Posts

@Background Pony #7DE6
AI is stuck in this dichotomy where borderline superhuman at adding non-specific detail, but it’s really bad at specific detail. Look at smaller elements of images. Depending on the art style, you’ll need to look at different things, but for most traditional anime-styled images, the most glaring issues are noticeable in hands, feet, eyes, and small accessories like jewelry.

Backgrounds are also often a dead giveaway, since AI is decent at contiguous spaces, but really bad at non-contiguous spaces. Look for dramatic shifts in the background where a character disrupts it. Smarter AI users will generate a full background first before layering a character on top of it, though, cuz it minimizes these problems.

There’s a lot more to it than just these, but those are two ways even inexperienced users should be able to see. More experienced users will know different models have a bias toward certain palettes and patterns, as well, which makes them even easier to identify. Illustrious is a very popular model, for example, but the art it produces is usually identifiable immediately just because of the linework combined with the palette, even if all the other details are sharp and ‘human-like’.

Of course, these flaws are improving on more recent models, but enough experience with them still reveals flaws.

@Thoryn
Personally, I strip my metadata because my workflows use a ton of custom stuff that people won’t have, and they’re pretty much completely unreadable to anyone that doesn’t have all my customs–and only barely readable even to me.

@Draco Dei
Sorry, I can’t speak for that site, and have no idea how their advertising works. I’ve just seen them telling people who upload AI Generated images there to come here.

@Background Pony #E516
As the objective, even when errors are involved (which they are not all the time when I post my vignettes) is turning their “bugs” into “features” it would SEEM to me someone would need to be very thin-skinned to take offense? Am I wrong about that?

@Nocturn
From what little I know about the split, the objection was mostly about the amount of yours the moderation team was having to put in to filter out very badly done AI art. If someone wanted to pay for an ad, why would they turn down the money? Do they garuntee their advertisers a certain PERCENTAGE of pages*map they can only have so many advertisers at one time?

*Mind you that might actually be a pretty morally upstanding way of preventing the problem of preventing part of the “increasoingly small slices of the same pie of user attention” that I heard contributes to the enshitification (spoilered for the official term including mild cussing) of YouTube, Facebook, etc.?

@Background Pony #CF65
The moderators there are asking people to move here if they upload any AI Generated images. But I don’t know if they would take an AI Generated advertisement.

@Star Streaker
I stand by my philosophy of sharing prompt info when people ask for it. I’ll gladly share any Text2Img info but the rest is really up to their discretion and know how. I started into Image Generation in the earlier days and info was hard to come by. and hated it so much because everyone was so obscure about their prompts and didn’t want to share anything. Not even models.

And ya im guilty of not commenting but i really just browse and upload here. But i do really appreciate the bros out there who comment on my stuff even if its just one or two words.

@Trademark2
Bionus imgbrd-grabber works just add your api key for nsfw

Someone with good art should advertise on Derpibooru

I’m in the same boat about not commenting much. I usually follow people who are further along in their workflows, so I don’t always feel like I have useful critique to add. A lot of the time I just like the result. My own process leans more on the AI’s RNG followed by cleaning it up, which has some mixed opinions here.

Also, the metadata isn’t always there to view. It gets lost when you run it through inpainting or editing in Krita or Photopea. Which makes the techniques seem opaque and hard to reverse engineer. I’d love to hear how others approach this. Are folks here open to sharing their processes?

I think that some people just don’t care. I wrote some about ghost signatures and most just keep chugging along. If I present something to the world I want it to be the best version it could be, so I don’t get it why people don’t fix obvious errors. And since text can’t convey my natural snark and sarcasm, I don’t write anything anymore, just edit the tags fitting the error.

@Draco Dei
It can be fun if it’s not meant to be “mean”.

(well, some people do deserve some “notice” about the things they’re not noticing… limbs merging and the like are hard to miss, and yet”

hi chat!!!

This site runs on basically the same code as Derpi, so pretty much everything that works there will work here.

e.g. >>55555t will result in

I have noticed this problem myself. It is a relief in a way to know that this is not some oddity specific to AI art, nor is the commenting likely ongoing on on some other site I was not aware of.
One trick I use to come up with comments is interpreting what is going on, especially in ways that explain the oddities/errors in the generation (or even intentional oddities such as anthropology characters casually interacting in Cthulthu-esque surroundings).

Do I need to explain specifics or can peoole look up my posts via my profile for details?

basically what the title asks, because I don’t know if it’s the same as on derpi

I thought it was because of the lack of actual users, the online censorship thats all happening at once, people be busy or like @thoryn said “I like this” is the comment, and a bunch of other factors. Its something we’re going to have to work on if as content creators want feedback or interaction with… well thats what I think.
Also, people just be busy.

I am still on SD Forge (which also embeds metadata), and it also has a metadata viewer, just have to drag and drop the png files into it.
There’s a few standalone ones as well, just search for prompt metadata viewer/reader.

@truekry
@Thoryn

You can also use this website for viewing image metadata: https://jimpl.com/

NovelAI (which I use) also embeds prompts in their images…

@truekry

Didn’t know people here kept the meta data intact, I always assumed it’d be purged either by the uploader or the site when processing uploads.
Now that I know, I’ll definitely take a peek at how you’ve done things - I’ve enjoyed so many of your images.
(Full disclosure, last time I tried to generate pony images with Illustrious checkpoints, I couldn’t even get characters to consistently walk like quadrupeds, so it feels a bit like cheating if your base tags just works.)

And back to the topic, I wonder if more people abstain from commenting because of the same reasons as you.. if so, then that’s a never-ending circle of silence.

I will try to keep it in mind for the future, but I am inherently an introvert - so I make no promises.

I leave a comment from time to time, but it feels like yelling into an empty room.

I personally never bother to add prompts and stuff since they are embedded into the pictures. Comfy does it and
Invoke too. Just open a AI picture in notepad and all the info is there.

99% of the time, my only comment would be “I like this.” - so I just hit the upvote/fav icon instead.
Many times, I’m wondering what resources and prompts were used - but I figure if they don’t share it from the get-go, it’s for a reason, so I don’t bother asking.

@Background Pony #8339
Generally when people don’t see comments they don’t comment themselves, even if there’s other sorts of activity going on.

This has been a growing problem on any new and small site for some 15 or so years, when Twitter, Tumblr, Instagram, etc. took over as “where people go to talk online”. Even “established” platforms like Deviantart, ArtStation, or Newgrounds are barren of comments, relative to their size.

Most images on this website don’t get any comments, what’s up with that?