Pony diffusion sometimes generates things that look like jpeg artifacts, this LORA has been trained on poorly compressed jpeg images and, when used as a negative, shoos away those artifacts.
Make sure to use this lora as a NEGATIVE prompt, if you use it in the positive prompt it will add jpeg artifacts.
Description
FAQ
Comments (10)
would you perhaps be able to show some examples of those images with artifacts or even better comparisons with or without your lora?
It'll need another version since this is a test kinda
since this model here makes absolutely zero sense, let me tell you that i already published an article for this:
(((Ultra-HD-details))) or (((Ultra-HD-quality-details))) in positive and you will never again see any of those artifacts....
Guess I should go take a peek at your stuff once again mang, haven't been genning in a while
Link it please i cant find it amidst your articles
@Malessar simply search for "details"? And minutes after that last article, he comes back with..... Best details you can get.... | Civitai
im still getting it on pony constantly. the model will seemingly just break randomly and any output i get is completely deformed or saturated or just a jpeg artifact mess. i cant i can load the exact same metadata but it suddenly just cant gen right
I tried this "Ultra-HD-details" prompting and it doesn't change anything with these subtle compression blocks and jpeg artifacts from Pony Diffusion. It would be also more rational to, instead of bashing, just point out that in the Automatic1111 web UI (and at the time of writing), LoRaAs can't be added to the negative prompt. People who seek information find such topics through Google after all.
Thanks for sharing! Great resource and an amazing tool for experimenting with on both XL and Pony equally.
You don't mention a tag anywhere. How do I put this in a negative prompt without a tag?