MiniMax releases open-weight H3-Image model; vLLM, SGLang, fal and ComfyUI add support
Article timeline
The articles in this story. Times are ET.
MiniMax Design (H3)OfficialMiniMax H3-Image accepts up to 16 reference images and generates in 10 secondsAIMiniMax says its H3-Image model accepts up to 16 reference images and generates output in 10 seconds. A blogger tested it by combining four animal characters, a steampunk airship and a cloud-floating kingdom into one cinematic scene.
- MiniMax Design (H3)OfficialMiniMax invites game developers to try its H3 Image model
AIHailuo AI, the account of MiniMax, invites game developers to start their next world with an #H3Image prompt. A quoted post from @Israa_Ali2077 says it tested MiniMax H3 Image on AAA-style game assets, 3D environments and gameplay-style firefight concepts.
- Comfy BlogOfficialComfyUI adds native support for MiniMax H3 Image
AIComfyUI now supports MiniMax H3 Image natively, according to its official blog. The open-weights model handles both text-to-image and reference-image editing in a single checkpoint, with six sampling steps and output up to 4096×4096. Users can run it on their own GPU with downloaded weights or through Partner Nodes via API.
falOfficialfal offers MiniMax H3 Image for text-to-image and image editingAIfal makes MiniMax H3 Image available on its platform, with text-to-image and image-editing models plus highspeed variants. The post links to separate endpoints for each mode and speed option.
- falOfficialMiniMax H3 Image delivers native 16MP resolution and 2K editing in about 15 seconds
AIMiniMax says its H3 Image model generates native 16MP images at any aspect ratio. The company also says 2K image editing takes about 15 seconds.
- falOfficialMiniMax H3 Image is now available on fal
AIfal makes MiniMax H3 Image available, the latest open-weights image model from MiniMax. The model generates and edits images in native 4K, combines multiple references, and renders complex text with precision.
SGLangOfficialSGLang shares a cookbook for running MiniMax-H3-ImageAISGLang publishes a cookbook for MiniMax-H3-Image, a diffusion model, on its documentation site. The post links to the guide at docs.sglang.io, with no further details given in the post.
- SGLangOfficialSGLang 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. It also notes support for native multi-GPU inference and fixed-seed reproducibility, and that AdaLN precomputation saves 24.6 GiB of VRAM.
- MiniMax Design (H3)OfficialMiniMax releases open-weight H3-Image model for 4K image generation
AIMiniMax releases H3-Image, an open-weight image generation model that produces 4K images at 16 megapixels in about 10 seconds. The model supports precise prompt following, editing, and up to 16 reference images, and is available through MiniMax's API, MiniMax Code, and Hailuo AI, with 50% off 16MP+ generations during launch week.
vLLMOfficialvLLM-Omni adds support for MiniMax-H3-Image on NVIDIA and AMDAIvLLM-Omni now supports MiniMax-H3-Image, which the project says is available as an official Docker image, vllm/vllm-omni:minimax-h3-image. The project says both NVIDIA and AMD accelerators are supported, including NVIDIA H200, RTX PRO 6000 Blackwell, and AMD Instinct MI350X, with per-hardware configurations and latency measurements listed on its recipes page.
- vLLMOfficialvLLM-Omni adds day-0 support for MiniMax H3-Image on NVIDIA and AMD GPUs
AIvLLM says its vLLM-Omni framework adds day-0 support for MiniMax's open-weight MiniMax-H3-Image, available on both NVIDIA and AMD GPUs. The model handles image generation and editing at native resolution of about 16 megapixels, with multi-reference editing of up to 9 images per request in the release recipe. It accepts prompts up to 4,096 tokens and renders dense text.
MiniMax (official)OfficialMiniMax releases open-weight H3-Image model generating 4K images in 10 secondsAIMiniMax releases H3-Image, an open-weight image generation model that produces 16-megapixel images in about 10 seconds. The model supports precise prompt following and editing, and accepts up to 16 reference images. It is available now through MiniMax's API, MiniMax Code, and Hailuo AI, with 50% off 16MP+ generations during launch week.
ModelScopeOfficialPickMiniMax releases H3-Image, an open-weight model for 4K image generation and editingAIModelScope says MiniMax has released H3-Image, an open-weight model for high-resolution image generation and editing. According to the post, it generates 4096x4096 images up to 6272x2688 directly, supports edits and compositions from up to 16 reference images, and runs on SGLang, vLLM, diffusers, and ComfyUI. The post also says it is distilled to 6 steps with no CFG, and links to the model and a prompt expander.
- MiniMax (official)OfficialPickMiniMax releases H3-Image model and its open weights
AIMiniMax announces H3-Image, a new image model built on its H3 architecture, and says it is releasing the model weights on Hugging Face. The company says the model generates 4K × 4K images in 10-20 seconds, supports up to 16 reference images, and is available via its API, MCode, and Hailuo AI, with 50% off 16 million pixel+ generations during launch week.
