ComfyUI MiniMax H3 cache node cuts generation time by 30–33%
Reddit user nnorbbi shares a lightweight, dependency-free ComfyUI patch node adding FirstBlockCache to MiniMax H3, cutting generation time 30–33% on an RTX 5090. Benchmark: 90.64s → 60.82s at 20 steps with INT8 ConvRot.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- CAS creates multi-agent coding factory for Claude Code
- MiniMax H3 update brings 2K, 5× turbo, camera previz
- MiniMax H3 clip chaining keeps motion and audio continuous across joins
- Chinese AI Chipmakers Poised to Gain From Beijing’s Tech Push
- Panther CEO Jack Naglieri: Using AI to build in the open is a good pattern