[throwaway9274]

Everybody’s too concerned about copyright for training data, no one’s concerned enough that the outputs are considered non-copyrightable.

$50mm pre-product valuations for startups producing 3D assets for games via prompting a foundation model.

Zero due diligence on the U.S. copyright office’s stance that every asset those companies generate will be open to reuse.

[circuit10]

I assume any tweaks to the assets and the way they’re put together will still be copyrightable though

[throwaway9274]

Maybe, depends on the threshold for human creativity, which is still in flux.

You could copyright the underlying character maybe. But its a major headache for these companies and a symptom of the broader fact that copyright needs legislative reform, not attempting to shoehorn AI into the current law.