Re:Stable Diffusion Textual Inversion
by karmasimida on Monday August 29, 2022 @09:54PM
(#32645811)
Well Google's paper showed you don't need CLIP anyway. T5 and other languages model can be used regardless.
CLIP isn't the true blocker to entry, the dataset and compute is.
StableDiffusion has an open dataset, was funded by one guy, and apparently took "much less than $600k" to train on AWS (https://twitter.com/EMostaque/status/1563965366061211660).
So it seems there actually aren't many barriers to entry at all. There's certainly a lot of legal questions, but if it's this easy to create your own model then it's hard to enforce anything…
I don't think there are actually major legal concerns. Copyright protects reproduction of a specific image. Looking at an image and producing something in a similar style is not copyright infringement, it's called being an artist. The law on this seems pretty clear.
The UK recently announced plans to make this completely explicit, to remove any remaining doubt: "For text and data mining, we plan to introduce a new copyright and database exception which allows TDM for any purpose. Rights holders will still have safeguards to protect their content, including a requirement for lawful access."
https://www.gov.uk/government/consultations/artificial-intel...
Not only that, but OpenAI didn't seem to know their CLIP model could be used to generate images (via Advad's CLIP+VQGAN) at all, otherwise they wouldn't have released it. So they did unintentionally start the "AI art" movement even if they didn't release a trained DALLE.