👨🏻👨🏼👨🏽👨🏾👨🏿 Adversarial Bros 👨🏿👨🏾👨🏽👨🏼👨🏻

A simple (my ass) Gradio app that generates face photos of delicious imaginary bros using BroGAN, a StyleGAN3 model.

It can generate and save single or bulk generate multiple images with different seeds and other settings.

I'm also planning some extra functionality for a later point in the future (if I don't get bored and abandon this), see ToDos below.

Info:

• I wrote this because I wanted to be able to use BroGAN without firing up A1111 and use the relevant extension for it, and without having to use the terminal and cli. Then I got interested in getting some more experience in Gradio and Python programming in general, so I kinda spent more time with it.
• It's still an exceptionally messy, probably buggy patchwork of mostly other people's code and a bit of my own (= I asked various LLMs to do stuff). But it somehow works?

Caution: It executes pickle code because it has to, and generally I give literally zero guarantees about the whole thing's safety and proper behaviour. Likewise I accept zero responsibility. So please read the code and use at your own risk, or don't.

Requirements:

• An NVIDIA card with CUDA support (can work with CPU if you change stuff in the code).
• A Python environment (I used 3.10 but others may work I guess) with uuuuh... gradio, torch, pickle and pillow and... diffusers? I might've completely failed to take a note of the packages I installed, so I don't remember all of them. Install them if it complains.
• It'll need to compile some CUDA binaries at some point, so you need nvcc and Visual Studio Build Tools.
• The whole thing was developed and run on Windows 11 but it should be relatively easy to run in Linux with minimal adaptations. I think.

Usage:

• Download BroGAN from https://huggingface.co/quartzermz/BroGANv1.0.0 and put it in the model folder.
• Run python bros.py or gradio bros.py.
• Use it. There's instructions inside for the various functions and settings.

ToDos:

• Implement CodeFormer integration so the images can be upscaled from the default 256x256 px.

Credits:

• Huggingface user quartzermz, who did God's work and created BroGAN, a StyleGAN3 model for PyTorch.
• NVlabs at NVIDIA, creators of the StyleGAN3 model that BroGAN is based on, and GitHub user PDillis and their stylegan3 fork, stylegan3-fun whose code I've used as a backend here.
• GitHub user ZerxiesDerxies, author of the very helpful rentry.co guide 'A men-only SD Webui Guide to running BroGAN', with a lot of info about BroGAN. And also author of the relevant A1111 extension, which was the initial inspiration for this project.
• The devs of CodeFormer, which is used to upscale the photos in this project.
• Users at the #men-only channel on the Unstable Diffusion Discord server, by who I was made aware of BroGAN.