>>40103
My basic input is something like:
"Turn this image into an indoor photograph, changing the cartoon character into a realistic human (a young woman). Preserve the exact same anatomy, pose, and composition exactly."
Then I include a bit about the scene I want. For the sooftly diner one, I used: "Make the scene a diner. Include a waitress standing to the right, as if she is taking their order. Include stacks of empty plates on their table. Include a guest in the booth behind these women, who is glancing over and surprised at their appetite."
I have to have it regenerate probably about 5-6 times until it lands on something good. I can make a few modifications to a new image in a follow up request, but I notice the image quality starts to degrade the more times you ask new stuff in the image, so I try to do it in one swoop.
Certain artist styles and forms lend themselves really well to making these great blobs. Also, for some of the original art that has nudity in it, I was able to use photoshop to apply some quick clothing/censoring to it, then use it in the prompt. Or, alternatively, sometimes if the generator was not wanting to make an image because it thought it was indecent, adding the phrase "change their clothing to an appropriate tank top and shorts" would work.