MiniMax H3 Image Conditioning - ComfyUI Built-in Node Documentation

Turns a prompt into conditioning and an empty latent for a still image with MiniMax H3 Image. Without images it is text to image; with images connected it edits them, and the prompt refers to them as <Picture 1>, <Picture 2>, and so on.

Inputs

Parameter	Description	Data Type	Required	Range
`clip`	The MiniMax H3 Image CLIP model, which tokenizes the prompt with any reference images and encodes the conditioning.	CLIP	Yes	-
`vae`	The MiniMax H3 Image VAE. With it the pictures are also anchored as reference latents; without it they only condition the text encoder.	VAE	No	-
`prompt`	What to generate, or how to edit the reference pictures.	STRING	Yes	Multiline text
`width`	Output width in pixels. 0 uses the width of `image_1` after its reference scaling, or 2048 without images. Default: 0.	INT	Yes	0 to MAX_RESOLUTION (step 32)
`height`	Output height in pixels. 0 uses the height of `image_1` after its reference scaling, or 2048 without images. Default: 0.	INT	Yes	0 to MAX_RESOLUTION (step 32)
`images`	Growable group of reference pictures, `image_1` upward; `image_1` is the picture to edit. The prompt addresses them as `<Picture n>`.	AUTOGROW	No	0 to 100 images

Outputs

Output Name	Description	Data Type
`positive`	Conditioning for the MiniMax H3 Image sampler, with the reference latents attached when images are connected.	CONDITIONING
`latent`	Empty latent for one image, 24 channels at a sixteenth of the output size.	LATENT

Note: The model is tuned for 2048x2048 pixels of area, so leave width and height at 0 unless a different canvas is needed.

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): b0927bbc44ac82e8a23b1edb5470c0cc8945844c326a352d75495ed801d11b1d