Tagging should follow the principle of duck typing. If a game is not meaningfully identifiable as AI-cg, then it should not be tagged as such, even if it is.
I feel like most of these assumptions are stuck in 2020 when SD 1.5 first hit the internet.
So is the below AI-Art or Advanced Creation Tools?
)
I'm aware of these, but good point. I'd say it depends on the effect it has on the image. If it's mostly just giving some extra touch to the original image, I'd consider it just an enhancing tool. If the model gives you a pretty different image, just with some similarities from the original (e.g. pose, objects in the scene, etc) I'd consider that AI-cg.
By my own definition, I'd say it still counts as AI-cg, since the output image is completely different from the original, and only ressembles it in concept. Of course, that's just my take on it, everyone has their own opinion.Say I wanted to generate Ahri with blonde hair in a pink maid outfit laughing with music notes and no cat ears.
I could try to prompt that and batch generate and hope for the best.. but that takes way too long and I use AI cause I'm lazy.
(Also the AI detection tools have almost no way of coping with this since they depend on noise distribution of pixels to detect AI images. This tool thinks your avatar is just as likely to be AI as mine is.
Good point, if it truly is impossible to distinguish, adding the tag is debatable. I still think artists should be honest when big part of their imagery is created with generative AI, but it's true that the addition wouldn't make as much sense from the perspective of the consumer.
Interesting.. okay.. what about now? Or now..?
Don't mind me, it's just my personal opinion, we can agree to disagree. Generally, if the result of the tool feels like a different image, I consider it AI-cg. If it looks like the original image but with some touch-ups, then I consider it just a standard creative tool.
Neither!Out of curiosity, are those all img2img generations with different strength factors, or is the latter actually hand modified? It looks almost identical
I'm not trying to argue your opinions - just understand where your line is. A lot of these AI arguments really do just come down to personal feels which you fully admit but that was my point in my original post.
I guess that just proves your point that my knowledge of SD is quite outdatedNeither!
I don't think anyone uses img2img anymore and these have zero hand modifications - that would be work and AI is supposed to remove work.
I haven't dabbled in it for a long while now.
I'd still consider both of these AICG, which isn't necessarily a bad thing.
Ok but your responding to my post out of context.
Lora's / Locon's are still insanely valuable and widely used today both for injecting bias into the diffusion process and for merging weights into checkpoints. It's actually way faster than training checkpoint weights natively so weight merging has completely dominated the checkpoint training scene for the most part.I guess that just proves your point that my knowledge of SD is quite outdated
It was more curiosity and novelty than anything else, so I got bored and eventually stopped. I remember dabbling in LoRAs as one of the last things I did, and those are already old
Personally no I wouldn't consider them AICG, but that's what the rest of my post was about. Just because I wouldn't consider that AICG doesn't mean someone else wouldn't. Which leads to it being up to both artist and consumer discretion on whether the artists feels they need to disclose it as AICG, or the consumer feeling that it should be labelled AICG.
Why not?
Because I base what I'd judge as a tool vs full ai generation based on how much of the creative process was lost by the use of a tool or generative ai.
No way haha, I wish I could draw like that! That pic is from a game called Bedabots, which has a very unique artstyle:
It indeed has created a stigma, since the majority of AI-cg games we get are, frankly, either boring or straight up trash. I think a lot of people find it insulting that these low-effort attempts at games flood the last updates page, though most probably already filter them out by now. Which is a shame, because it means actually high-effort games which are tagged as AI-cg are dragged into this mess.
The second image in the post you quoted matches the line work almost 1:1.
So adobe's tools already pretty much have some kind of generative AI buit-in? Interesting, I didn't know it was integrated to this extent
Watch almost any photoshop tutorial from the last 2 years. Artists who decried that AI was theft are now using these tools all the time. They have no idea what Adobe used to train their tools / models (Its all closed source / server hosted) and they don't seem to care either.So adobe's tools already pretty much have some kind of generative AI buit-in? Interesting, I didn't know it was integrated to this extent
I can see why it would be hard to draw the line then, if commonly used creative tools already provide these functions.
Indeed. Photoshop's best selling point may soon be how to easily touch up AI photo's since there is always something a little off with even the best generated ones. For instance blocking out a hand with 6 fingers, it should use AI to recreate it with 5.Watch almost any photoshop tutorial from the last 2 years. Artists who decried that AI was theft are now using these tools all the time. They have no idea what Adobe used to train their tools / models (Its all closed source / server hosted) and they don't seem to care either.
It's not just adobe either. It's baked into Unity, Maya, Blender, etc.
I mean, it would probably help people out if they are using web based generation (paid or unpaid), but that is fairly built in with stable diffusion and it's offshoots with mask painting and inpainting.
