Radar / AI media tools / MiniMax-H3-Image
MiniMax-H3-Image: a 33B open-weight image model that renders 16MP stills in 6 steps
MiniMax quietly shipped the weights for H3-Image, a 33.1B open-weight image model that generates and edits 16-megapixel images in six sampling steps with no classifier-free guidance, taking up to 16 reference images — though its community licence bars use in the US, EU, UK and South Korea.
Why it matters
Open-weight 4K image generation and editing in a single checkpoint, distilled to six steps so a 16MP image is an interactive request, not a batch job — with a territorial licence that excludes four major markets.
What you could build with it
An indie game studio could generate consistent 4K key art and character sheets with H3-Image: feed up to 16 reference images of a character, iterate in six-step generations, and keep a local SGLang server as the art pipeline's render farm.
Does it hold up?
Too early to judge: weights landed 10 November with no independent benchmark or image-board placement yet; the vendor showcase grid is the only quality evidence, and local use needs a 135GB download including the borrowed text encoder.
Built with MiniMax-H3-Image
- MiniMax-H3-Image vs Nano Banana 2.1: price vs proofarticle · Head-to-head on price ($0.0218 per 1K image vs $0.0336) and evidence: Nano Banana 2.1 has thousands of leaderboard samples, H3-Image has no Artificial Analysis entry at all yet.
- MiniMax-H3-Image vs MiniMax-H3: same family, stillsarticle · Architecture teardown: H3 with the audio branch removed, the 6-step distilled schedule, the borrowed Qwen3-VL text encoder, and the licence terms including the $1M revenue authorisation threshold.
- MiniMax releases open-weight H3-Image model with 4K output on Hugging Facearticle · Release coverage: 33.1B parameters, 4096x4096 native output, ~$0.022 per 1K image on the API, ComfyUI support at launch, and the licence excluding the US, EU, UK and South Korea.
- comfyanonymous/ComfyUIgithub · ComfyUI supported H3-Image immediately at launch, with text-to-image and edit workflow templates linked from the model card.
Learn more
First spotted on hf: source.
More AI media tools
Hyperframes: write HTML, render video — HeyGen's open video engine for AI agentsOpen-source renderer from HeyGen that turns HTML, CSS and JavaScript into MP4 video, built so AI coding agents can…media · JEV 0.78Video Shotcraft: AI video skill for Claude Code & Codex that directs cinematic product promos in RemotionA skill package for Claude Code and Codex that turns a product brief into a finished promo video: the agent captures…media · JEV 0.71Lemo-Opuscar: 59 code-directed film styles for Claude CodeA toolbox of 59 film styles, each a style prompt plus a demo film made entirely in code by Claude Opus 5.5 with no…media · JEV 0.71SolidCraft: open-source parametric CAD in Rust that agents can drive over MCPSolidCraft is an open-source parametric 3D CAD application written in pure Rust (desktop, browser via WebAssembly…media · JEV 0.67Concat: free, open-source, local-first CapCut replacement scriptable by agentsLocal-first, cross-platform video editor that replaces CapCut, scriptable through a JSON-RPC API and MCP so coding…media · JEV 0.66Qwen-Image-2.1: 7B open image model that tops Nano Banana 2.0Alibaba's unified generation-and-editing image model: native 2K, native transparent-background output, up to 10…media · JEV 0.66
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.