Long clips Any mode 4 steps
Long-clip recipe: SLA sparse attention + SLA Turbo LoRA
The longer the sequence, the bigger the win: measured 44→31/25 s per step on a 5090 at 768p/15s (0.85/0.90 sparsity). Must be used as the SLA kit.
How to run it
- 1 Install the SLA Attention node AND switch to the SLA-trained LoRA (both required)
- 2 Keep the audio block at 64 (128 makes dialogue sound muffled)
- 3 Skip SLA for short clips (<5s) — plain Sage is fine
Enabling the sparse route on non-SLA weights visibly degrades quality (see sla-pairing in the conflicts table).
Representative GPUs
864 × 480 · 8s · 32GB RAM / fast-disk; click a GPU for its full matrix.
| GPU | VRAM | Status | Per clip | VRAM peak |
|---|---|---|---|---|
| RTX 4060 8GB | 8GB | Slow | 5 min 8 s | 7G |
| RTX 3060 12GB | 12GB | OK | 4 min 53 s | 11G |
| RTX 4070 12GB | 12GB | OK | 2 min 52 s | 11G |
| RTX 4070 Ti Super 16GB | 16GB | OK | 2 min 7 s | 15G |
| RTX 4080 Super 16GB | 16GB | OK | 1 min 54 s | 15G |
| RTX 4090 24GB | 24GB | OK | 1 min 24 s | 23G |
| RTX 5090 32GB | 32GB | OK | 1 min 3 s | 26G |
| RTX PRO 6000 Blackwell 96GB | 96GB | OK | 59 s | 47G |
1344×768 · 5s: 48 of 49 listed GPUs workable, from about 6GB VRAM.
FAQ
What's the minimum GPU for this recipe?
About 6GB+ of VRAM for 1344×768 · 5s (48 of 49 listed GPUs workable). Short 736×416 clips run on less.
How does this recipe combine with other acceleration?
Step distillation picks one option; attention backends (Sage/Kitchen/SLA/VSA) and caches stack on top. See the conflicts table for exact rules.