Conversation
Notices
-
Embed this notice
@thendrix Flux seems to do an alright job but SDXL is really awful at it
-
Embed this notice
@mrsaturday @thendrix flux is amazing looking but also has a more narrow scope of what it can do sometimes in my experience
-
Embed this notice
@mrsaturday @thendrix I am working making loras for it right now. I'm having a little trouble but the potential is huge
-
Embed this notice
@sun @thendrix I'm hoping the autists start messing around under the hood. At least, I'm hoping it's scaleable like Stable Diffusion is and not locked down.
-
Embed this notice
@mrsaturday @thendrix I have been able to get it to imitate clothing based on a specially trained set I gave it. but I have been trying to train it on images of a model to make more pictures of her but it's been difficult.
-
Embed this notice
@mrsaturday @thendrix they definitely are doing something in the base model to prevent it in most cases but in some cases it is perfect so I don't think they are suppressing you from making your own
-
Embed this notice
@sun @thendrix I do wonder if they have built-in safeguards to keep people from making imitations of celebrities. I haven't looked terribly deep into what it's about from a technical standpoint and I'm stuck using web clients because I only have 8 GB of VRAM.