I just pasted the link like I had in the past. I don't know what civitai has done on their end, but on their site images don't even display metadata on post for whatever reason. But then they did the next day on different uploads 🤷♂️. I didn't realize this was a .jpg which is probably what they switched to I guess. Seemed like the full resolution though. @maimaimai
The difference between the image in the current post and e.g. this one https://civitai.com/images/1317705 (which is PNG) is that the former is a civitai on-site generation and the latter is an external generation. It seems that "domestic" civitai generations are now(?) only provided in a jpeg format for some reason. screenshot 1screenshot 2
However, this is still an AIBooru problem since the original=true link return a different image (even if it's a JPG) but the uploader claim to be calling an aibooru extractor by providing a link and not uploading the scaled version manually. Maybe there is still some difference in how /posts/, /images/ and direct image links are handled.
the uploader claim to be calling an aibooru extractor by providing a link and not uploading the scaled version manually.
I don't know what you mean by this tbh or know what you're getting at. But no I don't know which link I posted into the uploader @maimaimai, contact or mention a mod/admin of this site with any issues I guess.
This reminds me, I could not get Lovecraft on Dall E 3 some months ago.
I think he is a really die hard concerning AI generating. Well, I tried to get a light bulb in some cool mystic environment and accidentally summoned this abyss demiurge. Even Cthulhu might be scared of him.
I think he is a really die hard concerning AI generating. Well, I tried to get a light bulb in some cool mystic environment and accidentally summoned this abyss demiurge. Even Cthulhu might be scared of him.
I remember I got easily some Lovecraft's on LeonardoAI service a year ago, so I guess it is a thing of censorship.
I remember I got easily some Lovecraft's on LeonardoAI service a year ago, so I guess it is a thing of censorship.
Maybe. Anyway I wouldn't entrust this task to any base model because all base models on my memory didn't like any kind of likeness. In my opinion it's better to use LoRA or IP-adapter for sure.