The most well-known tool for AI-generated images isMidjourneyandStable DiffusionHowever, their advantages and disadvantages are quite distinct. Midjourney is expensive at $30 a month, while Stable Diffusion requires more configuration and is more complicated to operate.
Today we share an open source project that works very well and greatly reduces the threshold for AI mapping. It is:Fooocus
What is Fooocus
Fooocus It is a Gradio-based image generation software that rethinks the design of Stable Diffusion and Midjourney, aiming to provide high-quality text-to-image conversion while simplifying the user's operation process.
Fooocus learned about the offline, open source, and free features of Stable Diffusion, as well as the advantage of Midjourney that it does not require users to adjust parameters.
Fooocus integrates and automates many internal optimizations and quality improvements, allowing users to forget all the complex technical parameters and focus on the interaction between prompts and images to explore new ways of thinking and expand the imagination of the human species. The most important thing is that it is free and does not require configuration. It only requires 4G video memory, which is much lower than SD requirements!
Below is a screenshot of a system with 16GB of system RAM.and6GB VRAMThe speed on this machine is about 1.35 seconds per iteration. Pretty impressive - laptops with 3060s are generally very affordable these days.
Compare Midjourney
Although Fooocus simplifies the operation, it does not reduce the functionality. The following is a detailed comparison of Fooocus's functionality with Midjourney.
|
Midjourney |
Fooocus |
|
High-quality text to image without extensive on-the-fly engineering or parameter tweaking. |
High-quality text-to-image without a lot of on-the-fly engineering or parameter tuning. (Fooocus has an offline prompt processing engine based on GPT-2 and a lot of sampling improvements, so whether your prompt is as short as "house in the garden" or as long as 1000 words, the results are always beautiful) |
|
V1 V2 V3 V4 |
Input image -> Upscale or Variation -> Variation (Subtle) / Variation (Strong) |
|
U1 U2 U3 U4 |
Input image -> Enlarge or change -> Enlarge (1.5x) / Enlarge (2x) |
|
Fix/Up/Down/Left/Right (Pan) |
Input image -> Inpaint or Outpaint -> Inpaint / Up / Down / Left / Right (Fooocus uses its own inpainting algorithm and inpainting model, so the result is more satisfactory than all other software which use standard SDXL inpainting method/model) |
|
Image Tips |
Input image -> Image hints (Fooocus uses its own image hinting algorithm, so the result quality and hint understanding are more satisfactory than all other software using standard SDXL methods like Standard IP Adapter or Revision) |
|
--style |
Advanced -> Style |
|
--stylize |
Advanced -> Advanced -> Guidance |
|
—niji |
https://github.com/lllyasviel/Fooocus/discussions/679Fooocus supports SDXL models on Civitai (if you don't know, you can google "Civitai") |
|
- quality |
Advanced->Quality |
|
- repeat |
Advanced->Image Number |
|
Multiple prompts (::) |
Just use a multi-line prompt |
|
Prompt Weight |
You can use "I am (happy:1.5)". Fooocus uses A1111's reweighting algorithm, so if the user copies the prompt directly from Civitai, the result will be better than ComfyUI. (Because if the prompt is written with ComfyUI's reweighting, users are less likely to copy the prompt text, because they prefer to drag the file) To use embedding, you can use "(embedding:file_name:1.1)" |
|
- no |
Advanced -> Negative Prompt |
|
--ar |
Advanced -> Aspect Ratio |
|
Insights |
Input image -> Image prompt -> Advanced -> FaceSwap |
|
describe |
Input Image -> Description |
Install Fooocus
Windows
- First, you can download Fooocus directly from the download link on GitHub: https://github.com/lllyasviel/Fooocus/releases/
- After downloading the file, unzip it and run "run.bat".
- The first time you run it, the model files will be automatically downloaded and you will need to wait for a while. (If you already have model files, you can copy them to the following path, and the startup will be faster: .\Fooocus\models\checkpoints\)
Linux
- Clone the Fooocus GitHub repository, change into the Fooocus directory and create a new conda environment:
gitclone cdFooocus condaenv create -f environment.yaml- Activate the newly created conda environment and install the required Python packages:
condaactivate fooocus pipinstall -r requirements_versions.txt- Run the following command to start Fooocus:
pythonentry_with_update.pyThe following are the minimum resource requirements for installation. You can check them against your own platform.
Using Fooocus
After the download is complete, the drawing interface will automatically open. The default mode is simple mode, with only an input box, a generate image area and a generate button. Its basic interface is very simple. Checking "Advanced" will pop up an advanced settings window, where you can adjust the aspect ratio, style, number of images, seed value, reverse prompt word, model, lora weight ratio, image sharpness, etc.
Simply enter the prompt word in the text box and click the "Generate" button to generate the image. Check "Advanced" for more advanced settings. The generated image will be automatically saved in the outputs folder in the root directory.
Many contents in the models folder in the root directory of Fooocus are similar to those in SD WebUI, and are used to store various models such as checkpoints, lora, Vae, etc. If you have downloaded other stylized large models or lora models trained based on SDXL1.0, you can also put them in the corresponding folders and use them in Fooocus.
Let's look at the comparison examples with DALL-E3 and Midjourney:
Prompt: beautiful landscape with a lake in the foreground, with a dock, and mountains in the background, in the early morning with sunrise and mist on the lake.
Fooocus
DALL-E 3
Midjourney
Project Features
- Offline and open source: Fooocus runs completely offline, and users can use it without an internet connection, while also ensuring the open source nature of the software.
- Simplified installation and use: Fooocus simplifies the installation process, from downloading to generating the first image, the number of mouse clicks is strictly limited to 3. The minimum GPU memory requirement is 4GB (Nvidia).
- Automation Optimization: Fooocus automates many internal optimizations, including sampling improvements, to ensure that the results are always beautiful, no matter how short or detailed the user's prompt is.
- Easy to use: Fooocus has designed an easy-to-use interface and prompt system, so users do not need to make tedious parameter adjustments, but only need to pay attention to the prompts and the generated images.
Summarize
Overall, Fooocus is a powerfulImage generation toolsThe quality of the images generated is very high, with excellent details, which is not inferior to Midjourney at all, and it simplifies the operation, providing users with a platform that is both easy to use and powerful.
Whether you are an artist, designer or ordinary user, you can use Fooocus to create amazing visual works.