Originally posted by Lebowski
View Post
Photoshop has, for years now, had the "content-aware-fill" tool which is a sort of AI-based tool to speed up workflow. But that only uses data from within your image (or what you supply) and has to be used in the correct way, which takes some skill. It had no chance of "replacing" any creative; it mainly just allowed creatives to get a better result in a bit less time.
Adobe is closer than anyone else to getting past many of the ethical issues because of Adobe Stock; they're dealing with a huge dataset of people who have given permission to use their stuff. But I've used Adobe Stock prior to this and I know it has huge blind spots.
I'm seeing a lot of debate about this stuff online; same for ChatGPT - of whether AI is going to move in jumps, rather than a constant drip of improvement like most technologies. There's been the suggestion that ChatGPT might be nearing the "best it can be" for that type of language model, and while improvements are possible, they could offer diminishing returns until someone figures out the next big jump, and there's no idea when that will be.
AI generated imagery, especially for character illustrations from scratch, is the same; in terms of generating stuff from scratch, I keep up to date with this stuff, look at imageboards etc. where people post AI "art", and honestly, already, it's thousands and thousands of really quite similar images. In the anime space, you just see pages and pages of weirdly similar-faced anime women, usually with large boobs, rendered with backgrounds that look like knock-offs of Makoto Shinkai's work, all rendered in the same painterly style. And they all look dead; as-in, there's no through-line of action in the images. The character is just standing there.
Conversely, I'm seeing stuff which is interesting where people fold these tools into their workflows; but again, that's skilled work.
Comment