ComfyUI Node
🐧 WanSCAILToVideo (Latent Masked)
WanSCAILToVideo with optional original_frames/original_frame_masks inputs. White mask areas are hard-preserved from the original video in the latent instead of trusting the model to recreate them. The mask is snapped to the transformer token grid, with frame_mask_grow to shift the boundary.
🐧 WanSCAILToVideo (Latent Masked)
- positive
- negative
- vae
- pose_video
- pose_video_mask
- reference_image
- reference_image_mask
- clip_vision_output
- previous_frames
- original_frames
- original_frame_masks
- positive
- negative
- latent
- video_frame_offset
◄width512►
◄height896►
◄length81►
◄batch_size1►
◄pose_strength1.00►
◄pose_start0.00►
◄pose_end1.00►
◄video_frame_offset0►
◄previous_frame_count5►
◄replacement_modefalse►
◄enable_latent_masktrue►
◄frame_mask_grow0►
CategorySuperNodes/Video
Inputs (23)
| Name | Type | Default | Description |
|---|---|---|---|
| positive | CONDITIONING | — | |
| negative | CONDITIONING | — | |
| vae | VAE | — | |
| width | INT | 51232–16384 | — |
| height | INT | 89632–16384 | — |
| length | INT | 811–16384 | — |
| batch_size | INT | 11–4096 | — |
| pose_strength | FLOAT | 1.000–10 | Strength of the pose latent. |
| pose_start | FLOAT | 0.000–1 | Start step of the pose conditioning. |
| pose_end | FLOAT | 1.000–1 | End step of the pose conditioning. |
| video_frame_offset | INT | 00–16384 | Cumulative output frame this chunk begins at. Wire from the previous chunk's video_frame_offset output. |
| previous_frame_count | INT | 51–16384 | Tail frames of previous_frames to anchor. SCAIL-2 trained at 5 (81-frame chunks, 76-frame step). |
| pose_videoopt | IMAGE | Video used for pose conditioning. Will be downscaled to half the resolution of the main video. | |
| pose_video_maskopt | IMAGE | SCAIL-2 only. Colored per-identity SAM3 mask video at the same resolution as pose_video. | |
| replacement_modeopt | BOOLEAN | false | SCAIL-2 only. False = Animation Mode (pose_video_mask should have black background). True = Replacement Mode (pose_video_mask should have white background). |
| reference_imageopt | IMAGE | Reference image. The first image is the primary reference (composite all identities onto it). SCAIL-2: extra batch images are used as additional views (back view, close-up, occluded background), each needing a matching reference_image_mask in that identity's color. | |
| reference_image_maskopt | IMAGE | SCAIL-2 only. Colored reference mask, batch matching reference_image (first = primary reference mask, rest = identity masks for the additional reference_image). | |
| clip_vision_outputopt | CLIP_VISION_OUTPUT | CLIP vision features for conditioning. Model is trained with stretch resize to aspect ratio. | |
| previous_framesopt | IMAGE | SCAIL-2 only. Full decoded output of the previous chunk. Only the last previous_frame_count are used as the extension anchor. | |
| enable_latent_maskopt | BOOLEAN | true | Toggle the original-frame preservation feature. Off = original_frames/original_frame_masks are ignored entirely and the node behaves exactly like stock WanSCAILToVideo. |
| original_framesopt | IMAGE | Original video frames at the output resolution. Areas covered by original_frame_masks are encoded into the latent and locked so the sampler preserves them exactly. Offset by video_frame_offset like pose_video. Ignored if original_frame_masks is not connected. | |
| original_frame_masksopt | MASK | Per-frame masks matching original_frames (a single mask is broadcast to all frames). White (1.0) = hard-preserve the original video content, black (0.0) = generate normally. Snapped outward to the 16 px token grid. Ignored if original_frames is not connected. | |
| frame_mask_growopt | INT | 0-16–16 | Shift the preserve boundary by this many latent cells (8 px each) after token-grid snapping. Positive = dilate into generated territory (blending never touches the area you masked). Negative = erode into preserved territory (the model recreates the boundary strip, sacrificing some original content for a more seamless transition). |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |
| latent | LATENT | Latent of the generation size. Contains the encoded original frames (with matching noise mask) where original_frame_masks preserves them; empty elsewhere. |
| video_frame_offset | INT | Adjusted offset + length. Wire into the next chunk. |