**[MiniMax H3 Open Source Not Yet Complete: 2K Video Module Coming Soon, Local Generation to Speed Up]** According to monitoring by Dongcha Beating, the MiniMax H3 team revealed during a Reddit AMA that several capabilities of the H3 will continue to be opened up. The 2K video module, which currently requires the official API to use, has been confirmed for release. Additionally, an acceleration solution aimed at reducing computational load will also be made available, enabling H3 to run locally faster and more resource-efficiently. At present, H3 can only generate 768p videos locally. The upcoming release of the 2K module will allow existing videos and original reference materials to be reprocessed through the model, generating high-resolution videos that further restore text, characters, and visual details. In the future, it is expected that a complete 2K workflow can be run locally. The team also acknowledged two existing issues. When H3 generates content while referencing images, videos, and audio simultaneously, the visuals are indeed more prone to blurring. Additionally, distant scenes and small characters are more likely to suffer from blurred faces or distortions. These issues are inherent to the model itself, and MiniMax is actively working on optimizations. [Original Link]