders": encoders,
+    }
+
+
+@contextlib.contextmanager
+def torchcodec_import_compatibility() -> Iterator[None]:
+    """Temporarily provide upstream's unused decoder imports without loading FFmpeg."""
+    compatibility_modules = _compatibility_modules()
+    missing = object()
+    previous = {
+        name: sys.modules.get(name, missing) for name in compatibility_modules
+    }
+    sys.modules.update(compatibility_modules)
+    try:
+        yield
+    finally:
+        for name, module in previous.items():
+            if module is missing:
+                sys.modules.pop(name, None)
+            else:
+                sys.modules[name] = module

tests/test_package.py

@@ -1,9 +1,15 @@
 import sys
+import types
 from pathlib import Path
 
+import pytest
 import torch
 
-from sam_audio_comfy import audiotools_compat, xformers_compat
+from sam_audio_comfy import (
+    audiotools_compat,
+    torchcodec_compat,
+    xformers_compat,
+)
 
 
 def test_node_does_not_require_xformers():
@@ -21,6 +27,17 @@ def test_node_does_not_require_xformers():
         assert not any("xformers" in line for line in dependency_lines)
 
 
+def test_node_does_not_require_torchcodec():
+    root = Path(__file__).parents[1]
+    for filename in ("requirements.txt", "pyproject.toml"):
+        dependency_lines = [
+            line
+            for line in (root / filename).read_text().lower().splitlines()
+            if line.strip() and not line.lstrip().startswith("#")
+        ]
+        assert not any("torchcodec" in line for line in dependency_lines)
+
+
 def test_upstream_installer_uses_no_deps():
     root = Path(__file__).parents[1]
     installer = (root / "install.py").read_text().lower()
@@ -58,3 +75,33 @@ def test_audiotools_compatibility_is_import_only_and_scoped(monkeypatch):
         assert audiotools.ml.BaseModel.INTERN == []
 
     assert "audiotools" not in sys.modules
+
+
+def test_torchcodec_compatibility_is_import_only_and_scoped(monkeypatch):
+    monkeypatch.delitem(sys.modules, "torchcodec", raising=False)
+    monkeypatch.delitem(sys.modules, "torchcodec.decoders", raising=False)
+    monkeypatch.delitem(sys.modules, "torchcodec.encoders", raising=False)
+
+    with torchcodec_compat.torchcodec_import_compatibility():
+        decoders = sys.modules["torchcodec.decoders"]
+        assert decoders.AudioDecoder is torchcodec_compat.AudioDecoder
+        assert decoders.VideoDecoder is torchcodec_compat.VideoDecoder
+        with pytest.raises(RuntimeError, match="decoded ComfyUI"):
+            decoders.AudioDecoder("input.wav")
+
+    assert "torchcodec" not in sys.modules
+    assert "torchcodec.decoders" not in sys.modules
+
+
+def test_existing_torchcodec_modules_are_restored(monkeypatch):
+    original = types.ModuleType("torchcodec")
+    original_decoders = types.ModuleType("torchcodec.decoders")
+    monkeypatch.setitem(sys.modules, "torchcodec", original)
+    monkeypatch.setitem(sys.modules, "torchcodec.decoders", original_decoders)
+
+    with torchcodec_compat.torchcodec_import_compatibility():
+        assert sys.modules["torchcodec"] is not original
+        assert sys.modules["torchcodec.decoders"] is not original_decoders
+
+    assert sys.modules["torchcodec"] is original
+    assert sys.modules["torchcodec.decoders"] is original_decoders