DEV Community

Cover image for How to Set Up Fooocus Locally: Step-by-Step GPU Guide
Praveen Tech World
Praveen Tech World

Posted on

How to Set Up Fooocus Locally: Step-by-Step GPU Guide

#ai #tutorial

The short answer is: to set up Fooocus locally on Windows, you need an NVIDIA graphics card with at least 4GB of VRAM (8GB recommended). Download the official Fooocus entry package zip file from GitHub, extract it to a directory on your SSD, and double-click the run.bat file. The script will automatically download the necessary FLUX.2 and SDXL model weights and open a web-based local interface at http://127.0.0.1:7865.

Why We Set Up Fooocus on Our Local Workbench

Our team was looking for a way to generate unlimited, private graphics for our blog without constantly buying API credits. We tested several local interfaces (including ComfyUI and Automatic1111), but found their learning curves to be far too steep for daily content creation tasks.

Fooocus solves this. Developed by the creator of ControlNet, it brings a simplified, Midjourney-style prompt interface to your local machine while executing advanced backend optimizations. Under the hood, it uses the high-performance FLUX.2 and SDXL models to deliver stunning photorealism, clean typography, and prompt adherence. Here is how we configured it on our local developer rig.

Step-by-Step Installation Guide

Follow these steps to download, install, and configure Fooocus on your Windows system.

Step 1: Download the Fooocus Entry Package

Do not clone the entire repository unless you plan to develop custom extensions. Instead, download the pre-packaged setup zip:

  1. Navigate to the official Fooocus GitHub Repository.
  2. Scroll down to the Installation section.
  3. Click the direct download link for the Fooocus Entry Package. This downloads a 1.8GB file containing the portable Python environment.

Step 2: Extract the Package

Extracting to the correct directory is critical to avoid permission conflicts.

  1. Create a new folder in the root of your fastest SSD (e.g., C:\Fooocus or D:\LocalAI\Fooocus). Do not install it in the Program Files directory, as Windows will block script execution.
  2. Extract the downloaded zip file contents directly into your newly created folder.

Step 3: Run the Initialization Script

Fooocus uses a portable Python structure, meaning you do not need to install Python globally on your machine.

  1. Open your Fooocus directory in File Explorer.
  2. Double-click the run.bat file.
  3. A command prompt window will open and begin downloading the default model weights:
    • juggernautXL_v8.safetensors (approx 6.6GB)
    • sd_xl_offset_example-lora_1.0.safetensors (approx 700MB)
  4. Depending on your internet speed, this download may take between 10 to 30 minutes. Once complete, your web browser will automatically open to http://127.0.0.1:7865.

Customizing Your Setup for FLUX.2

By default, Fooocus uses SDXL. To get the high-fidelity photorealism of the FLUX.2 model, follow these steps:

  1. Download the FLUX.1-schnell model weights (specifically the FP8 quantized version to fit standard consumer GPUs) from Hugging Face.

  2. Move the .safetensors file into your models folder:

    // config path references
C:\Fooocus\models\checkpoints\flux1-schnell-fp8.safetensors

  3. Open the Fooocus web UI.

  4. Check the Input Option box under the prompt area, select Model, and swap the base checkpoint to your new FLUX model.

When Local Fooocus Works

Local Fooocus is ideal when:

  • You need to generate graphic assets for client work that must remain private and off cloud servers.
  • You are iterating on character consistency using local LoRA modules.
  • You have a modern NVIDIA GPU (like an RTX 3060/4060 or better) that can render images in under 15 seconds.

When Local Fooocus Fails

This local setup is not recommended if:

  • You are working on an AMD GPU or Intel integrated graphics card (support is experimental and extremely slow).
  • You need to generate images from a mobile phone or a tablet while away from your desk.

Decision Summary

  • If you have 8GB+ VRAM -> Run Fooocus with standard FLUX.2 models for best results.
  • If you have 4GB-6GB VRAM -> Edit the launch command to include the --lowvram flag to prevent system out-of-memory errors.
  • If you have No Dedicated GPU -> Skip the local install and use a free cloud provider like Leonardo.ai.

Frequently Asked Questions

Q: Can I run Fooocus on a Mac with Apple Silicon?
A: Yes, Fooocus supports Apple Silicon (M1/M2/M3) chips. You must install homebrew and run the install terminal commands outlined in the GitHub readme instead of using the Windows .bat file.

Q: Where are my generated images saved?
A: Every image you generate is automatically saved in your local directory under C:\Fooocus\outputs\. They are organized by date, making it easy to retrieve your historical assets.

Q: Can I run the web UI on a different port?
A: Yes. If port 7865 is occupied, open webui-user.bat in a text editor and add --port XXXX to the command arguments, replacing XXXX with your desired port.

Related Guides

Top comments (0)