Simplified Interface

Fooocus is designed to streamline the user experience by emphasizing prompt input without overwhelming users with numerous settings. It draws influence from the straightforward nature of tools such as StableDiffusionWebUI, ComfyUI, and Midjourney, focusing primarily on prompt usage while automatically managing other configurations. This redesign aims to deliver a user-friendly platform for generating high-quality images.

The user interface mirrors that of StableDiffusionWebUI, utilizing Gradio, but it features only a prompt input field and a Generate button, ensuring a clean and straightforward layout.

Reduced Memory Usage

A significant advantage of Fooocus is its low VRAM requirement for generating SDXL images. According to official sources, the software functions effectively with as little as 6GB of VRAM. This marks a notable advancement, as earlier versions of Stable Diffusion needed much higher VRAM for similar tasks. Consequently, users with limited VRAM can now smoothly create SDXL images.

This powerful yet simplified approach is a boon for many users. Are you excited to give it a shot?

GitHub - lllyasviel/Fooocus: Focusing on Prompts and Generation

Contribute to the development of Fooocus by joining GitHub. You can find it here: github.com.

Windows Version

1. Download the zip file from GitHub:

   https://github.com/lllyasviel/Fooocus/releases/download/release/Fooocus_win64_1-1-10.7z

2. Unzip and execute run.bat to install.

3. During the initial launch, the software will automatically download large files, including sd_xl_base_1.0_0.9vae.safetensors and sd_xl_refiner_1.0_0.9vae.safetensors, each exceeding 6GB in size. This may take some time.

MacOS / Linux Version

Before starting, please refer to the installation guide for StableDiffusionWebUI and Python setup. If Python is already installed, you can skip this part.

Ultimate Guide to Install StableDiffusionWebUI on MacOS

Utilize your Mac to freely generate AI Art: levelup.gitconnected.com

1. Use Git to clone the Fooocus repository:

   git clone https://github.com/lllyasviel/Fooocus.git

2. Create a virtual environment:

   python venv ./venv

3. Activate the virtual environment:

   python activate ./venv/bin/activate

4. Install the necessary Python packages:

   pip install -r requirement_versions.txt

5. Once the setup is complete, launch Fooocus:

   python launch.py

6. During the first run, the software will download large files, similar to the Windows version.

Setting

• Performance: The default Speed option generates images using 30 steps, while Quality uses 60 steps for enhanced results.
• Aspect Ratios: The standard configuration is 1152 x 896.
• Image Number: Control the quantity of generated images (up to 32).
• Random Seed: Default is -1 (random), but you can specify a custom seed.
• Negative Prompt: Specify elements you wish to exclude from the generated images.

Style

This section allows you to select from over 100 distinct image styles, with the default being cinematic-default. To aid in understanding the various styles, I generated reference images using the prompt 1girl at a resolution of 1024 x 1024. This approach provides a clearer comparison of the styles, especially since the official documentation lacks visual references.

Advanced

Here, you can tweak model-related settings and LoRAs.

• SDXL Base Model: Automatically downloads sd_xl_base_1.0_0.9vae.
• SDXL Refiner: Automatically downloads sd_xl_refiner_1.0_0.9vae.
• LoRAs: Choose up to five LoRAs simultaneously, along with their weights. The default sd_xl_offset_example LoRA enhances results compared to not using any LoRA.