Will AI free us from CUDA addiction?
Zoritae
Posts: 48
Hello.
I've seen how an application has abandoned rendering engines like V-Ray, Clicle, and Iray, entrusting rendering to ComfyUI.
Will AI free us from CUDA addiction?
That would be a huge step forward. What do you think?

Comments
My view.
ComfyUI is basically a AI like any other AI. You are describing your images with phrases, but you are not the one doing the image.
And an AI still needs a motor to run on (NVidia is having a large market related to AI).
I don't use ComfyUI but Automatic 1111 can run on CPU, though crawl would be a better term for it
just like iray can s l o w l y
I'd like to know what 3d application has abandoned actual render engines and switched to AI only, so i can avoid it.
I don't really see any future, AI or not, where Nvidia/cuda doesn't continue to dominate.
Depending on what benchmarks and who you want to listen to, Nvidia pretty much spanks amd all the way round.
Intel isn't even a factor.
That's mostly but not entirely true. You can use your renders as an input for ComfyUI and add some prompts to improve photo realism. Under some circumstances this actually works nicely.
That's right. I've looked into it further, and no, that "application" still uses commercial rendering engines, and yes, it uses ComfyUI to improve and add more realism to the final render.
Anyway, we'll have to keep waiting...
ComfyUI on my PC uses the specialized nVidia created SDKs/extensions via python & Windows libs quite extensively. It's DAZ CPU slow in imaging or LLM querying if you don't.
No, Comfyui won't replace Iray any time soon. They're not the same, or rather work differently. I don't think anyone would want a ray tracing engine that hallucinates and renders 10 different types of cubes. Iray does use a little bit of AI, but only as an enhancement. Until we get real-time AI rendering that is comparable to real-time ray tracing, that question is far in the future. But, perhaps someday, instead of sending in a list of vertices and faces, we can just describe the object we want, placed at coordinates (2,3,5), and have it render in real-time, but that is far off. But, before that, I think someone will create a AI pre-compiler. Something that takes a AI-scripted game and convert it to game that has renderable geometry.
AI is mostly if not entirely dependent on CUDA, so in that sense, far from it. To the actual question you're asking, that depends on whether one is comfortable using AI, given its horrific environmental cost, the disregard towards intellectual property the developers of AI engines have shown, and the questionable artistic merit of using AI in the first place.
No thanks. I'll pass on this every single time.