Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
Tags

Question about the AI resources

A topic by JasperDProfio created Jan 01, 2024 Views: 118 Replies: 5
Viewing posts 1 to 5

So I've been interested in using AI generation, specifically for backgrounds in stuff I'm planning, and I was wondering if anyone knew where the sites linked as recommendations in this jam sourced their data from? Like, I know that there was a big issue with artists having their work used to train AI without their permission, and I feel like it could be a really useful tool in art generation but I also don't want to be using someone's art without their consent. So, are the sites recommended for this jam getting their AI training data from artists that volunteered, or was it just kind of grabbed from the internet? I'm genuinely really interested in exploring using AI for areas that I'm not great in, but as an artist I don't want to be using anyone's art if they don't want me to, and it's hard to find details on where AI image sites trained their programs.

Submitted

Rarely do any AI companies share their data sources.  The way I think of it is...Don't ask for a specific artists' style or give it examples to copy.  Don't give trademarked character names, and if you're worried about it, do a reverse image search using tineye or similar to verify it's not a close copy.  If you're specific in your prompts, your response will be so specific that it's just yours.  It's extremely unlikely someone will get the same thing if you're specific enough in the context of your conversation. For example, here's "A visual novel scene with a detective standing in the rain looking at a body on the ground in front of a hotel".  That's quite specific, but you can get far more specific to be more...thematic.  I would recommend that you work on putting your story in and make whatever program you're using familiar.  Then try to get images that match your story with really long prompts that make it hard to be copies of something already out there.

I'll be honest, that still makes me a little uncomfortable. Sure, if I'm precise enough the chances of directly copying someone are lower, but it was still trained on stolen art and may still have elements of that unwilling artist that I can't pick up on. I'd rather go into it knowing that it was trained on art that they got permission to use first.

The only results I can find for fully artist-approved training is the new Adobe Firefly stuff, but I'm not sure how good that is because it's still pretty new and just uses Adobe stock images.

I'm thinking I may use Adobe Firefly, although the amount I can create without a monthly subscription is pretty low, the subscription seems pretty affordable (like $5/month) and I like the sourcing of images. I made this with the prompt "an abandoned plantation home at night with a large dead tree in front; there is a cemetery visible behind the house". It let me do the selective tweaking too that they recently integrated into Photoshop.

Submitted(+1)

Ultimately, you have to do what you feel comfortable with. 

I'll be honest, a lot of the conversations I've been having with other artists have actually made me personally a lot more comfortable with using AI. Not because anyone around me supports it, but because their arguments have shifted from "it's sourcing art without permission" to "it doesn't have soul and it's lazy", and imo those aren't real issues. I get the feeling for a lot of people loudly against it, it was less about the art theft that was happening and more about people thinking that art generation had to be difficult to "count". And as someone that regularly struggles with stuff only to be seen as "taking the lazy way out", that rubs me wrong.