ComfyUI Node
CS ProPainterX Inpaint
Erase masked regions from a video sequence with ProPainterX and paste the result back onto the source frames. Inference runs at the source size, one pass covers as many frames as the free VRAM affords, and long clips are run in overlapping passes that are cross-faded together.
CS ProPainterX Inpaint
- image
- mask
- IMAGE
◄deviceauto►
◄subvideo_length80►
◄neighbor_length10►
◄ref_stride10►
◄mask_dilation4►
◄force_unload_modelfalse►
◄batch_overlap4►
Category😺dzNodes/CineStyle
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Video frames as a ComfyUI IMAGE batch. | |
| mask | MASK | Matched per-frame masks marking what to remove; one shared mask broadcasts to every frame. | |
| device | COMBO | auto | 2 options: auto, cpu |
| subvideo_length | INT | 808–300 | Frames the transformer stage reasons over at once. This is what sets peak VRAM, and the node lowers it automatically when memory is tight. |
| neighbor_length | INT | 102–100 | — |
| ref_stride | INT | 101–100 | — |
| mask_dilation | INT | 40–64 | ProPainterX's own dilation of the removal region. |
| force_unload_model | BOOLEAN | false | Release the models when this run ends so other nodes can take the VRAM back. |
| batch_overlap | INT | 40–64 | Frames two consecutive passes share; they are cross-faded to hide the seam. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |