Sammy Kumar
692039e01b
docs: remove edge_threshold parameter from example workflow to align with new automatic edge thresholding feature docs(README): simplify and condense VACE Annotator Scribble documentation for clarity and ease of use docs: add help-page style documentation for VACEScribbleAnnotator node to improve user guidance and usability feat(annotator_models): add ContourInference and VendorResidualBlock classes for enhanced model architecture feat(annotator_models): implement postprocess_vendor_scribble_output function for vendor-specific output processing fix(annotator_models): update load_scribble_model to handle vendor state dicts and improve model loading logic refactor(preprocess_scribble_input): add normalize_mode parameter for flexible input normalization options fix(_quantile_threshold): improve scale handling for user-defined thresholds in quantile calculations refactor(vace_scribble_annotator.py): remove edge_threshold parameter and use a default constant for edge detection to simplify the code feat(vace_scribble_annotator.py): add support for vendor-specific postprocessing of scribble output to enhance functionality
1.8 KiB
1.8 KiB
Vendoring Notes
This repository vendors a minimal subset of the VACE-Annotators project (Apache-2.0) so ComfyUI users can run the scribble annotator without installing the full upstream repo.
Vendored Components
| File | Description | Upstream Source |
|---|---|---|
nodes/vace_annotators/annotator_models/scribble.py |
ResNet generator architecture plus preprocessing/postprocessing helpers. As of this update it also vendors the official ContourInference graph to mirror upstream behavior. |
Derived from the scribble generator in ali-vilab/VACE-Annotators (Apache-2.0). |
nodes/vace_annotators/annotator_models/__init__.py |
Shim exporting loader + helpers. | Local glue code referencing the vendored generator. |
nodes/vace_annotators/scribble_loader.py |
Loader, device handling, caching, and error plumbing. | Local adaptation based on SwissArmyKnife requirements. |
Updating the Vendored Code
- Fetch the latest
scribblegenerator implementation from the upstream repository. - Preserve the Apache-2.0 license header and attribution when copying/updating files.
- Keep imports self-contained (no upstream-relative imports remain after vendoring).
- Run
tests/annotators/test_scribble_inference.pyto ensure fallback + optional model inference still work. - Document changes in
docs/nodes/vace-annotators/scribble/README.mdwhen the vendored code diverges from upstream behavior.
Additional Assets
tests/data/sample_scribble_input.pngis a synthetic 128×128 RGB gradient used by the smoke tests to create deterministic scribble outputs.- Test artifacts are written to
tests/output/for visual inspection.
The upstream project is licensed under Apache-2.0. See the repository license for full terms.