VELA 1.0 speeds up MiniMax H3 video gen: 0.8MP render drops from 2:53 to 2:13
NoMouse9610 · reddit · 2026-10-08
After weeks of tuning MiniMax H3/VDN performance, a developer released VELA H3 1.0 — not a new model, but a smarter execution strategy: fast kernels only where numerically safe, exact paths where it matters.
Results (RTX 3090 Ti 24GB)
- 0.8MP / 5s: VDN 1.1 took 2:53, with VELA 2:13
- More striking: raw H3 at 20 steps vs VDN+VELA at 8 steps — anime 8:07 → 3:03, realistic 7:50 → 2:52, with near-identical quality
- Honest caveats: gains shrink with longer videos and slightly regress at 10s; the 2.7x figure is not VELA alone
The author rejected attention backends, quantization, low-rank approximations and other approaches that looked faster in isolation but hurt overall output. VELA is patch-free (plain ComfyUI custom node install), listed on Comfy Registry, with all benchmarks and failed experiments open-sourced.
More from Multimodal
- Midjourney prompt shares Kirlian-style glowing aura portraits with --v 8.2 — tisch_eins · 2026-10-08
- InSpatio-World 1.5 open-sources real-time 4D world model with multi-input support — liuziwei7 · 2026-10-08
- Opus 5.5 drives Blender+Suno via MCP, delivers sync'd 20s masterpiece in 3.5 hours with zero keyframes — sidahuj · 2026-10-08
- User shares strikingly realistic GPT-6 fantasy hunter character portrait — TheUltraBased · 2026-10-08
- LLM-coded Blender beats video models at controllable video generation — jyangballin · 2026-10-08
- KAIST LipForcing lip sync glitch: teeth blend into white block during fast facial movement — Neat_Lab_4435 · 2026-10-08