Has anyone tried isolated PNG objects, e.g. vegetables, fruit, cars, people, etc.?
Probably these are already also extremely oversaturated so I would have to find some niches.

I would have expected designers to need these resources for their projects like print, advertising, etc.

Is it worth the upload or are regular images better selling?

Today I played a little with a combination of Stable Diffusion XL and an AI upscaler.
I mean, considering that we are still at the beginning of the development cycle and I am just a newbie, who plays with AI tools for 3 months the possibilities are already now promising for the coming future in the next years.

I believe that future AI image models could be structured as "multi-layered" models.
In the first step a rough image sketch with a few diffusion steps (almost like a composition sketch) would be generated based on the prompt and subsequently each identified content of the image would be filled in with more details by using fine tuned AI upscalers.

So you will for sure have fine tuned portrait upscale models for skin, hair, lips, eyes, hands, fingers, etc. And probably also for other type of motifs, which will work like "inpaints". Like a painter, who would add details to an oil painting by several following steps.

This direction is already being explored with models like Stable Cascade.

I think the current diffusion technique is still very ineffective because of the current hardware limitations.
Certainly there will be many new techniques and model refinements in the future which we can't even imagine at the moment.
I also believe that the current generation errors will be spotted and corrected by the AI automatically.

So in a nutshell for me, there is no longer any uncertainty that (stock) photography will be replaced by AI. The question is not if, but when it will happen.

Some following examples.

Generated with SDXL:
https://ibb.co/9VwFMXS

AI enhanced:
https://ibb.co/pWZs2G9

Generated with SDXL:
https://ibb.co/H4skKNC

AI enhanced:
https://ibb.co/b5gZ5Qv

Generated with SDXL:
https://ibb.co/4FtjrL0

AI enhanced:
https://ibb.co/RNjqyDf

Generated with SDXL:
https://ibb.co/KxrRPmS

AI enhanced:
https://ibb.co/GHNT1hQ

Generated with SDXL:
https://ibb.co/HPC9Lm3

AI enhanced:
https://ibb.co/0XtDm3z

Generated with SDXL:
https://ibb.co/HHdjDJK

AI enhanced:
https://ibb.co/9mxFNt9

I think we can agree that landscape / widescreen format sells more than classical portrait format.
Which aspect ratio sells better for you in landscape format? Are there any noticeable patterns in terms of image categories or content?

Does the old 3:2 or the new 16:9 sell better?
I would have assumed that images for mobile use should sell better with 16:9 aspect ratio.

Found today this online AI Tool, which can generate image titles, image desriptions and image keywordings:

https://pixify.io/ai-keywording-tool

Tested it with some generic stuff like ethnical portraits with plants, candy stuff, etc. but also close ups of technical gadgets.
Worked better as expected, sometimes very impressive.
Would be insane, if Adobe or other agencies would implement this for contributors.

But seems to be expensive for batch processes, as it is only available as a subscription model.
So you have to be a pro contributor to get it paid off.