Using Dreambooth, a model can be trained using faces. After an hour (on a 4090), the trained data is added to an existing Stable Diffusion Model. I decided to give it a go, and the results were surprising to say the least.
First, I gathered 20 images from my Facebook Profiles pictures and added them in Dreambooth. Once it was ready, I could now use the model, including “markflorquin” in the prompts. Only 1 in 20 images turned out to be useable, so a big part of the process is triage.
I used keywoards like Star Wars, Skywalker, Cowboy, Sheriff, Cloak, Cyberpunk, Costume, Skirt and Tie, Bodybuilder, Skeleton, Hippie, Mad Hatter, Wizard, Photographer, and Firefighter.
It’s great to see AI bring it all together, but again it’s a hit and miss process. I have a lot of new potential Facebook profile pictures, but getting ripped as much as the Bodybuilder shots, that will take some time to match.
Thanks for watching, See you on the next project!