AI Demo Jam is a new event series organized by Dnipro VC together with Data Phoenix. On March 20, six AI startups presented their products live during the first AI Demo Jam. In this video, you'll see the demo from CopySight AI.
CopySight AI: first tool to enable safe genAI usage for enterprise-level content.
Transcript
Konstantin: I'm CTO of CopySight. My co-founder Tommy is now in LA talking to movie studios and fashion brands. And I know from personal experience of being a software engineer that Gen.AI cuts down development costs, but even so it cuts down the cost of creating digital content. But the issue is that most enterprises still don't allow it, especially in the production environments. And one of the biggest reasons is copyright. US Copyright Office says that AI-generated content is not copyrightable unless there is significant human involvement. And the other issue with it is when you generate content, you don't even know if you own it, because most of the models are trained on a very wide set of data that oftentimes includes legally obtained copyrighted content that the foundational model providers chose to train it.
So today I want to demo part of the pipeline that we're doing for the enterprise clients. It allows them to verify that the content they create is safe and legal to use. So let's say you're maybe a marketing agency for a fashion brand that's designing a new perfume ads and you want to throw something in there, it immediately highlights that you have here a couple of characters that are, sorry, it's my network filter. Privacy first, as you can see. Pokemons, Pikachu, and I forgot how to pronounce it, but anyways, owners, respective copyright, Chanel trademark and my co-founder traveled to New York to Chanel's headquarter last week to have a demo with them so it wasn't chosen for a reason. And then maybe you're doing things for the Nike. And what we did find is certain models, for example, Flux in particular, Flux are the people behind stable diffusion. Anytime you try to generate a shoe, it generates Nike shoe. Not much variety in the training data, you can tell right away. And this is one of our favorites. It's Punisher. It's a Winnie Pooh with a Punisher shirt on, stylish, but totally illegal to use in any commercial contents. Or maybe something like this. You know, cute, but it violates somebody's copyrights. So basically this is part of our pipeline and we first make sure all the content that's AI model generates is safe to use and then when they work through our pipeline, we document every single step of creation and creates basically document the proofs, this is the human involvement, this is what they created and they own it. And that's us. Thank you.
Audience: So what's the 95%, 85%? At what percent is it actually?
Konstantin: So kind of from what we determined, and this is so far, you know, we have a lawyer on staff. He's a former head of IP at Snapchat, and so is my co-founder. We settled on 70 - 75% just based on kind of like, you know, his experience dealing with copyright law, what it visually seems like. But we did notice that most of the things that are fridge there are in the 85-90% anyway so most of the time the guidance is if it's above your 75%, no if it's below 75%, consult a legal team, you know, we're not saying yes and no we're just giving companies guidance, so whether they can proceed with it.
Audience: You said that right now, it`s a lot of identifying, are you guys, also suggest something?
Konstantin: So, sure, yeah, good question. So most of our clients are well-established, and they usually know what they're doing. So our bigger pipeline is a tool that actually helps them go through the ideation and design stage where we can give them access to the models. Some of the models are fine-tuned, but it gives them that's their ability to put human creativity in that allows them to copyright it basically, you know. You're not just the prominent engineer, you can throw in a sketch and they will continue from the sketch. You can do in painting design and then we document all those steps. So it's a Photoshop plug-in but basically it leverages the identification to make sure you know the stuff they create is you know novel it doesn't violate copyright.
Audience: You mean it`s like images or videos?
Konstantin: Images, videos, gaming assets pretty much any kind of visual arts.
Audience: Does your pipeline also identifying? Do you differentiate them?
Konstantin: Yeah, that's actually a great question. We settled this completely by an accident. We had a conversation with… I guess I don't know if I'm allowed to say the name, but we had a conversation with General Counsel, one of the largest movie studios in LA. And in a conversation, she was like, what if our designers use AI and don't tell us? And we were like, okay, we can. We do identify it. It's part of the pipeline that we're building. But that task is actually pretty much solved. We have our own model. We have a couple APIs that we can call if the client allows sending their data to an outside API. It's somewhat a solved issue. We are not going after the market, it's just a value add that we have basically because a lot of executives sometimes worry that their designers will do something they're not supposed to. And we have a couple of clients that to be honest are borderline in that area. But we're guiding them towards a process where they can establish procedures that allow them to own the property that they create.
Audience: For the content: do you allign with the former or new model? And the second question: does it work on video?
Konstantin: Yes, this can work on video. Basically, video is basically images if you're, like we can identify when the frame jumps and then just take the stills of those images. We're thinking of even more robust pipeline, but that's what we can do for now. And we're definitely a company/designer focused product, but our idea is in creating new, not infringing on old. Like a lot of times, there's only this much that we've created. There's so much more opportunity to create more. So I think we help people understand what's their energy. We don't bar them from using this. So if they see something like a Spider Man, maybe there are ways for them to work with a respective copyright owner and obtain, that's something that we can expand, we can partner with somebody else to do. But I think our goal is to enable designers to create new things without getting in trouble and actually owning what they create. So in the future, when there is a market for that type of distributed royalties, they can claim the ownership.
Dmytro: Thank you.
Konstantin: Thank you.
Comments