- Replace SeedVR2Resize / SeedVR2ResizeAdvanced with a single pad-only node
SeedVR2Preprocess ("Pre-Process SeedVR2 Input"); resizing is delegated to the
native Resize Image/Mask node. Input resized_images -> output processed_images.
- SeedVR2PostProcessing consumes the native-resized image directly: inputs images
+ original_resized_images; de-pad the decode to the reference's unpadded
dimensions; color-correct against the reference pixels; apply alpha with the same
top-left de-pad crop as RGB. Output display_name "images".
- SeedVR2Conditioning latent input display "latent"; SeedVR2ProgressiveSampler
latent input/output latent_image -> latent (execute fn + _run_standard_sample).
- Update the SeedVR2 unit tests for the renamed API; remove dead area_resize.
- Strip third-party-node references from comments, a docstring, and a user-facing
error message.
John Pollock
1e08e8b724