pilk said:
I agree. NVIDIA is the better option for AI.

I'm just adding AMD is also viable if you happen to have it for any reason, you're not SOL and can still use all of the usual CUDA frameworks.
For context, on a 7900 XTX, it takes about a minute to do NoobAI with 25 Euler iterations at 2048x2048 base resolution or about the same time for 20 iterations at 1200x800 + hires fix. You don't even need all that for good results, I sketch with 10-15 and it takes seconds to see the changes.

tldr - don't be discouraged, anyone can gen

I know that it is possible to gen with AMD. But it is not recommended, still.
It just doesn't make sense to buy AMD with you wanna use it for AI. My 5060ti has cost only half the amount of money a 7900 XTX would. And it will gen an SDXL image at base res in a few seconds.
And we are only talking about most basic inference of SDXL here an old model with low requirements.
Can your run Wan14b-720p model with your GPU? Or what about Flux? You will have big trouble with that. While I can all do that without issues with my much cheaper video card.
Did you ever train LoRA with your GPU? Despite the high VRAM it will most likely go OOM. AMD memory management is awful.
AMD might appear cheaper at first glance, for seemingly better specs, but it's a trap. A GB of AMD VRAM is only worth half a GB of NVIDIA VRAM (tho that is not actually related to the VRAM itself).

So my advice stays the same: Don't buy AMD for AI.