You mention a prohibition on photorealistic faces, but none of these terms appear to say anything about this.
There is the prohibition “Do not upload images of people without their consent”, but this appears to be bound to the matter of actually-existing humans whose consent could be involved (and notably isn’t bound to what style actually-existing humans are depicted in, whether that’s photorealistic or otherwise).
DALL-E 2′s main page does confirm that measures were taken to prevent the AI from making “photorealistic generations of real individuals’ faces”—but this again seems to be specifically about actually-existing humans.
The guidance was in a google document they sent me in the email approving my access, which I think used to be the same as the document linked to in the “sharing publication” guidelines, but apparently now isn’t?
You mention a prohibition on photorealistic faces, but none of these terms appear to say anything about this.
There is the prohibition “Do not upload images of people without their consent”, but this appears to be bound to the matter of actually-existing humans whose consent could be involved (and notably isn’t bound to what style actually-existing humans are depicted in, whether that’s photorealistic or otherwise).
DALL-E 2′s main page does confirm that measures were taken to prevent the AI from making “photorealistic generations of real individuals’ faces”—but this again seems to be specifically about actually-existing humans.
Is this guidance given anywhere specifically?
The guidance was in a google document they sent me in the email approving my access, which I think used to be the same as the document linked to in the “sharing publication” guidelines, but apparently now isn’t?