[codazoda]

This is not at all surprising to me.

Just this week I asked for a picture of a cartoon Car. It produced an image so similar to Pixar Cars that I was surprised. I was hoping for something a bit more creative. I asked the AI a few follow-up questions about the first use of the windshield for eyes. That might not be a copyrightable thing but Pixar Cars have a certain look to them and these tools seem to produce a very similar look.

I didn't read the article due to the paywall.

[elpocko]

When you simply ask for "a picture of a cartoon car" and nothing else, these models will give you a bland and generic depiction of a cartoon car, something like the lowest common denominator among cartoon car images according to their training data. Pixar's Cars is a popular depiction of cartoon cars.

If you want something more sophisticated and creative then you have to be the source of creativity, you have to describe details of the cartoon car, the setting, and the style, whatever you can think of to describe the thing that you actually want. "Picture of a car" doesn't cut it. If you can't describe it in words, then you make a scribble instead (I don't actually know if Midjourney supports this, I only use ComfyUI + local models and tools).

Most people don't know how to use these tools properly and they don't care, all they want and all they know is a "Make image" button. More sophisticated users (dare I say "artists"?) use it like a renderer for their ideas, sometimes literally integrated into Blender or other creative software.

[Kim_Bruning]

How about "Cartoon car, not Pixar" for instance.

I think it's important for AI to learn from sources; just to know what not to do as well.

And I suppose it's legal to explore a visual space for your own personal use as well.

Actually publishing copyright infringing materials is a different story. Not sure you should blame the tool though.