Anime Elves have ears precisely like this.
Thank you. I am not into anime style so I didn't make the link. And it's very clear that the model was trained with lot of it, so it makes sense.
Meanwhile, I decided to subscribe to chat-gpt pro service. It's not the monetary amount (20 $/month) that blocked me but the fear of being let down. But I did it because I had spent 10 boost trying to recreate a very complex scene to illustration a specific scene that happened in last Friday's session. The players wanted to send a warning to a NPC far, far away. It wasn't D&D but I was inspired by Skywrite and I let them find a grimoire (they were in a magical university) to cast a ritual emulating that, except with risk and requirements in term of skill checks, including rare ones (Calligraphy and Danse, who does put points in those skills at character creation). So we had one of our character dancing and drawing his message in the air while scrying on their target (who was currently in an airship) in order to see clouds and write with them. I had a very specific image in my mid of what it must have looked like. I couldn't get the free D3 implementation to get it right, and I would have spent hours inpainting all the little details with SDXL.
Using chatgpt, I generated images very similar to what I had with D3 "free" but I could tell to add details. Here are the iterations of the image:
It took two tries, to get something that is very close.
I asked him for a wider view.
It's not just wider, it's another image, very close in spirit to the first one. I am rather unhappy with the sailing ship placement, because I preferred the first one, but holding the calligraphy pencil in the right hand is better.
But it looked like he was looking at the window, not a magical view, so I asked him to replace the window outline with glowing red runes.
It switched hands again, but the flying ship is better. Also, curiously failed at
red runes. It was probably the easiest.
In the end, I am quite happy as it adds precision to an already good (and free tool). It will still need inpainting because there is no way the character face will look like the PC (who is described "canonically" as resmebling a very dead French actor that ChatGPT doesn't want to recreate even in painting in what can't be mistaken as a deepfake ;-)