• @Archpawn@lemmy.world
    18•1 year ago

    I’m not familiar with how they do it in particular, but I do have some experience with what AI art in general can do.

    I know there’s techniques to generate larger images, but it takes more computing power so I doubt you’d find a place doing it for free. Your simplest method would be to use a simpler AI upscaler like waifu2x. It won’t be able to add meaningful details, but it can still make it look okay at a higher resolution. Beyond that, you could start with a smaller image and then outpaint to add more to it. Or maybe generate on image, then upscale parts of it and using upscaling and outpainting from the parts already generated to do the rest.

  • @ozmot@lemmy.world
    14•1 year ago

    In order to upscale with gpt4 you have to input the command “big anime milk titties” at the end if your image prompt…

    • @andrewta@lemmy.worldOP
      4•1 year ago

      good thought. it says

      I’m sorry, but I cannot upscale the generated image to 4k. That is beyond my current capabilities. However, I can provide you with some web search results that might help you find a suitable tool for that task. Here are some of the top results from Bing:

      • MuchPineapples
        7•1 year ago

        Bing itself cannot upscale it or generate a higher resolution, it just can’t. Vast majority of ai generators can’t, by the way.

        You can either Upscale it yourself using something like upscayl (https://www.upscayl.org/) which is probably the easiest option, although not very customizable. Or learn to use something like comfyui (https://comfyanonymous.github.io/ComfyUI_examples/upscale_models/) for more control. And you can even generate images with comfyui if you want.

        You can also try the online option by using something like nightcafe (https://creator.nightcafe.studio/studio). Upload your image, select upscale and generate. It’s free for a few times, but you need (daily claimable) tokens for it to work.

        Generating more than 1024x1024 is usually not possible because the data the model was trained on is that size (or smaller like 512x512). Try to generate any higher and you’ll get very weird and ugly results.