ChainCatcher report: MiniMax has officially released H3-Image and open-sourced its model weights. The model can directly generate images at 4096×4096 resolution—approximately 16.8 million pixels—without requiring upscaling. According to official data, generation takes less than 10 seconds, with speeds up to five times faster than mainstream alternatives. Text-to-image and image editing are integrated into a single model, supporting up to 16 reference images. H3-Image retains the H3 video model architecture, removes the audio component, preserves the original text encoder, and features a newly designed image decoder. The core image generation network contains 33.1 billion parameters and requires only six sampling steps to generate one image. On August 7, MiniMax researcher Kiro revealed in a Reddit AMA that the team is developing a dedicated image model based on the same H3 architecture, plans to open-source its weights, and proposed a workflow where the image model generates the first frame, which is then passed to H3 for video generation. On November 3, Pruna AI launched P-H3-Image based on the open-sourced H3 video model, supporting image generation and editing. The generation network was compressed from 33.1 billion to approximately 20 billion parameters, requiring only eight sampling steps. The smallest quantized version is about 20 GB in size and peaks at approximately 21 GB VRAM when generating 8-million-pixel images. The official MiniMax version retains the 33.1-billion-parameter generation network with six sampling steps and supports higher resolutions and more reference images; the combined original weights and reused text encoder total approximately 135 GB. The open-source license includes regional restrictions: the United States, United Kingdom, European Union, and South Korea are excluded from the community license; enterprises with annual revenue exceeding $1 million or offering model hosting services to third parties must obtain separate authorization.
MiniMax open-sources its H3 image model, generating 16.8MP images in 10 seconds
ChaincatcherShare
MiniMax has open-sourced its H3-Image model, part of a broader network upgrade in the AI + crypto news space. The model generates 4096×4096 images in under 10 seconds, supports up to 16 reference images, and uses a 331-billion-parameter core network. Pruna AI has released a 200-billion-parameter version. The open-source license excludes the US, UK, EU, and South Korea. Commercial use with annual revenue exceeding $1 million requires separate approval.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.