The face was first generated with StyleGAN 2 and the rest was photoshopped (including a final pass on the face)

Indeed.

This seems to be the exact source of the right half www.discoverimages.com/kiyomizu-der...

Looks like the photographer did a simple mashup of real images then did a style transfer of sorts.

Maybe AI infiltrating Shutterstock through what looks like a real photographer www.shutterstock.com/image-photo/... uploaded Dec 2024.

Based on a low resolution scan of a bust at MET museum (left). Going to release a better one soon (right).

I read it as: "please create more data that we can train our thingamajigs on"

Top 5?

Image was generated on a server somewhere in the Bay area

If technology moved to the point of making a Ghibli-acceptable conversions there should be a licensing model.

Ballpark idea: 1M for the ability to train on their content and 10% of whatever users are paying OpenAI when making them.

My stance is not that AI image generation is immoral but that Studio Ghibli should be making money from this.