🎨
SDXL 1.0
by Stability AI
Versatile open-source model with a massive ecosystem of fine-tunes, LoRAs, and community tools.
Overview
Stable Diffusion XL (SDXL) 1.0 by Stability AI is the most popular open-source image model. Fully customizable with LoRAs, ControlNet, and local deployment. The foundation of the open-source AI art ecosystem.
Key Strengths
- Open source: Run locally, fine-tune, modify — no API costs
- LoRA ecosystem: Thousands of community-trained style adapters
- ControlNet: Precise control over composition, pose, depth
- Negative prompts: Explicitly exclude unwanted elements
- Customization: Fine-tune on your own images/style
- Speed: Fast generation on consumer GPUs (RTX 3060+)
Prompting Tips
Prompt Structure
SDXL uses a weighted keyword system, not natural language:
masterpiece, best quality, highly detailed, [subject], [style], [lighting], [composition], [technical]
Example:
masterpiece, best quality, highly detailed, portrait of a young woman with flowing red hair, oil painting style, dramatic Rembrandt lighting, soft bokeh background, 8k uhd
Negative Prompts (Essential!)
Negative: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, deformed
Prompt Weighting
Emphasize or de-emphasize terms:
(important keyword:1.3) — 30% more emphasis
(less important:0.7) — 30% less emphasis
((very important)) — double emphasis
Key Parameters
| Parameter | Recommended | Effect |
|---|---|---|
| Steps | 25-35 | More = more detail, diminishing returns after 35 |
| CFG Scale | 6-8 | Lower = creative, Higher = strict prompt adherence |
| Sampler | DPM++ 2M Karras | Best quality/speed balance |
| Size | 1024x1024 | SDXL's native resolution |
Best Use Cases
| Use Case | Quality |
|---|---|
| Artistic/stylized images | Excellent |
| Custom-trained styles (LoRA) | Excellent |
| Controlled generation (ControlNet) | Excellent |
| Batch generation (volume) | Excellent |
| Photorealistic portraits | Good |
| Text in images | Poor |
LoRA & ControlNet
- LoRA: Style adapters trained on specific aesthetics (anime, pixel art, specific artist styles)
- ControlNet: Use reference images for pose, depth, edges, or composition
- IP-Adapter: Transfer style from reference images
Running Locally
- Minimum: RTX 3060 (12GB VRAM), 16GB RAM
- Recommended: RTX 4070+ (12GB+ VRAM)
- Software: ComfyUI (node-based) or Automatic1111 (web UI)
- Generation time: ~5-10 seconds at 1024x1024 on RTX 4070
Community Tips
Quick tips from the community about what works with SDXL 1.0 right now.
Sign in to share a tip.
No tips yet. Add a tip for this model.
Prompts for SDXL 1.0 (10)
Terrazzo Pattern — Modern Interior Tile+52Seamless Wood Texture — PBR Material+20Rusty Metal Texture — Industrial Decay+17Renaissance Oil Painting — Classical Portrait+65Weathered Stone Texture — Ancient Wall+24Synthwave Retro Portrait — Neon Grid Aesthetic+68Seamless Marble Texture — Carrara White+47Cinematic Portrait — Dramatic Lighting
