Skip to content
View original post on X: SGLangOfficial
Published
62/100

SGLang Diffusion adds Day 0 support for MiniMax-H3-Image

AISGLang says its SGLang Diffusion runs MiniMax-H3-Image at full precision on Day 0, locally and in the cloud. It reports 4K generation in 10s on a single H200, 18.5s on a single RTX PRO 6000, and 85s on DGX Spark.
SGLangVerified on XOfficial
@sgl_project
View on X

MiniMax-H3-Image is an open-weights model that generates and edits images at native 4K.

We made a full set of SGLang swag with it within 5 minutes on one RTX PRO 6000.

On Day 0, SGLang Diffusion runs it at full precision, locally and in the cloud:
• DGX Spark: 2K image in 16s, 4K in 85s
• Single RTX PRO 6000: 2K in 3.4s, 4K in 18.5s
• Single H200: 2K in 1.9s, 4K in 10s
• Native multi-GPU inference support
• Reproducible results with fixed seeds
• AdaLN precomputation saves 24.6 GiB of VRAM
• Recipes for DGX Spark, RTX PRO 6000, H200 and B300

MiniMax (official)@MiniMax_AI
4K image generation in 10 seconds! ⚡ Meet MiniMax H3-Image. 16 million pixel resolution. Stunning realism. Precise prompt following & editing. Up to 16 reference images. All in 10 seconds. And it's open-weight. 🚀 Available now via our API, MiniMax Code, and Hailuo AI — with 50% off 16MP+ generations during launch week! Can't wait to see what you create. 🔥 Hugging Face: http://huggingface.co/MiniMaxAI/MiniMax-H3-Image API: http://platform.minimax.io/docs/guides/image-generation MiniMax Code: http://code.minimax.io/about
View quoted post on X
Source details

SGLang · x.comPublished · added here

Original title: MiniMax-H3-Image is an open-weights model that generates and edits images at native 4K.

Topics

Image generationDeployment & engineeringChina AIProduct updates