by Stability AI
Next-gen architecture with improved coherence and detail over SDXL.
Stable Diffusion 3.5 by Stability AI is the latest in the SD family, featuring the MMDiT (Multimodal Diffusion Transformer) architecture. Better prompt understanding and image quality than SDXL, while remaining open-source.
Unlike SDXL's keyword-heavy approach, SD3.5 understands natural descriptions:
SDXL style: "masterpiece, best quality, highly detailed, portrait, woman, red hair, oil painting"
SD3.5 style: "A portrait of a woman with flowing red hair, painted in rich oil paint style with visible brushstrokes and dramatic lighting"
Negative: blurry, low quality, distorted, deformed, ugly, watermark, text overlay
| Parameter | SD3.5 Large | SD3.5 Turbo |
|---|---|---|
| Steps | 28-40 | 4-8 |
| CFG Scale | 4-7 | 1-2 |
| Sampler | Euler | Euler |
| Size | 1024x1024 | 1024x1024 |
| Variant | Size | Speed | Quality | Best For |
|---|---|---|---|---|
| Large | 8B params | Slow | Best | Final renders, portfolio work |
| Medium | 2.5B | Moderate | Very Good | Daily use, iteration |
| Turbo | 2.5B | Fast (4 steps!) | Good | Rapid prototyping, previews |
| Use Case | Quality |
|---|---|
| Natural language prompts | Very Good |
| Character faces & anatomy | Good (improved from SDXL) |
| Text in images | Good (improved from SDXL) |
| Fine-tuning / LoRA | Developing ecosystem |
| Production volume | Excellent (especially Turbo) |
Quick tips from the community about what works with Stable Diffusion 3.5 right now.
Sign in to share a tip.
No tips yet. Add a tip for this model.