It could be your prompt that is producing it like that. Post your prompt here. Also are you using midjourney or stable diffusion?
I was joking. The AI would not have a problem. You can set the variation level to add more detail. I'll look for some half decent photos. I can scale up some of the low resolution stuff with Gigapixel. But better quality images is never a negative. As I mentioned, the images online of the 90s are piss poor. You'd think something as major as the 1993 Olympia would have better quality pics. Those photographers must have the ability to release better images. Maybe that's an idea for someone, to start a web site like Getty Images - but related to all things Bodybuilding / Physique - photo sets, contests, events, etc. With Getty you can search a person and if they are a big enough celebrity you can sort by events.
Forgot to mention, I am using Stable Diffusion XL on Fooocus. Fooocus is much more user friendly. With A1111 you have to add the loras manually in your prompts. With Fooocus you just choose them from a dropdown. I also find inpainting or img2img works better in Fooocus compared to A1111. But I am sure an expert on A1111 will be able to get great results. SDXL is also newer than the original Stable Diffusion. Stable Diffusion 3 is almost ready for release to the public, its still only available to insiders. I never got into Midjourney because I prefer to have everything local on my computer.
Comfyui is also another alternative to Fooocus and A1111. It works with nodes. I have gotten pretty good lately with Davinci Resolve (I noticed Markus Ruhl uses that) and it works with nodes in the Fusion and Color Correction tabs. Once you get the hang of nodes it works fantastic. A lot of software these days have nodes in some parts of the interface. Blender, 3ds max, etc. I'll have to checkout Comfyui at some point.