MiniMax H3 roundup: 1.6x deep-reuse caching, torchao quantization, manga LoRA and faster training
optimisticalish · reddit · 2026-09-12
- ComfyUI FirstBlockCache: cross-step caching for 20-step users, 30% faster natively; a new fifth experimental deep-reuse cache mode hits 1.6x speed and runs on an RTX 3060 12GB.
- MiniMaxH3Torchao018: a quantized build targeting torchao >=0.18 that significantly cuts inference VRAM while keeping fidelity.
- Manga Tone Rendering LoRA: teaches H3 Japanese b&w screentone shading that holds up under dynamic camera motion; trigger words: manga-tone rendering.
- 15+ reference image workflow: a ComfyUI patch removes the nine-reference limit for Ref2VA (ships with an undo script; .BAT patch caution).
- Plus an age-slider LoRA (toddler to elderly demo), a style-transfer LoRA with matching workflow, and the Fizgig LoRA trainer's new ultra quality mode — better visual/audio results and 30% faster training (2.6→3.4 steps/s on int8, now default; 16GB card required).
More from Multimodal
- NVIDIA open-sources NemotronLabs RE-USE, a 9.6M-param multilingual speech denoiser — NVIDIAAI · 2026-10-31
- AI-generated fake trailer imagines a 1999 Dino Crisis movie with retro VHS feel — noizlab_studio · 2026-10-31
- Spatial Hub aggregates ~5000 public COLMAP-compatible spatial datasets for world models — janusch_patas · 2026-10-31
- Single image to walkable 3D Gaussian splat scene using World Labs Atlas and WebGPU — XRarchitect · 2026-10-31
- AI video reimagines Suicide Squad's Joker as a rapper in a full music-video treatment — Silver_War5476 · 2026-10-31
- 60+ community-made ChatGPT slash-style words turn one word into a full visual style — heyshrutimishra · 2026-10-31