WanHuMoImageToVideo - ComfyUI Built-in Node Documentation

The WanHuMoImageToVideo node prepares conditioning data and an empty latent video for the Wan HuMo video generation pipeline.

The WanHuMoImageToVideo node prepares conditioning data and an empty latent video for the Wan HuMo video generation pipeline. It can attach a reference image and audio embeddings to the positive and negative conditioning inputs, and it creates a zero-filled latent sized from the requested width, height, length, and batch size.

Inputs

Parameter	Description	Data Type	Required	Range
positive	Positive conditioning input that guides the video generation toward desired content.	CONDITIONING	Yes	-
negative	Negative conditioning input that steers the video generation away from unwanted content.	CONDITIONING	Yes	-
vae	VAE model used for encoding reference images into latent space.	VAE	Yes	-
width	Width of the output video frames in pixels. Default: 832.	INT	Yes	16 to MAX_RESOLUTION, step 16
height	Height of the output video frames in pixels. Default: 480.	INT	Yes	16 to MAX_RESOLUTION, step 16
length	Number of frames in the generated video sequence. Default: 97.	INT	Yes	1 to MAX_RESOLUTION, step 4
batch_size	Number of video sequences to generate simultaneously. Default: 1.	INT	Yes	1 to 4096
audio_encoder_output	Optional audio encoding data that can influence video generation based on audio content.	AUDIOENCODEROUTPUT	No	-
ref_image	Optional reference image used to guide the video generation style and content. Only the first image in the batch is used.	IMAGE	No	-

Note: When a reference image is provided, the first image in the batch is upscaled to the requested width and height using bilinear interpolation and encoded with the VAE. That reference latent is attached to the positive conditioning, while a zero-filled latent of the same shape is attached to the negative conditioning. When audio_encoder_output is provided, the audio embeddings are interpolated and attached to the positive conditioning, while a zero-filled audio embedding is attached to the negative conditioning. If either optional input is omitted, zero-filled placeholder tensors are used: a zero reference latent of shape [batch_size, 16, 1, height // 8, width // 8] and/or zero audio embeddings of shape [batch_size, latent_t + 1, 8, 5, 1280], where latent_t = ((length - 1) // 4) + 1.

Outputs

Output Name	Description	Data Type
positive	Modified positive conditioning with reference image and/or audio embeddings incorporated.	CONDITIONING
negative	Modified negative conditioning with reference image and/or audio embeddings incorporated.	CONDITIONING
latent	Zero-initialized latent representation for the video sequence, sized according to width, height, length, and batch_size. Shape: [batch_size, 16, latent_t, height // 8, width // 8], where latent_t = ((length - 1) // 4) + 1.	LATENT

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): db674a4a00729a8715988030083e2858f958cd21de73bbbe4ed6d76f5f539419