And to Aitrepreneur and his great guide regarding TI:
I've used training data from @Athos252 (big thanks for it, I have no clue why other people DO NOT INCLUDE training data, after all, we are here to share the models AND the knowledge, right?
For the first positive result, I decided to not stray away from the suggested parameters from the guide, so everything was set that way.
To be honest I am quite happy with the result and I do not understand why we have on Civitai so many TIs of famous people that BARELY resemble them (yes, the similarity, for the most part, is there, but nobody will be fooled in thinking that it could be an actual photograph of such person).
I, by no means, claim that mine is that good (I still believe that perfect Dreambooth beats perfect TI, however good enough TI can beat good enough Dreambooth, so - in a sense, @scruffynerf is correct that usually, a TI is sufficient) but I think it's good enough :)
Also, shoutout to lovely Chrissie Mayr which I was a fan of before seeing her TI on Civitai. She has a great sense of humor and does not beat around the bush. Her youtube channel is worth checking out!! (https://www.youtube.com/@ChrissieMayr/videos)
They all have their own purposes (600 is the most flexible, while the others can be used for more photorealistic versions with a degree of variety).
As a showcase of the model, I've uploaded my generations that were done purely with AUTOMATIC1111 (no photoshop or any other improvements). However, some were direct generations and some were generations with inpainting on top (easy to find out the inpaintings as I was using Huen @ 150 for them)
发表于 2023-6-7 23:02:41