Going to be real: how is this flagged as AI? I've been looking close at all the places where AI would show up and be messed up but I can't find any exact detail. In fact I'm seeing a lot of details that show PostProcessing or human input.

SilverSenpai said:

Going to be real: how is this flagged as AI? I've been looking close at all the places where AI would show up and be messed up but I can't find any exact detail. In fact I'm seeing a lot of details that show PostProcessing or human input.

Check the windows on the building on the left. The distribution and shape patterning is uneven on the right hand side, particularly the large windows. Then whatever the hell is attached to the building on the background on the right side: it attaches thin-side onto the wall, where realistically you'd put the side with larger surface area. Then lastly the size and shapes of the windows on the same building are uneven and the particular one partly hidden behind Golshi's right side hand has a messed up perspective shadow-wise.

looking at other works from this creator it could be ai-assisted, playing devil's advocate i'd say the background element probably comes straight from a prompt with some minimal touch ups, golshi looks to me like she is a 3d model posing on an angle from a virtual camera that had toon shading, may or may have been put into an ai pipeline for processing and then had some amount of drawing over (no idea how much or how little) to add the blush and color detail on the elbows

however a couple of works from the creator seem like they may have no post-processing nor drawing over the ai generated image so it is a good idea to look at the works from this creator more closely to be sure

tho i say that as the "3d quality" of the art reminds me of bekobeko's 3d model + ai and touch-up + draw over pipeline

for context, bekobeko started doing 3d modeling apparently as a step towards getting into animation and then drawing over the 3d models to produce single image and manga works, little over a year ago bekobeko began training an ai model with his pre-3d model drawover works to skip the whole sketching and pose sketches phase by simply posing a 3d model, screenshotting it, processing with ai to get a "fine sketch" and then do from more invested to minimal draw over (many doujin works from bekobeko received criticism from how bad the ai processing was and how "lazy" he was for not doing a better draw over), later last year some fan with more experience in ai trained a better model on bekobeko's older work and made it publically available at civitai, like 2 or 3 months later bekobeko switched to using that model for multiple doujin manga works (altho not openly stated but the background and shading artifacts from the fan made model are consistent with those found on recent bekobeko works) while his single image and some doujin covers seem to have no ai processing and be drawn over a screenshot of a 3d model.