santa hat
deerdeer nosedeer glow
Sign In

Stable Diffusion: text-to-person | How to Generate High Quality Real People in Stable Diffusion

Stable Diffusion: text-to-person | How to Generate High Quality Real People in Stable Diffusion

Hello! I made a huge guide on training and using LoRA models of real people. It covers almost everything you should know. The main thing it contains is visual comparison of all the LoRA training parameters so you can see how changing every parameter affects the result. I had to make more than 130 LoRAs and more than 400 X/Y/Z Plots to compare everything! It also contains a lot of tips and tricks on how to automate image generation the most. I will also show you how you can train one or two LoRAs and use them on dozens of checkpoints. I'm pretty sure everyone will find something previously unknown and useful!

I tried to monetize my knowledge and skills, but it didn’t work out, so I decided to create the most complete guide possible :)

It is so big it wouldn't fit on a single page so I made a Github Wiki for it.

I used Billie Eilish as a character most of the guide. Here's one of the final comparisons. The first two images are the models we started with, the second two are intermediate models, the last two are the final models. But both intermediate models and final models turned out to be pretty good :)


full size

full size

full size

full size


And some of the most interesting results!


Feel free to suggest any improvements! Ask your questions if you have some. You can write here in the comments or create a GitHub Issue.


Thanks ♥

25

Comments