Low VRAM Any mode 4 steps
Low-VRAM recipe: GGUF Q3 + 4-step (8–12GB cards)
The most realistic combo for 8GB-class cards: Q3 model (8.9GB) + Q2 encoder + 4-step distill; 540p short clips render. Full download stays under ~25GB.
How to run it
- 1 Install ComfyUI-GGUF and drop the .gguf files into diffusion_models / text_encoders
- 2 Replace the checkpoint loader with Unet Loader (GGUF)
- 3 Launch with --fast-disk --disable-mmap; start at 736×416
Sage gains are inconsistent on GGUF, and some memory-optimization nodes conflict with it (see 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 | OK | 6 min 5 s | 7G |
| RTX 3060 12GB | 12GB | OK | 5 min 46 s | 11G |
| RTX 4070 12GB | 12GB | OK | 3 min 20 s | 11G |
| RTX 4070 Ti Super 16GB | 16GB | OK | 2 min 26 s | 14G |
| RTX 4080 Super 16GB | 16GB | OK | 2 min 10 s | 14G |
| RTX 4090 24GB | 24GB | OK | 1 min 33 s | 14G |
| RTX 5090 32GB | 32GB | OK | 1 min 7 s | 14G |
| RTX PRO 6000 Blackwell 96GB | 96GB | OK | 1 min 3 s | 32G |
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.