Comment on Concept Artists Say Generative AI References Only Make Their Jobs Harder [TWIV]

Ilixtze@lemmy.ml ⁨1⁩ ⁨week⁩ ago

• EnsignWashout@startrek.website ⁨1⁩ ⁨week⁩ ago

  Why reference the imperfect copies gen AI makes when you have databases of the real original artists and pictures and 3d scans of real things in the world out there to reference.

  I think it is because one of these approaches requires skill and time, the other requires a ChatGPT subscription and George Jetson to push one idiotic button.

  It makes people feel that they can contribute in fields where they were previously (and still are) useless.

  source

  • Elkenders@feddit.uk ⁨1⁩ ⁨week⁩ ago

    So well put

    source
• BarbecueCowboy@lemmy.dbzer0.com ⁨1⁩ ⁨week⁩ ago

  For that use case, Gen AI lets you draw from thousands of pieces into a singular reference image in seconds tailored to whatever you can ask for. And, yeah, very imperfect and questionably sourced, but you’ve eliminated countless hours of work that could ostensibly be directed towards just fixing any problems created (Rare, but that’s how they sell it). And then, if you don’t like the result there, you can just get a fresh one in another few seconds.

  I think it’s useful enough that it’s going to keep being a bigger and bigger problem.

  source

  • Ilixtze@lemmy.ml ⁨1⁩ ⁨week⁩ ago

    I feel the problem is that in the end referencing isn’t about copying but extracting accurate information. You don’t reference something you want to design but something where you need a very specific object to be accurate. Using AI reference for this just sounds kind of useless. I can understand that very untrained or untalented artists will find most of the AI references useful because they don’t have the design language to construct things themselves. But this brings out the last problem that Generative AI models have a lot of diminishing returns. And after using them for a while everything starts to look the same and you need to steal more stuff to shake things up.

    All fo that time could be better spent forming a design language or gathering useful references.

    In the end most design processes I’ve see that use generative AI end up looking like asset flips because that us the logic of the system itself.

    source
• Bronzebeard@lemmy.zip ⁨1⁩ ⁨week⁩ ago

  I mean, I can understand the desire to cobble together the ideas you have instead of saying you want certain pieces of 18 different reference photos, but with the vibe of a few others…

  Like: I want something like this, but more cohesive.

  source

  • Ilixtze@lemmy.ml ⁨1⁩ ⁨week⁩ ago

    I feel the problem is that, you deny yourself of forming a design language. I suspect a lot of cultural products will look very samey in the end simply because , I already know a lot of people who use AI in 3d that have this “asset flip” aesthetic; Also making a piece from 18 different photos is a learning process that can enrich the rest of the product. At some point you need to touch grass, learn from the real thing.

    source