SCAIL-2 Video Generation Takes 20 Mins on RTX 6000: Dev Seeds Speed Optimization
Cloud9_pilot · reddit · 2026-07-30
A developer took to Reddit seeking help for slow local inference, noting that generating a 10-15 second video clip using the SCAIL-2 model in ComfyUI takes up to 20 minutes on an RTX PRO 6000 Blackwell (with 96GB VRAM).
Given the high-end hardware, they feel this is significantly slower than it should be. They asked the community for practical ways to cut down generation time, such as:
- Using quantized model variants
- Tweaking LiteX2V distillation settings
- Adjusting TeaCache configurations
- Switching attention backends
More from Infra
- US Commerce Dept Allocates $874M to Accelerate Semiconductor R&D — imjustnewatai · 2026-07-30
- Qualcomm Q3 revenue beats estimates but weak Q4 EPS guide weighs — firstadopter · 2026-07-30
- Sam Altman Understands Why People Don't Want AI Data Centers in Their Backyards — businessinsider · 2026-07-30
- Zuckerberg: We're getting compute offers at a significant premium — firstadopter · 2026-07-30
- Dual GPU inference with RTX 4090 + 3060: speed impact and optimization tips — cosmoschtroumpf · 2026-07-30
- Future 100T Param Model to Cost >$250B to Train, Says Joseph Jacks — JosephJacks_ · 2026-07-30