Stable Diffusion 3.5 Large Turbo
Stable Diffusion 3.5 Large Turbo is the distilled Large model. Adversarial Diffusion Distillation enables four-step sampling, with local weights and a named Stability API option.
Model design and access
The official Diffusers example sets guidance_scale to zero. The card includes low-bit loading instructions for the diffusion transformer and text encoder. Turbo is a separately trained distilled release, not a speed parameter that can be applied to the ordinary Large checkpoint.Stable Diffusion 3.5 Large Turbo official model documentation
Practical use
Practical guidance. Turbo can support a fast preview loop where a user compares several compositions. Use its own recommended settings before comparing it with Large. Reducing sampling work does not remove the memory cost of loading the pipeline, so a four-step model can still require substantial device memory.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | sd3.5-large-turbo |
| Sampling | Four-step generation.Stable Diffusion 3.5 Large Turbo official model documentation |
| License | Stability AI Community License.Stable Diffusion 3.5 Large Turbo official model documentation |
| Architecture | Distilled MMDiT with three text encoders and query-key normalization.Stable Diffusion 3.5 Large Turbo official model documentation |
Related models
- Stable Diffusion 3.5 Flash
Stability AI · Stable Diffusion
A distilled Stable Diffusion 3.5 Medium image model with four-step generation.
Launch date not confirmedAPI - Stable Diffusion 3.5 Large
Stability AI · Stable Diffusion
Stable Diffusion 3.5 Large is a text-to-image diffusion transformer with downloadable weights and hosted API access.
Launch date not confirmedAPI · Downloadable