h3 studio: native Metal web UI for MiniMax-H3 video gen on Apple Silicon, model stays resident
janishar · reddit · 2026-09-12
Doing video diffusion on a Mac, the author kept hitting the same wall: no real MLX support in ComfyUI, and PyTorch's mps backend is slow and eats far more unified memory than needed — painful when that memory is shared with everything else.
h3 studio is a Go server, stdlib-only, wrapping h3.c (a native Metal engine for MiniMax-H3) into a day-to-day UI:
- Model loads once and stays resident — iterate on prompts without repaying load cost
- Explicit drag-to-reorder reference ordering for multi-image conditioning
- Three ways to continue a shot: chain the last frame, use it as a reference, or reuse the whole clip
- Timeline panel to stitch takes into one sequence
- Full parameter sidecar per render
- Live profiling with load vs. denoise cost broken out
Everything runs locally; MIT licensed.
Related event: h3 Studio Runs MiniMax-H3 Video Generation Natively on Apple Silicon(2 posts)→
More from Infra
- Aristotle system machine-checks Ibex RISC-V core MUL operation formally — satnam6502 · 2026-10-31
- Clover-35B-A3B Quantized to 29.4 GB Runs at ~42 tok/s on a 64 GB Mac — viveky259 · 2026-10-31
- Liquid Agent brings fully local LFM3 agent to Linux with one-command install — JosephJacks_ · 2026-10-31
- A fully local office AI stack: MiniCPM, Qwen3, LFM2.5 and Whisper — HowDevelop · 2026-10-31
- a16z: GPU rental prices up 106% in 3 months while token prices fall 58% — a16z · 2026-10-31
- vLLM nightly adds PLE disk offload, with a telling [Qwen4Exp] tag on the PR — rmhubbert · 2026-10-31