Inputs
| Parameter | Description | Data Type | Required | Range |
|---|---|---|---|---|
positive | The positive conditioning input. | CONDITIONING | Yes | - |
negative | The negative conditioning input. | CONDITIONING | Yes | - |
vae | The VAE model used for encoding images and video frames. | VAE | Yes | - |
width | The width of the output video in pixels (default: 512). Must be divisible by 32. | INT | Yes | 32 to MAX_RESOLUTION |
height | The height of the output video in pixels (default: 896). Must be divisible by 32. | INT | Yes | 32 to MAX_RESOLUTION |
length | The number of frames in the video (default: 81). Must be divisible by 4. | INT | Yes | 1 to MAX_RESOLUTION |
batch_size | The number of videos to generate in a batch (default: 1). | INT | Yes | 1 to 4096 |
pose_video | Video used for pose conditioning. Will be downscaled to half the resolution of the main video. | IMAGE | No | - |
pose_video_mask | SCAIL-2 only. Colored per-identity SAM3 mask video at the same resolution as pose_video. | IMAGE | No | - |
replacement_mode | SCAIL-2 only. False = Animation Mode (pose_video_mask should have black background). True = Replacement Mode (pose_video_mask should have white background). Default: False. | BOOLEAN | No | - |
pose_strength | Strength of the pose latent (default: 1.0). | FLOAT | Yes | 0.0 to 10.0 |
pose_start | Start step of the pose conditioning (default: 0.0). | FLOAT | Yes | 0.0 to 1.0 |
pose_end | End step of the pose conditioning (default: 1.0). | FLOAT | Yes | 0.0 to 1.0 |
reference_image | Reference image, for multiple references composite all on single image. | IMAGE | No | - |
reference_image_mask | SCAIL-2 only. Colored reference mask at the same resolution as reference_image. | IMAGE | No | - |
clip_vision_output | CLIP vision features for conditioning. Model is trained with stretch resize to aspect ratio. | CLIP_VISION_OUTPUT | No | - |
video_frame_offset | Cumulative output frame this chunk begins at. Wire from the previous chunk’s video_frame_offset output (default: 0). | INT | Yes | 0 to MAX_RESOLUTION |
previous_frame_count | Tail frames of previous_frames to anchor. SCAIL-2 trained at 5 (81-frame chunks, 76-frame step) (default: 5). | INT | Yes | 1 to MAX_RESOLUTION |
previous_frames | SCAIL-2 only. Full decoded output of the previous chunk. Only the last previous_frame_count are used as the extension anchor. | IMAGE | No | - |
pose_video and pose_video_mask inputs are processed only for the first length frames. The reference_image is processed only for the first image in the batch. When reference_image is provided, it is encoded into a latent and embedded into both positive and negative conditioning. When clip_vision_output is provided, it is applied to both positive and negative conditioning. The pose_video is downscaled to half the resolution of the main video before encoding. When previous_frames is provided, only the last previous_frame_count frames are used as the extension anchor, and the video_frame_offset is adjusted accordingly. In Replacement Mode (replacement_mode=True), the reference image is composited on a black background using the reference image mask as an alpha matte.
Outputs
| Output Name | Description | Data Type |
|---|---|---|
positive | The modified positive conditioning, potentially containing embedded reference image latents, CLIP vision output, pose video latents, driving masks, reference masks, or previous frame latents. | CONDITIONING |
negative | The modified negative conditioning, potentially containing embedded reference image latents, CLIP vision output, pose video latents, driving masks, reference masks, or previous frame latents. | CONDITIONING |
latent | An empty latent tensor of shape [batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8]. When previous_frames is provided, the latent is partially filled with encoded previous frames and a noise mask is included. | LATENT |
video_frame_offset | Adjusted offset + length. Wire into the next chunk for sequential video generation. | INT |
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
Source fingerprint (SHA-256):
30e14959248c46e624e2ce2e3d079cd5aad94c12b66d74d4979ef70143b871e3