Sparse attention node gives H3 Minimax in ComfyUI up to 2.5x speedup
Plague_Kind · reddit · 2026-08-21
A new sparse attention SLA node in a ComfyUI node pack speeds up H3 Minimax video generation by up to 2.5x, without requiring the SLA LoRA and compatible with any turbo.
Usage notes:
- If you hit OOM, add comfykitch attention before it for another 5-10% speedup;
- Requires a recent PyTorch and CU130;
- Place it after your LoRA loader for now; other positions untested.
Repo: ComfyUI-PlagueKind-Nodes; design credited to pl0x.
Related event: New ComfyUI Sparse Attention Nodes Speed Up H3 Video Generation(2 posts)→
More from coding & agent
- Leerob's Stanford Lecture Breaks Down Always-On Agents, Harnesses and Context Engineering — dean_rie · 2026-10-16
- Dev Praises Codex Local Computer Use: 'Every Other Harness Disappoints Me' — craigsdennis · 2026-10-16
- Decision models match LLMs on routing and safety — and are 7-8x faster — randal_olson · 2026-10-16
- Temporal Serverless Workers on Cloud Run: queue-driven autoscaling, scale to zero — rseroter · 2026-10-16
- Meta's Superintelligence Lab Runs 200+ Direct Reports—A Lesson for Flat Multi-Agent Systems — peterjliu · 2026-10-16
- LangChain CEO: the most interesting agent harness work lives at the harness-infra boundary — hwchase17 · 2026-10-16