> XL Extra Large? 40 times?
1024 x 1024 instead of 512 x 512.
I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
11–20 of 66 posts
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#12Please consider posting the LoRa on civitai.com as well as the stable diffusion Reddit. These results look pretty good, looking forward to trying it out. I hadn't realized that the generative images buzz was dying out, since I'm using it regularly I guess it is always in buzz for me.
Civitai has syndicated the LoRA: https://civitai.com/models/128708/sdxl-wrong-lora
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#13>A minor weakness with LoRAs is that you can only have one active at a time Uh this isn't true at all, at least with auto1111.
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#14I don't know how these models are trained, but hopefully future models will include bad results as negative training data, baking it into the base model.
It's only mentioned in passing in the article, but apparently it's possible to merge LoRAs? How would you do that, I'd like to use one LoRA to include my own subjects, this LoRA to make the results better, and maybe a third one for a particular style.
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#15This is really interesting. Like mentioned in the article, this is a kind of RLHF, and that's what takes GPT3 from a difficult to use LLM to a chat bot which is able to confuse some people into thinking is has consciousness. It makes it much more usable. I don't know how these models are trained, but hopefully future models will include bad results as negative training data, baking it into the base model. It's only m…
diffusers is working on a PR for it: https://github.com/huggingface/diffusers/pull/4473
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#16Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#17Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#18Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#19This concept is not new. Lots of "negative embeddings" that you put into negative prompts to fix hands and bad anatomy on civit.ai
This submission is about a negative LoRA which does not behave the same way at a technical level.
Re: I Made Stable Diffusion XL Smarter by Finetuning It on Bad AI-Generated Images
#20Earlier quoted context omitted.
XL more likely refers to the parameter count, which is 3 billion instead of <1 billion
No, I think it is mainly because it's optimized for 1024 x 1024 images, rather than 512 x 512 as the previous version was.