I’m not going to link here because I don’t want to be accused of brigading or singling anyone out.
But, the amount of AI generated content on the asset store is starting to trickle in before the inevitable flood.
It’s trivial to make and, in my opinion, not worth paying money for if I can produce similar results in minutes AND my results will be relatively unique as opposed to an art pack purchased off the unity store that everyone has access to.
I don’t have a particularly huge problem with it existing for people who disagree with that sentiment, but it’s only a matter of time before I get misled into buying it if it isn’t clearly marked as AI generated art.
Not to mention the potential copyright issues that have yet to be ironed out in regards to this space.
Is there any word at all from Unity on how they plan to address this, if at all?
It takes between one or two months to get hang of stable diffusion with default checkpoint and learn how to produce anything decent. Then complexity goes up because there’s huge number of tools that affect quality of the result. Using combination of control net, img2img, hypernetworks, embeddings with textual inversion, LoRA and custom checkpoints, a skilled image generation “operator” will absolutely make results that you will NOT be able to reproduce in minutes.
Long story short, unless you spent ridiculous amount of time using image generators, your results will not be unique, they will not be good, and they’ll be easily recognizable as AI art. Basically, they’ll look like something akin to Daz render, which is infamous for being related to a huge number of horrendeous looking comics.
So, in regards to asset store.
If you do not like what’s being offered, I’d advise not to buy. If you believe you can create the same thing yourself, I’d advise to do it. In the end people will vote with their wallet. Low effort assets will be buried, high effort assets will remain. “AI generated” does not mean “low effort” by default. Making the tool do what you want and ,making it produce high quality result does take time. And effort.
If you want to buy AI art, then buy AI art. I however do not agree with your sentiment that it takes any great length of time to churn out relatively good looking results, and have been doing so since day one. I can produce portrait quality images in under ten minutes using the right models. I have been creating these as placeholders for my current project and have debated just keeping them.
As the quality gets better and better and the freely available models get more and more distinct, the visual distinction is going to get harder. What I’m asking is not for a blanket ban on AI art, but for some kind of distinction to be made on the asset store so that people like me, who for their own reasons want to avoid it, can do so.
I’m speaking from experience, and as far as I’m aware my statement is correct. I think I’ve generated ten thousands if not tens of thousands images by this point.
It’s been a while since stable diffusion and midjourney were released. There were ton of people using those to produce the artworks for books, etc. Right now those artworks are recognizable as “quick ai art” and they look cheap. I call this “Daz effect”.
The problem is that anything you can produce quickly is going to be recognizable as AI art that can be produced quickly. So, if you have “quick placeholders” there’s a really good chance people are gonna spot that those were generated.
Anything that is not recognizable immediately and looks good requires work and can easily take hours to the point where it would’ve been faster to paint it yourself from scratch. For example that award-winning ai painting took 80 hours. When I’m doing anything serious for myself, I’d need to clean it up with img2img to death, fix all the tiny glitches, etc. And that also can take hours. Even finding a good image takes time. And if the “operator” uses poseable templates, you will not be able to replicate the result easily.
This is what I’m seeing flooding the asset store at this point, so we’re in agreement there at least. That’s going to be difficult to spot however when it comes to the “lower detail” things like RPG spell icons, and I’ve noticed you can get much better results for anime/toon effect images from stock model and plain English descriptions than from the photorealistic models.
My concern is that a good amount of the 2D content being uploaded recently is clearly AI generated art and it’s only going to end up swamping the asset store to the point where finding anything else will be like a needle in a haystack situation if some kind of control measure isn’t put in place.
One would like to at least hope the asset store was a curated resource of quality goods.
If you cannot spot that it was generated, then it is good. If it is good, then it took time to make it. If it took time to make it, then what is the problem? There was always plenty of poor content on asset store. I’d expect people to just vote with their wallets.
Because as I have mentioned several times already, I do not want to buy AI generated art. Not sure why you’re pushing this point as you are not going to change my mind. I am not interested in debating the nuances of quality vs talent.
To actually answer a question (even if with an option) and not debate…
I don’t really see Unity as a company that will enforce a tag like “AI Generated” like a couple art sites have done.
In my opinion that would not be fair anyways since it would give both the people who put real effort into it the same stigmata as quick-prompt-n-dump assets.
Indeed the market will have to solve this if it turns out to be an issue.
The only ways to absolutely ensure you don’t have AI work in your project are to either create all yourself or hire an artist and then look over their shoulder (I’d not trust people on Fiverr for example if I cared that much about how the result came to be) xP
EDIT: I am curious whether quick dumps of AI art will actually appear. After all the Asset Store DOES have minimum requirements on effort and that is actually being applied (as one can see by the occasional complaint thread).
Asset store always had a few low quality low effort assets available.
There were also accidents. For example I recall a guy who used Giger’s works to make textures. Which is not allowed (as copyright has not yet expired).
How about artwork that was created with the assistance of AI? I’m not just talking giving words to a machine learning model and letting it do the actual work. I’m talking giving various intermediary stages of your art to AI models and letting it make fine adjustments.
I’m a programmer not an artist so I can’t speak to the efficacy of using it for artwork but what I do know is that transformers like ChatGPT have been very helpful these past few months. If I were to remove it from my use it would require more work on my part without it than it with it which translates to a higher cost for my work.
I think AI is here to stay one way or another and (not that you suggested in anyway), the idea of an outright ban just isn’t going to work. It’s part of the workflow now.
My issue is I’d like if we were able to get some agreement from Unity themselves on what qualifies for sale.
An artist that uses AI in subtle (but powerful) ways to save time using the cut tool, or add texture to a flat object, or recolour something, would be acceptable to me. Grudgingly I’ll admit, but ultimately there’s an actual artist at work there at least and it’s difficult to distinguish that on a moral level from something like hue shifting.
Typing “(male | female), {long | short} {red | green | blue} hair” into waifu diffusion then outputting a batch of 500 while they go make a coffee however, is not in my eyes good content and because just about anyone can do so, the asset store is poised to be flooded with this kind of content once the 5 step guide to get the webui running goes mainstream.
I’ve spent several hours with Stable Diffusion on PlaygroundAI as well as on my own computer. Yes, you can use very basic keywords to get results, but all of the very competent ones out there are not as simple as that. It’s usually dozens of keywords both for positive and negative prompts and searching hundreds of seeds.
Here’s the one that I’ve been using. It’s a one-click installer followed by double-clicking a batch file.
Why would they be significantly restrictive in that regard? In the end it’s revenue. Plus one should expect buyers on a store like the Asset Store to be responsible enough to look closely at what they buy and whether it fulfills their needs. It suffices if the store keeps blatant lying at bay (hence why some verification and a reporting feature is in place).
Then let’s see if we can define something less arbitrary, but you repeatedly interjecting to say “no” is neither helpful nor constructive.
It’s pretty unfortunate that there’s no incentive for them to have any quality control, as I’m a buyer rather than seller I’d hoped there would at least be some in a more broadly general sense.
Though the verification and reporting is what I’d like to get some kind of discussion on for this topic. Clearly though it doesn’t seem to bother anyone else.
Is it that bad? I keep a close eye on the store (like the unfiltered preset with “newest first” is one of my main bookmarks) and cannot really complain. Sure occasionally packs of 200 icons or 50 static character portraits pop up and I do really wonder who the heck would pay money for these?
But it’s not that much that good things would be entirely submerged.
It’s an oncoming problem I think. AI models are getting better at generating passable content, and more people are clocking onto the idea that they can make a quick buck this way.
I’ve noticed more this month than there was last month, and I suspect it’s going to get progressively worse fairly quickly.
Honestly, I like browsing the asset store for interesting content and I’ve found myself in a handful of occasions now wasting time looking at something that’s clearly AI generated once you get past the thumbnail.
I’m worried that this time next year it’ll be 10 to 1 and the new tab is just going to be flooded with this stuff.
I’d recommend automatic1111 build. Though it doesn’t have built-in lora training support.
This sort of preference kinda boils down to “a living human has to suffer for X hours to produce the piece, otherwise it is not worthy of purchase”.
In real life quality tend to matter more and not the time spent, and more skilled artist can be paid more while spending less time. I’ve also seen people make amazing pieces using photobashing.
Also, “anyone can do the same” kinda does not hold, as even if person comes up with the same prompt, the result will be different.
I can also easily provide a couple of examples that will be quite hard to replicate…
If that’s what it is, then that’s what it is. I am entitled to feel that way, regardless of whether or not you agree with it.
I feel like this discussion with you is going in circles. I have made my position clear and have repeated it in multiple posts as you have repeated the same arguments in kind. I am not interested in debating my morality on the topic - simple figuring out if we can get Unity to listen and give us a way to filter it out in some capacity so I don’t have to sift through pages of low effort, low quality work that will hit the store by this time next year.
I went with the one I did because I couldn’t get anything else to work. I don’t know what’s up with my current installation of Windows but I kept getting errors no matter what I tried. Easy Diffusion keeps its own copies of everything in its installation folder.