MiniMax releases the model weights for H3-Image, a system that generates images from text and edits existing images. Multiple outlets report that the release includes performance claims such as producing high-resolution outputs—around a 16-megapixel image (about 4,096×4,096 pixels)—in roughly under 10 seconds, depending on the workflow.
The coverage also highlights how the model can use multiple reference images in a single request for editing or guidance. TechNode emphasizes practical speed and that the weights are made available on Hugging Face, while Pandaily focuses on the model’s architecture and generation flow, describing it as a large diffusion transformer and noting it can produce images in six steps. Pandaily also adds that the release is governed by a community license with territorial restrictions.
Overall, both accounts agree the weights are public and that H3-Image supports both text-to-image generation and image editing using up to 16 reference images.