Fooocus is an open-source AI image generation tool that hits a sweet spot between Midjourney’s ease of use and Stable Diffusion’s local control. Unlike ComfyUI or Automatic1111 (which require understanding nodes or settings), Fooocus asks for only a text prompt and produces excellent results out of the box. Built on Stable Diffusion XL, it runs locally with no cloud costs.

System Requirements

NVIDIA GPU: 4GB VRAM minimum (6GB recommended for SDXL)
AMD GPU: supported via DirectML on Windows; ROCm on Linux
CPU: works but very slow (minutes per image)
RAM: 8GB minimum, 16GB recommended
Storage: 15–20GB for base models

Installation

Windows (One-Click)

Download the latest release from github.com/lllyasviel/Fooocus/releases
Extract the zip
Double-click run.bat (NVIDIA) or run_amd.bat (AMD)
Fooocus downloads required models (~7GB) on first run
Browser opens at http://127.0.0.1:7865

Linux / macOS

git clone https://github.com/lllyasviel/Fooocus
cd Fooocus
pip install -r requirements_versions.txt
python entry_with_update.py

Basic Usage

Fooocus’s interface is intentionally minimal:

Type a prompt in the text box
Click Generate (or press Ctrl+Enter)
Wait 15–60 seconds depending on your GPU
Results appear below

Example prompts that work well

"cinematic photo of a lone wolf in a snowy forest, golden hour lighting, 8k, dramatic"

"portrait of a cyberpunk hacker, neon lights, detailed face, futuristic city background"

"a cozy reading nook with warm library vibes, oil painting style, detailed books"

Fooocus automatically applies a default style (Fooocus V2) that enhances prompt interpretation — you don’t need to add trigger words or model-specific tags.

Style Presets

Click the Styles button to select visual styles that modify the output:

Style	Effect
Fooocus V2	Default enhanced realism
SAI Photographic	Photorealistic portraits and scenes
Anime	Anime-style illustration
Cinematic	Film-like color grading
Digital Art	Digital illustration style
Oil Painting	Traditional art aesthetic

You can select multiple styles and they compound. Example: SAI Photographic + Fooocus Cinematic for dramatic portrait photography.

Advanced Settings

Click Advanced to expand settings:

Image Resolution

Pre-defined options: 1024×1024 (default SDXL square), 1152×896 (landscape), 896×1152 (portrait).

For specific ratios, use custom width/height inputs.

Performance

Quality: best results, slower
Speed: faster, slightly lower detail
Extreme Speed: fastest (uses LCM sampler), noticeably lower quality

Negative Prompt

Describe what you don’t want in the image:

ugly, blurry, low quality, deformed, extra fingers, watermark, text

Sampling

Steps: 30 default; 20 for speed, 50–60 for maximum quality
Sampler: Euler a, DPM++ 2M, DDIM — each produces slightly different aesthetics
CFG Scale: 7 default; higher = more literally follows prompt, lower = more creative

Inpainting (Edit Specific Areas)

Generate an image
Click Inpaint on the result
Paint over the area to change
Write a new prompt for just that area
Generate

This replaces only the painted region. Useful for fixing faces, adding elements, or swapping objects.

Image Upscaling

Fooocus includes an upscaler:

Generate an image at 1024×1024
Click Upscale → select 2× or 4×
The image upscales to 2048×2048 or 4096×4096 with AI-enhanced detail

Custom Models (Checkpoints)

Fooocus works with any SDXL-compatible checkpoint:

Download a model from civitai.com (SDXL models only)
Place in Fooocus/models/checkpoints/
Restart Fooocus
Select the model from the Model dropdown

Popular models:

Juggernaut XL: photorealistic humans
DreamShaper XL: versatile, artistic
RealVisXL: documentary/photographic style
Pony Diffusion XL: illustration, NSFW capable (separate download)

LoRA Models (Style Add-ons)

LoRAs are small add-on files that nudge the model toward a specific style or character:

Download LoRA files (.safetensors) from CivitAI
Place in Fooocus/models/loras/
In Fooocus: scroll to LoRA section → select your LoRA → set weight (0.5–1.0)

Example LoRAs: specific artist styles, character LoRAs, lighting effects, clothing styles.

Fooocus vs. ComfyUI vs. A1111

Feature	Fooocus	ComfyUI	Automatic1111
Ease of use	Excellent	Difficult	Moderate
Workflow customization	Limited	Unlimited	Moderate
SDXL support	Native	Native	Extension
Video generation	No	Yes (AnimateDiff)	Extension
Learning curve	15 min	Days-weeks	Hours

Fooocus is the right starting point for most users. Graduate to ComfyUI when you need workflow customization, ControlNet, or video generation.