Official Multi-reference 20 steps
Official Ref2V template (multi-reference control)
Assign roles to images, video and audio (identity / style / motion / voice) and write the prompt in the six-section reference structure; best for character consistency and motion transfer.
How to run it
- 1 Pick Reference to Video from the template library
- 2 Define each reference asset's role first (label + purpose), then describe the target timeline
- 3 Audio cannot be the only input — it must accompany image or video material
More references mean more encoder pressure; on small cards keep references to 1–2.
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 | 12 min 30 s | 7G |
| RTX 3060 12GB | 12GB | OK | 11 min 51 s | 11G |
| RTX 4070 12GB | 12GB | OK | 6 min 49 s | 11G |
| RTX 4070 Ti Super 16GB | 16GB | OK | 4 min 56 s | 15G |
| RTX 4080 Super 16GB | 16GB | OK | 4 min 23 s | 15G |
| RTX 4090 24GB | 24GB | OK | 3 min 7 s | 23G |
| RTX 5090 32GB | 32GB | OK | 2 min 13 s | 26G |
| RTX PRO 6000 Blackwell 96GB | 96GB | OK | 2 min 4 s | 47G |
1344×768 · 5s: 46 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 (46 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.