Self-Host Fooocus: Replace Midjourney and DALL-E With Free, Private AI Image Generation

Midjourney costs $10-120 per month. DALL-E charges per image. Both send your prompts to servers you don’t control, where they might be used to train future models or kept indefinitely.

Fooocus offers a different approach: a free, open-source image generator that runs entirely on your own hardware. No subscriptions, no prompt logging, no usage limits. Once installed, it’s yours.

What Fooocus Actually Is

Fooocus is a Stable Diffusion interface designed by lllyasviel (creator of ControlNet) with one goal: make image generation as simple as Midjourney without the complexity of other local tools. You type a prompt, you get an image. No nodes to connect, no settings to tweak unless you want to.

It’s currently in long-term support mode at version 2.5.5, meaning active development has slowed but it remains stable and functional. Under the hood it runs SDXL models, delivering quality comparable to commercial services.

What You Need

Minimum hardware:

4GB VRAM (NVIDIA GPU recommended)
8GB system RAM
24GB free disk space

Recommended:

8GB+ VRAM for faster generation
16GB+ system RAM
SSD for model loading

Operating systems:

Windows 10/11 (easiest)
Linux with NVIDIA GPU
macOS with Apple Silicon (experimental, slower)

If you don’t have a dedicated GPU, Fooocus can run on CPU only, but generation will be roughly 17x slower and you’ll need at least 32GB of RAM.

Installing on Windows

The Windows installation is straightforward:

Download Fooocus from the GitHub releases page. Click the download link in the latest release.
Extract the archive using 7-Zip or Windows built-in extractor. Right-click the downloaded file and select “Extract Here.”
Run the batch file. You’ll find three options:
- run.bat - General-purpose model
- run_realistic.bat - Optimized for photorealistic images
- run_anime.bat - Tuned for anime/illustration style
Wait for first-time setup. On first run, Fooocus downloads the JuggernautXL model (about 6.6GB). This only happens once.
Open the interface. Your browser will automatically open to http://localhost:7865. If not, navigate there manually.

That’s it. Type a prompt and click Generate.

Installing on Linux

For Ubuntu or other Debian-based distributions:

# Install prerequisites
sudo apt install python3 python3-venv git

# Clone the repository
git clone https://github.com/lllyasviel/Fooocus.git
cd Fooocus

# Create virtual environment
python3 -m venv fooocus_env
source fooocus_env/bin/activate

# Install dependencies
pip install -r requirements_versions.txt

# Run Fooocus
python entry_with_update.py

To access Fooocus from other devices on your network:

python entry_with_update.py --listen

Installing on Mac (Apple Silicon)

Mac support is experimental and slower than dedicated GPUs, but it works:

Install Miniconda for macOS ARM64.
Set up PyTorch with MPS support:

conda create -n fooocus python=3.10
conda activate fooocus
pip install torch torchvision torchaudio

Clone and install Fooocus:

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

Note: 8GB RAM Macs will be slow due to memory swapping. 16GB or more is recommended.

Generating Your First Image

Once Fooocus is running:

Enter a prompt in the text box. Something like: “A cozy coffee shop in autumn, warm lighting, rain on the windows”
Click Generate. Default settings produce good results without tweaking.
Wait 10-60 seconds depending on your hardware. The image appears when done.

Want to adjust style? Click Advanced and look at the Style dropdown. Fooocus includes presets like “Fooocus V2,” “MRE Surreal Painting,” “MRE Dark Dream,” and many more.

Features That Match Midjourney

Image prompts. Upload a reference image and Fooocus will incorporate its style or content into your generation. Drag an image into the interface or use the Image Prompt section under Advanced.

Inpainting. Select a region of an existing image and regenerate just that area. Fooocus uses its own inpainting algorithm that works with any SDXL model—something other tools struggle with.

Outpainting. Extend images beyond their original borders in any direction: left, right, top, or bottom. The AI generates seamless new content that matches the existing image.

Face and hand improvement. The “Improve Detail” mode specifically targets common weak points in AI generation. Select a face or hands in your image and let Fooocus regenerate them with more accuracy.

Negative prompts. Tell Fooocus what to avoid. Common additions: “blurry, low quality, distorted, extra limbs.”

Privacy Comparison

Feature	Midjourney	DALL-E	Fooocus
Prompts logged	Yes	Yes	No
Images stored on servers	Yes	Yes	No
Data used for training	Possibly	Possibly	No
Works offline	No	No	Yes
Monthly cost	$10-120	Pay per use	Free

With Fooocus, your prompts never leave your computer. The models run locally. There’s no server to send data to.

Realistic Expectations

Fooocus isn’t perfect:

Speed varies wildly. A high-end GPU generates images in 10-15 seconds. A 4GB card might take a minute or more. CPU-only could be several minutes per image.
VRAM limitations. Lower VRAM means lower resolution outputs and potential crashes with complex prompts. The --lowvram flag helps but slows things down.
Model size. SDXL models are 6-7GB each. Want multiple style options? Plan for 20-30GB of model storage.
No cloud backup. Your images live on your computer. If you want them accessible elsewhere, you’ll need to set that up yourself.

Adding More Models

Fooocus supports custom checkpoints. Download SDXL-compatible models from Civitai or Hugging Face and place them in:

Fooocus/models/checkpoints/

Restart Fooocus and they’ll appear in the model dropdown under Advanced settings.

Popular alternatives to the default JuggernautXL:

DreamShaper XL - Versatile, good at many styles
RealVisXL - Photorealistic focus
Animagine XL - Anime and illustration

When to Use Something Else

Fooocus prioritizes simplicity. If you need:

FLUX.1 models - Use ComfyUI or WebUI Forge instead
Complex workflows - ComfyUI’s node-based system offers more control
Video generation - Look at separate tools like AnimateDiff

For most image generation needs, Fooocus hits the sweet spot of quality versus complexity.

The Bottom Line

Fooocus delivers Midjourney-competitive results without the subscription, the privacy concerns, or the Discord requirement. Setup takes about 10 minutes on Windows, and first-time model downloads take another 10-15 depending on your connection.

If you’ve been paying for AI image generation and wondering if there’s a better way, this is it. Your computer, your images, your control.