How Stable Diffusion prompts work
- Structure: quality tags → subject → style and medium → details → camera and lighting. Order matters — earlier tokens get slightly more weight.
- Weight with (token:1.2) to emphasise; [token] or (token:0.8) to de-emphasise.
- A negative prompt removes what you don’t want: worst quality, low quality, blurry, deformed.
- Match the style to your checkpoint: SD 1.5 / SDXL use descriptive tags, Pony uses score tags, Illustrious uses danbooru tags, Flux prefers natural language.
Example Stable Diffusion prompt
masterpiece, best quality, a fox spirit in a misty bamboo forest, golden hour, (detailed fur:1.2), cinematic, 35mm
negative: worst quality, low quality, blurry, deformed, extra limbs
Build Stable Diffusion prompts faster with Prompt Builder
- A color-coded snippet library — click quality, subject, style and lighting blocks into place.
- Per-checkpoint template packs (SD 1.5, SDXL, Pony, Illustrious, Flux).
- One-key attention weighting and a live CLIP token counter.
- Drag in any SD render (A1111 / ComfyUI) to recover its exact prompt.
Frequently asked questions
What makes a good Stable Diffusion prompt?▾
Clear structure (quality, subject, style, detail), moderate attention weighting, and a focused negative prompt — with the style matched to your checkpoint.
How do I make part of a prompt stronger?▾
Wrap the key tokens in attention weights, e.g. (detailed eyes:1.3), and remove competing concepts via the negative prompt.
Do all Stable Diffusion models use the same prompts?▾
The syntax is shared, but conventions differ: SD 1.5 / SDXL use descriptive tags, Pony uses score tags, Illustrious uses danbooru tags, and Flux prefers natural-language sentences.
Keep exploring
- Prompt weighting & LoRA syntax guide
- Prompt glossary
- Other models: SDXL · Flux · Midjourney · Pony Diffusion · Illustrious · ComfyUI · Automatic1111