Compare the official MiniMax H3 repository, h3.c, ComfyUI, Turbo, Motion Context, Audio T8, low-VRAM, and training projects by use case and hardware.
- OpenMontage
An open-source agentic video-production system spanning storyboards, assets, editing, quality gates, and multiple pipelines. It does not replace H3 inference; it is an architectural reference for putting H3 inside a production system.
Star snapshot: 49499 · 2026-08-23 - MiniMax-AI / MiniMax-H3
The official entry point for weights, deployment documentation, and nine Agent Skills. h3-prompt-writing structures shots, dialogue, and environmental audio in agents including Claude and Codex.
Star snapshot: 6732 · 2026-08-23 - h3.c / h3-metal
Antirez's native MiniMax H3 inference engine for Apple Silicon, built without Python, PyTorch, or ComfyUI. Text-to-audio-video, first/last-frame control, and ordered Ref2VA references work end to end.
Star snapshot: 2358 · 2026-08-23 - ComfyUI Wiki
A bilingual knowledge base for ComfyUI setup, nodes, models, workflows, and troubleshooting. It is not H3-specific, but fills the ComfyUI fundamentals assumed by many H3 guides.
Star snapshot: 1150 · 2026-08-23 - MiniMax H3 Audio T8
An 86-node native ComfyUI H3 audio-video suite covering stable dual-clock sampling, audio conditioning, mixing, trimming, and preflight, plus experimental long-video, speech, keyframe, and source-AV paths.
Star snapshot: 788 · 2026-08-23 - ComfyUI MiniMax H3 Director
A ComfyUI director that combines H3 shot planning, conditioning, sampling, audio-video decoding, and export. It covers T2V, FL2V, R2V, V2V, and RV2V with timelines, selective runs, inter-shot guidance, and optional refinement.
Star snapshot: 774 · 2026-08-23 - MiniMax H3 Director Workflow Pack
A reproduction-oriented pack of five Director JSON workflows for T2V, FL2V, R2V, V2V, and RV2V. It depends on the upstream AIMixer plugin and is not a separate inference engine.
Star snapshot: 722 · 2026-08-23 - ComfyUI H3 Motion Context
Carries the previous clip's H3 visual latent and audio context into the next clip so motion, rhythm, and sound continue across the join instead of restarting with a similar take.
Star snapshot: 697 · 2026-08-23 - ComfyUI Spectrum MiniMax H3
Forecasts hidden states for H3 target audio-video rows to skip selected expensive transformer evaluations while keeping the native output head, decode, and save path. A default 20-step run commonly becomes 11 actual evaluations plus nine forecasts.
Star snapshot: 555 · 2026-08-23 - ComfyUI MiniMax H3 Easy
Combines images, videos, and audio in one Media input with @ references, dialogue blocks, raw-prompt inspection, multi-provider prompt optimization, and a compact node set spanning text, keyframe, and full-reference modes.
Star snapshot: 521 · 2026-08-23 - MiniMax-H3 Turbo LoRA
Compresses the usual roughly 20-step synchronized audio-video sampling path to four to eight steps. The current v4 EMA suits most work; six to eight steps are steadier, while four is primarily for previews.
Star snapshot: 495 · 2026-08-23 - Awesome MiniMax H3 Resources
A maintained map of the official and community H3 ecosystem: base and quantized weights, text encoders, VAEs, Turbo and style LoRAs, ComfyUI nodes, tutorials, and importable workflows.
Star snapshot: 352 · 2026-08-23 - ModelTC MiniMax H3 Turbo
Provides four- and eight-step Turbo LoRAs for FL2VA and Ref2VA, Diffusers batch-inference scripts, and importable ComfyUI workflows for text, first-frame, and multimodal reference generation.
Star snapshot: 303 · 2026-08-23 - BeatAPI MiniMax H3 Prompt Gallery
A use-case-organized H3 prompt gallery with 300 prompts, WebM examples, and creator attribution across stories, action, ads, music, and social video.
Star snapshot: 117 · 2026-08-23 - MiniMax H3 FineTuning
Fills the trainer gap left by the official release with supervised rectified-flow training on the official Diffusers integration, cached video and stereo-audio latents, LoRA, and ZeRO-3 multi-GPU paths.
Star snapshot: 66 · 2026-08-23