# MiniMax H3 Image is natively supported in ComfyUI on day-0 — retained record

> Retained reddit record from reddit:comfyui.

Canonical: https://highsignal.app/data/records/5695f28f3721370b

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:comfyui)  published 2026-11-10 16:39 ingested 2026-11-11 02:42

[original source ↗](https://www.reddit.com/r/comfyui/comments/1xsrnze/minimax_h3_image_is_natively_supported_in_comfyui/)

## retained text



```
MiniMax H3 Image now runs natively in ComfyUI as soon as it drops! One checkpoint does both text-to-image and reference-image editing: connect images and it's an edit workflow, leave them off and it's text-to-image.

# Highlights

* **Six steps, nothing to tune.** The model is CFG-distilled and samples in 6 steps at guidance 1. No negative prompt.
* **Native high-res output.** 4096×4096 at 1:1, 4736×3552 at 4:3, 5024×3360 at 3:2, 5472×3072 at 16:9, and 6272×2688 at 21:9. Generated at that size, not upscaled.
* **Multi-reference editing.** Connect several images and refer to them in the prompt as `<Picture 1>`, `<Picture 2>`, etc. Numbering follows input order.
* **Long prompts.** Up to 4096 tokens, and detailed descriptions work best. Use the Prompt Enhancer for optimal results.
* **Shared text encoder.** It uses the same Qwen3-VL text encoder as the MiniMax H3 video model, so if you already run H3 you have that file.

# Tips for the MiniMax H3 Image Conditioning node

* Leave width and height at 0 for automatic sizing: 2048×2048 for text-to-image, or the size of `image_1` when editing.
* `image_1` is the picture being edited. The other slots are references.
* Connect the VAE when you edit, so reference images are anchored as latents and not only read by the text encoder.

# Getting started

1. Update ComfyUI to the latest version.
2. Download the diffusion model, VAE, and Qwen3-VL text encoder from [Hugging Face](https://huggingface.co/Comfy-Org/MiniMax-H3-Image).
3. Load the MiniMax H3 Image template from the Templates panel, or grab the [text-to-image](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_minimax_h3_image_t2i.json) and [image edit](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_minimax_h3_image_edit.json) workflows.

If you'd rather not download weights, it's also available on [Comfy Cloud](https://cloud.comfy.org/?template=image_minimax_h3_image_t2i).

Full blog post: [https://blog.comfy.org/p/minimax-h3-image-is-now-available](https://blog.comfy.org/p/minimax-h3-image-is-now-available)

Let us know what you make! 
```



document fetched 2026-11-11 00:42 · canonical https://reddit.com/r/comfyui/comments/1xsrnze/minimax_h3_image_is_natively_supported_in_comfyui
