MiniMax H3-Image for precise edits, GPT Image 2.6, and FLUX.3 [klein]
Since our November 11 update, giz.ai has a new default image model that edits pictures with remarkable precision, plus a new GPT Image version and an upgraded klein model.
MiniMax H3-Image: edits that keep what matters
MiniMax H3-Image is now the default for new images and image edits, served by giz.ai on our own GPUs. It is at its best on the edits that are hardest to get right: changing one thing in a picture while keeping everything else exactly as it was.
Every picture below was made on giz.ai with MiniMax H3-Image, each in a single run of about 15 to 25 seconds, with no retouching.
Rewrite the text in a photo, in another language, and change the season. The English sign became Korean in the same brush-script style on the same board, the scene turned to a snowy evening, and the bicycle, plants and storefront stayed as they were.


Same person, new outfit, new place. Her face stays hers while the red raincoat becomes a navy wool coat with a knit scarf, and the street becomes a Paris café at golden hour.


Combine pictures into one scene. Give it a person and a product and it puts them together: the same woman on a skate ramp at sunset, wearing the exact sneaker from the product photo, green stripe included.


- Up to 16 reference pictures in one edit; refer to them as Picture 1, Picture 2 and so on.
- 1 to 16 megapixels. Without a choice it makes 1K images; pick a larger size for print or detail.
- Write in any language. Prompts in Korean, Japanese or any other language are carried over precisely, and words you want drawn in the picture keep their own script.
- The headshot, product photo and clothes changer tools now run on MiniMax H3-Image too.
GPT Image 2.6
GPT Image 2.6 joins the GPT Image versions, in two styles: Flare and Sunburst. Paid plans include it at 1K sizes, and 2K and 4K use credits. When you or your Agent pick a GPT Image version, that exact version now makes the image.
FLUX.3 [klein] replaces FLUX.2 [klein]
The klein model included in plans is now FLUX.3 [klein]: one model for both text-to-image and edits with up to four reference pictures. FLUX.2 [klein] 9B remains available with credits.
Thank you for using GizAI.