Flux AI model, install it on ComfyUI

Flux is a series of image diffusion models developed by Black Forest Labs. It is undoubtedly the best open source image diffusion model deployed locally and running on a PC. Its rendering effect and stability surpass SDXL and SD3. I wonder if the just-released SD3.5 can compete? The Flux.1 AI model has very good compliance with prompt words, can accurately generate a person image with the correct finger, and is very good at generating text information in the image The following will start a tutorial on how to install the Flux model in your local ComfyUI. Flux is currently not supported in webUI Automatic1111. If you haven't installed ComfyUI yet or Forge, the first step is to deploy ComfyUI locally. If your graphics card has low Gpu memory, it is recommended to use Flux in Forge. 1. Deploy ComfyUI locally It is recommended to install Autumn Leaf's integrated package. The download address is: pan.baidu.com/s/1W-pzVF_o_grH5ACJgVDbZQ About 4.4 GB. After downloading, decompress it and find the Emi launcher. Double-click to open it, you will see the interface. If an error is reported, please check whether the downloaded package is complete. I just interrupted it once and the downloaded package is decompressed before it can be used. If there are no prompts and interfaces pop up, please check the anti-virus software. Sometimes you will accidentally kill the startup shortcut. You can start normally after temporarily quitting anti-virus or joining the whitelist. You can see the interface shown in the figure. Congratulations, you have successfully installed it. Click one click to launch, and if you can see the browser window below, it means that the local ComfyUI environment has been configured. There are also some other setting items, which you can set as you gradually use and become familiar with, so I won't go into detail here. 2. How to share model files with WebUI? ① Open the comfyUI-aki-v1.4 folder, find this file, and delete the suffix.example. ② Use Notepad to open the file just now, and change it to the folder path of the webUI integration package. For example, mine is C:Program Filessdsd-webui-aki-v4.2. Don't forget the space in front of the path!!! ③ After restarting the launcher, click Models in the upper right corner of the browser. If you can see the model preview, it means that you can use the model files in the WebUI. 3. Install Flux model files Download the required model file attachments. ① Download the Flux model file. Flux has many versions of model files, and there are many introduction articles that will not be detailed here. It is recommended to use the version of flux1-dev-fp8, which is about 17 GB in size. Download address: https://huggingface.co/Comfy-Org/flux1-dev/tree/main ② Download the CLIP model file, and after downloading, place the two CLIP models in the ComfyUI-aki-v1.4modelsClip directory. Download address: https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main ③ Download the VAE model file and place it in the ComfyUI-aki-v1.4modelsvac directory Download address: https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main 4. Running the workflow Download and install the model file, you can import the workflow to start generating diagrams. There are two ways to import workflow, one is to load json format files, the other is images generated by workflow, click load or drag. As shown in the figure, I directly drag the generated image into the workflow. Most of your canvas will have such missing nodes. This is because some nodes are not caused locally. You can solve the missing node problem by installing missing nodes in the manager. Click install after the displayed missing node, and then refresh the workflow interface. The last step is to enter text in the CLIP text prompt to generate an image. In 1. Model Loading, fill in the three model files you just downloaded respectively. In 3. Parameter Settings, you can adjust the size, size and number of images generated, and then click on the right to add prompt queue. The generated real character's fingers were correct, and there was no obvious collapse. In the next article, I will learn with you about the workflow construction of graphical drawings and partial redrawing.