SeedVR2Preprocess - ComfyUI Built-in Node Documentation

This node prepares a resized image or video for the SeedVR2 model by padding it to the shape the model expects.

This node prepares a resized image or video for the SeedVR2 model by padding it to the shape the model expects. It drops the alpha channel during processing; the companion Post-Process SeedVR2 Output node restores it later from the original resized image. Pixel values are clamped to the 0-1 range, the height and width are padded to multiples of 16, and the frame count is padded by repeating the last frame when needed.

Inputs

Parameter	Description	Data Type	Required	Range
resized_images	The resized image to process.	IMAGE	Yes	-

Note: The input can be a single frame, a sequence of frames, or a batch of videos. If it has more than 3 channels, the alpha channel is dropped and only RGB is kept. The shorter edge of the input must be at least 2 pixels. Spatial padding is filled with black (value 0), and valid frame counts follow a 4n+1 pattern (1, 5, 9, 13, …).

Outputs

Output Name	Description	Data Type
images	The padded image for VAE encoding.	IMAGE

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): f4fa433d299feba40696f27ff365c59988e5102112f09536724b5db5b09416bb