Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/lllyasviel/Fooocus/llms.txt

Use this file to discover all available pages before exploring further.

Fooocus makes high-quality text-to-image generation accessible without requiring extensive prompt engineering or parameter tuning. Its offline GPT-2 based prompt processing engine, combined with a suite of carefully tuned sampling improvements, means that results look great whether your prompt is as short as “house in garden” or as detailed as a thousand words.

​
How It Works

Type your prompt into the main text field and click Generate. Fooocus handles the rest — expanding your prompt, applying your chosen styles, and using its optimized SDXL pipeline to produce sharp, high-quality outputs.
1

Enter your prompt

Type a description of the image you want. Fooocus works well with both short and long prompts — no special syntax required for basic use.
2

(Optional) Add a negative prompt

Expand the Negative Prompt field to describe what you want to avoid in the output (e.g., blurry, watermark, text).
3

Choose an aspect ratio

Select your target resolution from the Aspect Ratios dropdown. These are fixed SDXL-native resolutions.
4

Click Generate

Fooocus runs the full pipeline: prompt expansion, style injection, sampling, and optional refinement.

​
Prompt Expansion (GPT-2 / Fooocus V2)

Fooocus includes an offline GPT-2-based prompt expansion engine exposed as the Fooocus V2 style. When active, it automatically enriches your short prompts with additional descriptive detail before sending them to the model — similar in concept to Midjourney’s hidden pre-processing or LeonardoAI’s Prompt Magic, but running entirely on your local machine with no internet connection required.
Prompt expansion is enabled by default via the Fooocus V2 style entry in Advanced → Style. You can disable it by deselecting that style.
Fooocus also applies Negative ADM guidance: because the highest resolution level of SDXL Base lacks cross-attention layers, CFG contrast at that level is weak, leading to plastic or overly smooth results. Fooocus compensates by modifying the ADM conditioning on both the positive and negative sides, ensuring natural-looking outputs.

​
Prompt Weighting

Fooocus uses A1111’s reweighting algorithm, so prompts copied directly from Civitai work as expected. 
a photo of a (happy:1.5) dog in a (sunny:0.8) park
To use an embedding, reference it with the embedding: prefix: 
(embedding:file_name:1.1), portrait of a woman
A1111 reweighting produces better results than ComfyUI reweighting when using prompts sourced from Civitai, because those prompts are typically written in A1111 syntax.

​
Multi-Line Prompts

Each line of text in the prompt box is treated as a separate sub-prompt. Use multi-line entries to blend concepts or describe a scene from multiple angles: 
a misty forest at dawn
golden light filtering through ancient oak trees
a lone deer standing by a stream

​
Wildcards

Fooocus supports wildcard substitution using __wildcard_name__ syntax. A random entry from the corresponding file in the wildcards/ folder replaces the token at generation time (randomness is seeded): 
a __color__ sports car on a mountain road
You can enable Read wildcards in order in Developer Debug Mode to process wildcard files sequentially rather than randomly.

​
Aspect Ratios

All supported aspect ratios correspond to fixed SDXL-native resolutions. Images generated at these exact sizes look slightly better than arbitrary resolutions because SDXL’s positional encoding was trained on these specific values.

Portrait (tall)

704×1408, 704×1344, 768×1344, 768×1280, 832×1216, 832×1152, 896×1152, 896×1088, 960×1088, 960×1024

Square

1024×1024

Landscape (wide)

1024×960, 1088×960, 1088×896, 1152×896, 1152×832, 1216×832, 1280×768, 1344×768, 1344×704, 1408×704, 1472×704, 1536×640, 1600×640, 1664×576, 1728×576
Using one of these fixed resolutions produces slightly better results than arbitrary sizes. The Fooocus UI hard-codes these values specifically to take advantage of SDXL’s positional encoding.

​
Image Count

To generate multiple images in a single run, go to Advanced → Image Number and set the desired count. All images share the same prompt and settings but use incrementing seeds (unless seed increment is disabled).

​
Output Format

Fooocus supports three output formats, configurable in Advanced → Output Format:
FormatNotes
pngLossless, largest file size. Best for archiving or further editing.
jpegLossy compression, smaller file size.
webpModern format with good compression and quality.

​
Seed Control

Each generation uses a numeric seed that determines the random noise pattern. By default, a new random seed is picked for each image.
  • Fix the seed to reproduce the exact same image given identical settings.
  • Increment seed (default) generates a unique image for each output in a batch.
  • Disable seed increment to use the same seed for all images in a batch, useful when comparing style or prompt changes.

Build docs developers (and LLMs) love

Get started for freeTalk to us