Fooocus Image Generation

What is Fooocus?

Fooocus is a free and open-source text-to-image generation tool that leverages AI to create images based on text descriptions, or prompts. It combines the best features of Stable Diffusion (SDXL 1.0) with a user-friendly web interface and requiring minimal manual tweaking.

There is also a free, online version of Fooocus available at fooocus.one with limited functionality.

Key Features:

• Ease of Use: Designed to be more accessible than other image generation tools, automating many settings so users can focus on crafting great prompts.
• Offline Capability: Can run offline on your own computer, providing more control and privacy compared to web-based tools.
• High-Quality Images with Stable Diffusion XL: Optimized to deliver excellent images without the need for extensive parameter adjustments.

System Requirements:

• Operating System: Windows, macOS, or Linux
• Hardware: A modern GPU with at least 4GB of VRAM is recommended for optimal performance.
• Software: Python 3.7 or higher, with the required dependencies installed.
• Disk Space: Approximately 8GB of free disk space is needed for the model and data.

Fine-Tuning models

• Fine-tuning models can be downloaded from the Civitai website.
• Download a pre-trained model and place it in the Fooocus/models/checkpoints directory.
  • Filter on Checkpoints and SDXL 1.0

Example Usage

Simple prompt

• Start with the program by running run_realistic.bat
• Check the Advanced option
• My settings:
  • Preset: realistic
  • Performance: Extreme Speed
  • Aspect Ratio: 1:1
  • Number of Images: 1
  • Seed: random
• Prompt: "A field full of sunflowers, a colorful rooster on the road and a snowy mountain in the back, vivid colors"

Add some weight to the prompt

• If you want to add more focus to specific parts of the prompt, you can use the weight parameter.
• Select a word or phrase in the prompt and click Ctrl + arrow up/down to change the weight.
• Some examples with the image of the previous prompt:
  Note: deselect the Random checkbox to keep (almost) the same image (= the same seed) when re-generating the image.

• Prompt: "A field full of sunflowers, a colorful rooster on the road and a snowy (mountain:1.4) in the back, vivid colors"
  

Using LoRA's

• LoRA (Low-Rank Adaptation) is a technique used in image generation models to fine-tune specific elements like distinct characters or unique styles by adapting only a small subset of parameters, making the process faster and more efficient
• LoRA's for Fooocus can be downloaded from the Civitai website
• One of the available LoRA's is the Potato Chips LoRA, which can be used to generate images of potato chips
• Info: Potato Chips LoRA
  • Download the LoRA inside the Fooocus/models/loras
  • Always look at the Trigger Words to activate the LoRA
  • For the Potato Chips LoRA, the trigger words are p1nkch1ps or made out of potato chips
• Let's generate an image of a potato chip rooster:
  • Prompt: the same as the original prompt, but at the end with the trigger made out of potato chips
    "A field full of sunflowers, a colorful rooster on the road and a snowy mountain in the back, vivid colors, made out of potato chips"
  • Seed: random

• Nothing happened because the LoRA was not activated
  

Styles

• Under the Styles tab, you can select a variety of pre-defined styles to apply to the generated image
• Experiment with different styles to see how they affect the final output

• Styles: Fooocus V2, Fooocus Photograph, Fooocus Negative

Tip

Generate a prompt from an image

• Upload an image to Bing Copilot and ask for a prompt:

Create a stable diffusion prompt for this image. 
Describe the content of this image in one sentence and use keywords, separated by a comma, to describe the style in detail.

•