Heads up: statuses are automated heuristics from commit dates, registry metadata and issue titles. "Stale" can mean "finished and stable"; "Error spike" means error-titled issues are unusually high for that pack lately — worth checking recent issues, not a verdict that it is broken. Always check the repo.

ComfyUI-MiniMax-H3-Turbo Active

4-step turbo sampler + LoRA loader for MiniMax-H3 joint video + audio generation.

Why: last commit 66 days ago

Repo	Larryvrh/ComfyUI-MiniMax-H3-Turbo (default branch main)
Registry	comfyui-minimax-h3-turbo · publisher larryvrh · latest 1.2.3 (2026-08-07) · in Manager list: no
Popularity	rank #209 · 62,058 downloads · 606 stars
Tags	Video LoRA management Audio Loaders, quantization & speed Samplers & schedulers Animation Generic / any
Star velocity	collecting (nightly snapshots started 2026-10-19)
Error issues	0 error-titled issues opened in last 30 days (open+closed) · 0.0 per 100K downloads · baseline 1.5/month over prior 6 months (9 total) · 0.0× baseline (Error spike needs ≥3, ≥2× baseline, ≥0.5 per 100K)
Last commit	2026-08-14 (66 days ago)
Open issues	13 issues, 5 PRs (fetched 2026-10-19)
pyproject.toml	present · version 1.2.3 · requires-comfyui — · frontend — · requires-python >=3.10 · 0 deps
Registry ComfyUI req.

Error-titled issues, last 30 days

None

Open issues mentioning errors / breakage

#28 Community LoRAs (realism/tutu) silently fail on pruned int8 base - double diffusion_model. prefix in key map 2026-08-15
#27 Reference audio crashes the sampler on pruned checkpoints (adaln row count mismatch) 2026-08-14
#17 custom turbo lora loader broken after comfyui update 2026-08-10
#14 Doesn't work on Apple silicon 2026-08-10

Recent commits

2026-08-14 [main] Merge pull request #16 from afkaf/main
2026-08-12 [main] Merge pull request #10 from bkunbargi/fix-audio-ref-adaln-rows
2026-08-10 [main] Fix modulation segment calculation with lookup from `adaln_t_table`
2026-08-08 [main] Fix adaln injection row mismatch with audio reference conditioning
2026-08-07 [main] README: v4-600 vs v1-850 decision tree — v1-850 is friendlier for 4-step heavy motion