I tried to use getimg.ai to turn a picture of an anime girl sinking in quicksand into a different anime girl. The results were mixed. The AI understands how to represent fear if you put that on the prompt, but it can't draw crying in a convincing way. If you use "image to image" in reference, then you will never get consistent depth. To do that you have to use "Hard Edges" set at about 85. Still it's not perfect.
>If you add more than one girl, it generally gives up and gives them whatever emotion it wants.
>If the girl has a stupid hat or ponytail, the head warps into something ugly
>Sometimes it will draw lazily with little details and there's nothing you can do about it.
This was the only interesting one I got after wasting like 50 credits. The neck was close to ruining everything and as you can see, the background was totally butchered.