Wuerstchen v2
A unique 3-stage cascade architecture with 42x latent compression. CLIP-G text encoder feeds into Prior → Decoder → VQ-GAN stages.
- Developer: Wuerstchen Team
- License: MIT
- HuggingFace: warp-ai/wuerstchen
Variants
| Model | Steps | Size | Notes |
|---|---|---|---|
wuerstchen-v2:fp16 | 60 | 5.6 GB | Full cascade pipeline |
Defaults
- Resolution: 1024x1024
- Guidance: 4.0
- Steps: 60
Recommended Dimensions
| Width | Height | Aspect Ratio |
|---|---|---|
| 1024 | 1024 | 1:1 (native) |
Using non-recommended dimensions will trigger a warning. All values must be multiples of 16.
Notes
Wuerstchen includes a default negative prompt. The 42x latent compression means the diffusion process operates in a very compact space, which allows for efficient generation despite the multi-stage pipeline.
Negative prompts are supported and effective with this model.