Prompt Writing Guide¶
Tips for writing effective image generation prompts, tailored to each provider.
For per-model strengths, weaknesses, and prompt examples, see the Model Catalog.
OpenAI (natural language)¶
OpenAI models work best with natural language descriptions. Be specific about what you want.
General tips¶
- Describe the scene in complete sentences
- Include style, lighting, and composition details
- Specify camera angle or perspective when relevant
- Mention the medium (photograph, oil painting, digital art, etc.)
Examples¶
Product photography:
A professional product photo of white sneakers on a clean white background,
studio lighting, sharp focus, commercial photography style, high resolution
Logo design:
A minimalist logo for "Acme Corp" with clean sans-serif typography,
blue and white color scheme, modern design, vector art style
Illustration:
A whimsical children's book illustration of a fox reading a book under
a large oak tree, watercolor style, warm autumn colors, soft lighting
Negative prompts¶
When using OpenAI, negative prompts are appended as an "Avoid:" clause. Keep them concise:
Avoid: blurry, low quality, distorted text, watermark
SD WebUI / Stable Diffusion (CLIP tag format)¶
Stable Diffusion models respond best to comma-separated tags ordered by importance.
Tag structure¶
subject, medium, style, lighting, camera, quality tags
Examples¶
Portrait:
1girl, long hair, blue eyes, school uniform, standing, cherry blossoms,
soft lighting, detailed face, masterpiece, best quality
Landscape:
mountain landscape, sunset, dramatic clouds, lake reflection,
cinematic lighting, wide angle, 8k, highly detailed
Product shot:
white sneakers, product photography, studio lighting, white background,
sharp focus, commercial photography, high resolution
Anime:
1boy, silver hair, red eyes, dark coat, standing on rooftop, city skyline,
night, wind, detailed, anime style, masterpiece
Weighted tokens (emphasis)¶
Use parentheses to increase or decrease emphasis on specific tokens:
```
Increase emphasis (1.1x per level of parentheses)¶
(detailed face) # 1.1x emphasis ((detailed face)) # 1.21x emphasis (detailed face:1.5) # 1.5x emphasis (explicit weight)
Decrease emphasis¶
[blurry background] # 0.9x emphasis (blurry background:0.5) # 0.5x emphasis (explicit weight) ```
Weights are relative to the baseline of 1.0. Values above 1.0 strengthen a concept; values below 1.0 weaken it. Extreme weights (above 1.8 or below 0.3) often produce artifacts.
Quality tags¶
Add these to improve output quality:
| Tags | Effect |
|---|---|
masterpiece, best quality |
General quality boost |
highly detailed, sharp focus |
Detail enhancement |
8k, ultra high res |
Resolution boost (use sparingly) |
professional, award winning |
Style refinement |
Negative prompts¶
Always include a negative prompt with SD WebUI to avoid common artifacts.
General-purpose:
lowres, bad anatomy, bad hands, text, error, missing fingers,
extra digit, fewer digits, cropped, worst quality, low quality,
normal quality, jpeg artifacts, signature, watermark, blurry
Add for photorealism:
cartoon, anime, illustration, painting, drawing, art, sketch
Add for anime/illustration:
photo, realistic, 3d render
BREAK syntax¶
Use BREAK to separate concepts into different CLIP token chunks. This is useful when your prompt is long or has distinct elements:
1girl, detailed face, blue eyes BREAK
forest background, sunlight through trees BREAK
masterpiece, best quality, sharp focus
CLIP token limits¶
- SD 1.5: 77 tokens per CLIP chunk
- SDXL: 77 tokens per chunk, but uses two CLIP encoders (ViT-L + ViT-bigG)
Front-load the most important tags -- tokens beyond the first 77-token chunk have diminishing influence.
Provider-agnostic tips¶
Aspect ratios¶
All providers support these aspect ratios:
| Ratio | Best for |
|---|---|
1:1 |
Avatars, icons, social media posts |
16:9 |
Landscapes, banners, desktop wallpapers |
9:16 |
Phone wallpapers, stories, portraits |
3:2 |
Photography standard, prints |
2:3 |
Book covers, posters |
Common use cases¶
| Use case | Recommended provider | Prompt approach |
|---|---|---|
| Logo with text | OpenAI | Natural language, specify text content exactly |
| Photo-realistic portrait | SD WebUI | Tag format, include quality and lighting tags |
| Anime character | SD WebUI | Tag format, include character details |
| Quick placeholder | Placeholder | Any prompt (generates solid color) |
| Product photo | SD WebUI or OpenAI | Describe studio setup and product details |
| Landscape art | SD WebUI or OpenAI | Include composition, lighting, and style |