Stable Cascade
Stable Cascade is a research image model based on Würstchen. It separates generation into three stages and uses a highly compressed latent representation for the text-conditioned stage.
Research release
Model design and access
The architecture can represent a 1024 × 1024 image in a 24 × 24 latent grid. Stage C generates the compressed representation; the other stages reconstruct the image. The card recommends the larger Stage C and B variants for its strongest results.Stable Cascade official model documentation
Practical use
Practical guidance. Cascade is relevant when studying compression and staged image generation. Its stages should be evaluated together: replacing one component changes the complete pipeline. Check the research license before adapting it for a commercial project, and avoid assuming that its terms match those of current Stable Diffusion releases.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | stable-cascade-prior |
| Stage C | 1B and 3.6B parameter variants.Stable Cascade official model documentation |
| Stage B | 700M and 1.5B parameter variants.Stable Cascade official model documentation |
| Stage A | 20M parameters.Stable Cascade official model documentation |
| License | Stable Cascade Non-Commercial Community License.Stable Cascade official model documentation |
Related models
- Stable Diffusion XL 1.0
Stability AI · Stable Diffusion
SDXL 1.0 is a latent-diffusion image model that can run as a standalone base or with a separate refiner.
Launch date not confirmedAPI · Downloadable