4 ms·
> I doubt you can do it in 8K at that speed All inference of Stable Diffusion is done in FP16 at best, with multi-step models, and models themselves are highly
by terafo 3y ago
> I doubt you can do it in 8K at that speed
All inference of Stable Diffusion is done in FP16 at best, with multi-step models, and models themselves are highly inefficient. I would be surprised if in May 2024 4090 isn't capable of generating 8k(8192x8192) Stable Diffusion quality image in less than 1 second. And also you wouldn't realistically need to
> also why not just compress it? Any quality degradation from compression will be less than the artifacts produced by an AI image generator
Because it allows for much more variance in content. I'm a bit sick of seeing tiled textures everywhere TBH. Also virtually all textures would be AI generated in less than 5 years.
- WithinReason 3y agoI think this is exactly right. The next step is generated meshes/textures and generating entire worlds. You can do generation in a smart way so that performance is not an issue. Predict what you will need before you need it. Generally you would almost never need to generate an 8K texture from scratch, you would generate it from a 4K texture you already generated earlier, which would be much faster. You could have specialised asset generation HW on a GPU.