I work for a company that uses Unreal to train AI. The NoAI tag assets are not allowed to be trained with AI. When I look through my library of assets, 1) cannot see their tags 2) I can’t search for tags in my library, but 3) what I really need is to know which assets DON’T have the NoAI tag.
Right now to solve this, I have to log out and look at the store at an asset that I have. It’s kind of frustrating. Any work around?
Did you find any solution? I am also trying to create synthetic dataset and to find assets that do not contain the ‘noAI’ tag is impossible. Also do you know if the noAI tag restricts the generation for Predictive AI (limited to predictions such as labels) applications since the EULA only restricts Generative AI (learns to produce new content).
No, I never found an answer.
Yeah, we treat #NoAI strictly according to the EULA which specifically says it’s not for generative AI. It doesn’t call out other kinds of AI training.
So if i understand correct you use ‘noAI’ assets for e.g. classification or other tasks. I contacted a seller and said that he forbid the use of its assets at any kind of AI. Most creators seems to dont understand whats the ‘noAI’ intended for and the fact that when they upload their creations they agree to specific terms. I saw the PUG dataset from META and they have a disclaimer that the dataset cannot be used for Generative AI. I am also confused about MetaHumans since there is a general restriction (in the metahuman EULA) for AI/Machine Learning/Deep Learning but its not specified for linear media (images). I want to use the characters from the City Crowd Sample but i am not sure. The City Crowd Sample itself has no ‘noAI’ tag and the characters are modified metahumans.
Because artists realize that this is self defeating and hate it.
Oh they do. They don’t want to have anything to do with AI at large due to how generative AI companies (and their fans) treat them. Abuse all the way. Also, with the amount of bad faith now that generative AI companies gathered the feelings are extended to AI as a whole.
You say you want to make a dataset for predictive AI, but nobody will trust that this dataset will not be used for generative AI down the line. If you think it won’t happen if you put disclaimers etc., look at LION-5b. There is a big disclaimer there that they “do not recommend using it for creating ready-to-go industrial products” and that this data should be used for research purposes only. And yet, most of the major commercial image generators use it.
If you find it hard to find data then you have a few ways to proceed:
It doesnt matter what they think or want but what the license states is forbidden or not. I read another forum where it was said that a seller was saying no for a use case but an EPIC employee responded that he should do whatever the EULA allows. I also contanted two more sellers and they ended up removing the ‘noAI’ tag. They had no idea what was its actual intention at the first place. In addition the dataset especially from Universities/Research Institutes are not provided freely in Github or any other source but you need to ask for it and specify your job title, company/university/insitute and your use case. Generative AI will also require specific annotations based on the architecture, which if the dataset does not include because it was intended for another task its useless anyway. There are plenty assets without ‘noAI’ tag but EPIC should had a filter to save us some time.
this is the attitute I was talking about… but also…
… it doesn’t matter what you think or want too, right?