According to CoinMeta, the MiniMax H3 team revealed in Reddit AMA that they are developing an image model based on the H3 architecture and plan to make the weights of the model publicly available. This model integrates text-to-image generation with general image editing into a single framework and has currently entered the post-training optimization phase. The model builds upon the H3's VAE and encoder, and includes specially designed VAE and decoder for image generation. The envisioned workflow is to first use the image model to generate the first frame, which is then passed on to H3 to continue generating the video. The team also mentioned that H3 itself has already demonstrated certain capabilities for image generation and editing. Although it was previously trained only to predict subsequent frames based on the "first frame + text description," it still showed strong zero-shot capabilities in various image editing evaluations. This is an important basis for MiniMax to continue expanding this architecture to an image model as well.