Direct with references
Group pictures of one character, add instructions, guide from storyboards and preview selected video or audio ranges.
SELF-HOSTED • WORKFLOW-AWARE • OPEN SOURCE
A complete browser workspace for MiniMax H3 video with native audio and Qwen Image creation. Work on your own NVIDIA machine or GPU server, with no commercial API key required for generation.
ONE CONNECTED WORKSPACE
Supported selections configure the connected ComfyUI workflow. Compatibility rules apply required settings, preserve saved choices and block unresolved conflicts before rendering.
Group pictures of one character, add instructions, guide from storyboards and preview selected video or audio ranges.
Text, first/last frames, references, supported LoRAs, ControlNet, continuation and refinement have explicit workflow contracts.
Projects, scenes, reusable Elements, independent Reuse setups and portable exports preserve your creative setup.
Use a separate Qwen Image creation and editing workspace, then bring the approved image into a shot.
Automation uses deterministic configuration rules. Optional AI prompt preparation requires a configured local or remote model. Output quality still needs creative review; unknown LoRA requirements cannot be inferred automatically.
RUN IT YOURSELF
Start from the source outside ComfyUI’s custom_nodes folder. The launcher prepares supported dependencies and resumes model downloads. Model weights need substantial storage, RAM and VRAM.
git clone https://github.com/underworldhistory1-ctrl/minimax-h3-interface.git
cd minimax-h3-interface
bash launcher.sh --comfy-root /path/to/ComfyUI --models-root /path/to/modelsWindows: ./launcher.ps1 -ComfyRoot C:/path/to/ComfyUI
INSPECTABLE BY DESIGN
The installer source, compatibility pins, tests and release history are public. The default updater follows official main; use --no-update when you need to retain reviewed source. Model checks and Git pins do not replace a fully locked dependency supply chain.
Source tests, browser acceptance and GPU output evidence are reported separately. No software project can promise every possible combination is free of bugs. Default access is local; public GPU deployments need access protection.