DAINAPP
patreon
SD Dreambooth working < ~12vram
🕑 Added 2022-11-18 12:15:51 +0000 UTCHi all, managed to make dreambooth successfully train new models with my code.
In the meantime, been helping the libraries to fix this a small thing or two.
Also implemented Clip Skip, that is what NovelAI uses. It kinda of ignore the finer details in a image for a better overall quality. For real-life art it can look a little washed , but it can look better for anime/cartoon.
Dreambooth training will work with any SD model, vae model, clip skip, scheduler, hyperlink.
For those unfamiliar with model training, training models can break and result in garbage very easily, so i'm currently testing a good default configuration to help the training to give good results.
So right now there is three things to do:
- Test the best values for training, to use them as default
- Add the options to the GUI
- Make a tutorial on how to train a model on Dreambooth
Dreambooth let you train a model with your images. It can work with very little images or lots of images.
Comments
Rai T00n
So nice. I havent tried yet but I will be really happy if it works. Thank you!
GRisk
Yes. Model training can become really complex, so eventually will be more and more options for it, but you will be able to train a model on top of a existing model. (Training from scratch is impossible on a consumer computer)
Leon
Hi, Can we use this SD Dreambooth to train our own models?
