Can someone explain to me why AI can't retopologize 3D models yet? Retopology sucks and everyone I know has wanted the machines to automate the process for years now, yet for some reason it seems impossibly hard to train an AI to do it? What gives?
>>108280132If you can't define a mathematical algorithm to retopoligize 3D models then you don't actually know what you want and you're operating purely on vibes, how do you expect a computer to recreate that?
>>108280132Auto retopo was good enough before AI craze was even a thang. You just had to fix minor issues to match your deform case. Absolutely no professionals have manually retopo'd in the last 10 years. Auto unwrappers and packers have been extremely good too in the last decade. I don't see the point in either getting any better.
>>108280132bruh...Just write your own fuckong algorithm for thatStart with a naive one like vertex reduction and go from there.
>>108280185what do people on /g/ have to gain from lying like this?
>>108280142Heuristic algorithms provide useful solutions for a lot of non-verifiable problems. For example, optimal packing of products for shipment. An exhaustive brute force search is very expensive and you can't verify that the result is optimal without performing the entire search again. So you take the first result that meets some criteria. Retopology might be similarly probabilistic and imperfect but it would still be useful.
>>108280142Why hasn't anyone written the algorithm yet then? Aren't vertices inherently mathematical in nature?
fuck off kris
>>108280132> Can someone explain to me why AI can't retopologize 3D models yet? > yet for some reason it seems impossibly hard to train an AI to do it? Most modern AI is trained based on existing data, usually free or cheaply available online. Hence why most content creation generative AI is text, audio or raster images.I **ASSUME** a lot of the good 3D data, specially regarding retopology, is either expensive or straight up not available.In case of retopology you would need not only the original models but the optimized models as well. Artists and corportations would have to publish a good chunk of their workflow online. Save of some ultra-rare exception, I don't think there's anyone doing that. It's virtually impossible to train good extensive models.
>>108281589There's always the models resource site, though that's just the already retopologized models, not the base meshes they're made from. There's probably also potential legal concerns too, I'm assuming.
>>108280132AI is already pretty good at making a retopo base mesh from flat images which saves a massive amount of time. i dont mind doing the retopo work by hand after that
>>108280443guess what ai isn't good at
>>108280132it can
>>108280132Cris? WTF are you doing here? Go back to >>>/3/
>>108280575>>108282367Who the fuck is kris/cris? I'm guessing a lolcow of some kind?
>>108282594cris is the schizo of stagnation. He's been at this for 13 years minimum. https://m.youtube.com/@ca3games/videosHe has the miraculous ability to never improve.
>>108280132The way 3D graphics and AI is headed is mostly>make an Bethesda level unoptimized model>use an incredibly optimized rendering pipeline (like nanite) to render itIts a real shame, its really impressive what can be done with low poly
>>108280132>3D asset creation has long been the bottleneck that frustrates the creative and the technical alike. Wireframes. Topology. ReTopology. UVMapping. Texturing. Procedural or Not. Time. Budget. Tencent’s new Hunyuan3D-PolyGen AI is changing the game—not with flashy gimmicks, but with real, working solutions built for people 3D Technical Artists like me who actually develop content.
>>108283548>an Bethesdagood morning sir
>>108283631Fuck you bloody blanchod
>>108283596>built for people 3D Technical Artists like me
>>108281655>There's probably also potential legal concerns too, I'm assuming.As if that stopped any AI company
