• @[email protected]
          link
          fedilink
          English
          324 hours ago

          Yeah, though would it be issue with the training data rather than the model? For example details on her hair and face are almost 10/10, and it’s not like those are easy to get right

          • @[email protected]
            link
            fedilink
            English
            324 hours ago

            I suppose you’d have to feed it a critical mass of every kind of detailed object. Which isn’t viable if you’re trying to create a general-purpose image or video generator. If you want to forgo perfection to get a product that’s just ‘good enough’ that’s what you’d do, and that’s what model trainers are doing.

        • merde alors
          link
          fedilink
          English
          124 hours ago

          they’re dreaming. Details are not necessary for dreams.

          • @[email protected]
            link
            fedilink
            English
            123 hours ago

            Dreams don’t make for a good product. Unless the models produce useful outputs, it’s either a research project or a waste of GPU/CPU cycles.

            • merde alors
              link
              fedilink
              English
              219 hours ago

              “product”

              “useful”

              “waste of GPU/CPU cycles”

              we don’t speak the same language.

              • @[email protected]
                link
                fedilink
                English
                318 hours ago

                Oh didn’t they tell you? Making a profit is the most important thing in this world. Stop doing stuff that does not increase shareholder value!