I have been using Auto1111's webui for getting started with diffusion models and LoRas. I want to create my own text-to-image website where users can play around with different models and LoRas that I upload on the server, like a playground of sort.
Hence, I plan on building my own custom frontend ccomponents in React for the same instead of using gradio elements(since it does not give much UI customisation control), which is the case for Auto1111's repo. I have been trying to disintegrate the backend for the txt2img generation and it is very tricky sincy all the files are contained inside a single modules directory which contains the backend and frontend logic.
I tried understanding the entire pipeline of Auto1111's application and figured that the main files I require for the txt2img generation are:
- modelloader.py -> Basically takes the .ckpt and .safetensor files from the models>StableDiffusion directory and loads inside the gradio list component
- processing.py
- txt2img.py
- sd_models and a few more.....
Is there a simpler and less complex way to create an API for the difffusion models? Otherwise what approach should I use for making my custom frontend with Auto1111's complete backend logic for the text-to-image generation models.
