zxcholmes/minimax-h3-studio is a self-hosted tool that runs MiniMax-H3 audio and video models on a single GPU server and provides a browser UI for the work. There is no publicly available latest release. The project description says it loads ComfyUI as a Python library without a screen interface, and makes use of quantized checkpoints and dynamic VRAM offloading.
Besides video generation, the tool offers image generation and editing with several image models, 3D generation from a single image, speech and music generation and transcription, and video upscaling. It supports text and images as input, as well as reference images and videos, and also documents turbo LoRA. Jobs enter a persistent queue and run one at a time. The README also explains that scripts are included to help with deployment to remote GPU environments and with checkpoint downloads. The repository is written in Python, is licensed under Apache-2.0, and has 100 stars.
With no published release information, there is no basis for judging version-by-version changes or compatibility. Also, because it handles jobs one at a time, it may not suit environments that need concurrent job processing. Its scope is presented as a tool for using several generation features in one web studio on a single GPU, but operational suitability depends on how the queue works and on the deployment requirements.