MiniMax H3 Image

作者：CivitaiOfficial

MiniMax H3 Image is a text-to-image and image editing model from the MiniMax H3 family. One model does both: it creates images from a prompt, and it edits or combines up to 16 reference images. It renders about 16 megapixels natively (4096x4096 at 1:1, up to 21:9) with no upscaler, and it follows long, detailed instructions closely, including exact text.

Originally released by MiniMax as open weights on Hugging Face. All credit for the model goes to the MiniMax team. Civitai runs it on its own GPUs so you can generate on-site with no setup. Head to the original repo for the weights, the license and updates.

What it does

• Text-to-image: high-resolution images at aspect ratios from 1:3 to 3:1, in 1K, 2K or 4K.

• Image editing: change part of an image and keep the rest. Swap an object, restyle a scene, replace text, or turn a 3D blockout into a film still.

• Multi-reference composition: combine up to 16 reference images in one generation, for example an empty room plus nine pieces of furniture into one furnished living room.

• Text in images: signs, posters, labels and layouts with the exact words you give it.

On Civitai

MiniMax H3 Image is available in the Civitai generator for text-to-image and editing. Generation runs on Civitai's own GPUs, so there is no download on this page; get the weights from the original repo. LoRA training is not available for it yet.

How to prompt it (tips from MiniMax)

• Describe the task before the style. Say what the image is for and who or what is in it, then add style.

• Set the layout before the details. Give the number and placement of subjects, the angle and the whitespace first, for example "Title at the top; image on the left and text on the right below."

• Show style through visible details. "Soft side light, metallic reflections, clean whitespace" works better than "premium feel."

• Put exact text in quotes, and say where it goes and how it breaks across lines.

• Give each reference image a role. "Keep the product from image 1, use image 2 for color, and image 3 for layout."

• Say what changes and what stays. "Change the cup on the left to blue, keeping its shape, material, position and size the same."

• Spell out grids as rows and columns: "2 rows and 3 columns", not "2x3", which can flip.

Links

• Source: MiniMaxAI/MiniMax-H3-Image on Hugging Face

• Code: MiniMax-AI/MiniMax-H3 on GitHub

• API guide: MiniMax image generation docs

• Community: MiniMax Discord