Tech info/text encoder info.
Hey, I'm interested in the fact that you've trained this model with licensed data but I'm struggling to find any technical info on the model's training. Are you able to provide a link to a paper or anything that goes into more detail please?
If that isn't possible, would you be able to let me know what text encoder you trained the model with.
Appreciated.
Hi
@IlluminatiAI
- Thanks for the question.
Bria 2.3 was trained with the same architecture as SDXL starting from random weights and based on licensed data only.
Currently the text encoders we use are CLIP based but we are going to release a T5 version within 2 months.
We plan to release Bria 3.0 by end of Q3 - this model will be based on transformers, T5 as text encoder and brand new trained VAE based on our data
Do you have a specific use case that you wish to use our models for?
Color me excited about this release