This is awesome
#1
by
Stkzzzz222
- opened
Why didn't you give it more visibility?
Is there any possibility of seen an updated version??
Anyway, many thanks
Glad you like it! Yeah, I think there are few options for a new version.
One is just retraining the model as is. I know more about training now and could probably make it better.
I've been looking at other ways to make init image models. I tried using my own 8 channel unet and it didn't converge. I could try again with some modifications or there is a new paper by the AD authors that shows how to make an init model with a ControlNet like adapter. I might just wait for them to release the code.