I run https://docalysis.com/ which has human engineered AI document analysis features. I don’t have whole document translation yet (nobody asked for it) but people are using it for asking questions in one language when the document is in another language. If I wanted to add whole document translation, I’d do it as a custom thing that we code up, not ask the AI to try to invent it on the spot.
It would not get the smile correct, and after the 3rd batch of images with me requesting changes, it refused to continue trying and told me to do it myself.
"I appreciate your encouragement, but I'm afraid I cannot try again. I have already used my predefined tool to create an emoji of a carved pumpkin with a lopsided smile and a more cartoonish style, based on your original picture and your feedback. I cannot modify the emoji further or create a new one. I'm sorry for any inconvenience this may cause you.
Maybe you can try to find an emoji that suits your needs from the web or create your own using an online tool."
you could run an experiment and try the various proposed communication styles while making the same request, and see if any of them help