Skip to content

How-ToGPU

How to Run MiniMax H3 Image Locally in ComfyUI

ComfyUI natively supports MiniMax H3 Image, an open-weights model for text-to-image and reference-image editing in six fixed steps at up to 4096x4096. Update ComfyUI, download the diffusion model, VAE and Qwen3-VL text encoder, load the template, and connect images and the VAE to edit.

By , Editorial team

3 min read

How to Run MiniMax H3 Image Locally in ComfyUI

MiniMax H3 Image is MiniMax's open-weights image model, and since 10 November 2026 ComfyUI supports it natively. One checkpoint handles both text-to-image and editing with reference images, it always samples in six steps, and it can output images as large as 4096×4096. This guide is based on the Comfy team's announcement and ComfyUI's update guide; Forge PC hasn't run the model on its own hardware, so we can't give speed or VRAM figures, and Comfy's announcement doesn't list hardware requirements.

What makes it different

According to Comfy:

  • Generation and editing in one model. Connect reference images and it edits; leave them disconnected and it's text-to-image. Same checkpoint, same graph.

  • Six steps, nothing to tune. The model is CFG-distilled and samples in six steps at guidance 1. Comfy says those values are fixed by the checkpoint, so there's no negative prompt to write.

  • Native high resolution. Supported sizes are 4096×4096 (1:1), 4736×3552 (4:3), 5024×3360 (3:2), 5472×3072 (16:9) and 6272×2688 (21:9). The 4:3, 3:2 and 16:9 sizes also run in portrait. Comfy says images are generated at that size, not upscaled.

  • Long prompts. Up to 4,096 tokens, and Comfy says long, complete descriptions work best. MiniMax offers an H3 Image Prompt Enhancer model to help write them.

You can also run it through ComfyUI's Partner Nodes via the API, with no weights to download, but this guide covers running it locally.

Step 1: Update ComfyUI

Comfy's first step is to update ComfyUI to the latest version. ComfyUI's update guide covers each install type: in ComfyUI Portable for Windows, run update_comfyui.bat (latest development version) or update_comfyui_stable.bat (latest stable) from the update folder; in Comfy Desktop, open the Manage panel and use the Update tab; for a manual Git install, run git pull and then pip install -r requirements.txt inside ComfyUI's virtual environment. If you haven't installed it yet, see how to install ComfyUI on Windows.

Step 2: Download the three model files

Comfy says to download the MiniMax H3 Image weights from Hugging Face, and that you need:

  1. The diffusion model

  2. The VAE

  3. The Qwen3-VL text encoder

The text encoder is shared with MiniMax H3, MiniMax's video model that ComfyUI already supports, so if you already run that you have this file.

Step 3: Load the template and prompt

Load the MiniMax H3 Image template from ComfyUI's Templates panel, or download the text-to-image and image edit workflows from Comfy's repository. Write your prompt and run.

In the MiniMax H3 Image Conditioning node, width and height default to automatic: left at 0, text-to-image produces a 2048×2048 image. Set both values to choose a different size.

Step 4: Edit with reference images

To edit, connect images to image_1 onward:

  • image_1 is the picture being edited; the others are references.

  • Refer to them in the prompt as <Picture 1>, <Picture 2> and so on. Numbering follows the input slot order.

  • Connect the VAE to the node when you edit. Comfy says that with it, reference images are anchored as latents as well as read by the text encoder; without it, they only condition the text encoder.

  • With images connected and size left on automatic, the output follows image_1's size.

If you're short of VRAM

Comfy's announcement doesn't list hardware requirements. If you run into memory errors, try the default 2048×2048 output first and see our guide to running ComfyUI on low VRAM. For other local image tools, see ComfyUI vs Forge WebUI.

Our verdict

Update ComfyUI, download the diffusion model, VAE and Qwen3-VL text encoder, load the MiniMax H3 Image template and write a long, detailed prompt. For edits, connect your images from image_1 and remember to connect the VAE. Start at the default size and work up if your GPU copes.

The bottom line

Update ComfyUI, download the diffusion model, VAE and Qwen3-VL text encoder, load the MiniMax H3 Image template and write a long, detailed prompt. For edits, connect your images from image_1 and remember to connect the VAE. Start at the default size and work up if your GPU copes.

Sources

  1. 01Comfy Blog, MiniMax H3 Image is now available in ComfyUI
  2. 02ComfyUI Docs, How to Update ComfyUI

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Speed Up LLM Text Generation in ComfyUI With MTP

How-To

How to Speed Up LLM Text Generation in ComfyUI With MTP

ComfyUI v0.45.0 rebuilt local LLM text generation for Qwen3.x models with multi-token prediction and adaptive tree drafting, which Comfy says can roughly double token rates on the same GPU. Update ComfyUI, leave mtp on auto in the Text Generate node and measure your own gains.

Forge PC Staff3 min

How to Use the LM Studio CLI (lms): Load, Chat and Serve

How-To

How to Use the LM Studio CLI (lms): Load, Chat and Serve

LM Studio's lms command-line tool lets you list models with lms ls, see loaded ones with lms ps, load with context length, GPU offload and TTL options, check memory with --estimate-only, chat with lms chat and start the local API with lms server start.

Forge PC Staff3 min

NVIDIA Releases GeForce Hotfix 621.35 to Fix Excessive SSD Writes

News

NVIDIA Releases GeForce Hotfix 621.35 to Fix Excessive SSD Writes

NVIDIA's GeForce Hotfix Driver 621.35, posted on its Customer Care site on 20 November 2026, fixes excessive background writes to the nvlog.nvlgstg file seen with drivers 621.14 and 621.24. NVIDIA says it's an optional beta based on its latest Game Ready driver; unaffected users can wait for the next WHQL driver.

Forge PC Staff2 min