GitHub radar

Generate and edit high-resolution images from text instructions

MiniMax-H3-Image is an image model from the makers of Hailuo: it turns a text description into a detailed, high-resolution picture and follows editing instructions precisely, even across several reference images at once.

01MiniMaxAI/MiniMax-H3-Image 934437k downloads/mo33.1B paramstext-to-image

An AI model that draws images from text and edits existing photos by instruction — both in one tool.

When it helps

Useful when you need large, highly detailed images without extra processing elsewhere, or text-guided edits that draw on several references at once. Not an option if you want to run it on your own computer — it only works on server-grade hardware.

Pros

  • Generates and edits images in a single tool
  • Produces up to 16-megapixel images with no extra processing
  • Follows text instructions better than other open models
  • Supports up to 16 reference images for edits

Cons

  • Can't run on a regular computer — needs server-grade GPUs
  • Model files total around 130 GB combined

How to set it up — step by step

  1. 1Open the model card on Hugging Face: huggingface.co/MiniMaxAI/MiniMax-H3-Image.
  2. 2Send your agent the text from the block below — it will explain what the model does and how to try it without your own server.
  3. 3Open the Hailuo AI web app and try generating an image from a text description for free.
  4. 4To integrate it into your project, get a MiniMax API key and read the docs for image generation and editing.
  5. 5If you plan to self-host it, confirm you have GPUs like NVIDIA H200 or AMD Instinct MI350X — otherwise it won't run.

Text for your agent

Copy this and send it to your agent — Claude Code, Codex, any of them:

Open the MiniMax-H3-Image model card: https://huggingface.co/MiniMaxAI/MiniMax-H3-Image — explain what it can do (text-to-image generation and instruction-based photo editing), and tell me if I can try it without my own server, through the Hailuo AI web app or the MiniMax API.

This is a server-grade model. The generator file alone is about 66 GB, and it needs the text encoder from the base model — another roughly 67 GB, so more than 130 GB combined. A home computer or laptop can't handle this: the authors recommend professional GPUs like the NVIDIA H200 or AMD Instinct MI350X, sometimes several at once.

Open on Hugging Face