V2 Restricted Content Filter Suggestions
Michael
A few things would be EXTREMELY helpful when allegedly restricted content is generated in Version 2.
1) One is for the error to describe what content was generated or prompt criteria was used that didn't pass the filter.
Currently it tells us to review the input and try again, but review for what? How can we fix an issue that isn't addressed?
It would be like if every feature request you got just said "Please fix issue" and then charged you for the suggestion. You wouldn't know what to do, but you would sure know you were losing money for the experience. I bet you would find that very frustrating.
Basically, the AI filter found something objectionable. The AI knows what it was. The AI has the capability to put it into words. It could translate that into a description of which part of the image failed which criteria so we would know what to refine.
It would at least make people less suspicious that it is failing content arbitrarily just to steal credits. I've dabbled with other services that credit the user back when restricted content is generated because they don't automatically assume the user is being malicious.
More than half the times I've gotten a restricted content notificationI've literally changed nothing in the prompt (or changed one word like from "replicate" to "duplicate") and had it come back with a valid result. This tells me either the filter is too strong, the AI goes off-script way too often, or the "restricted content" error is being applied completely arbitrarily.
Knowing at least what criteria it found offensive would help identify what the problem is or what needs to be fixed on the user's end.
2) Leave the "restricted content" error in the gallery until the user deletes it. Right now it stays up for two seconds and just disappears and I can't even read the message.
If I look away from the computer for five seconds I don't even realize what just happened; sometimes I think my request didn't send and resend it, wasting multiple credits.
Leaving the error in the gallery would be most helpful if error descriptions are implemented.
3) Of course crediting the user back, even if with a limit that could vary with each tier, would show the end users a gesture of goodwill. I'm pretty sure most of the users are aware of the community guidelines (or if not you could provide a side panel link to them) and aren't actively trying to violate them because that would be pointless.
Michael
Also sorry for the tone of the original request and the comments. I wrote it while extremely frustrated with the entire issue which I shouldn't have done.
I really want the criticism to be constructive and suggestions to be helpful and I hope this rewrite better conveys that.
It's a good site with just the right blend of functionality and simplicity for my taste and I've always gotten the impression that the creative team is very respectful and passionate about the site and the user base.
Michael
An example of the filter being too strict is having it block the result when I ask to duplicate a reference image in a different style (such as oil paint or pixel art or animated) and the reference image is a headshot that contains bare shoulders above the clavical.
The person might be naked; they might be wearing a tube top. It's irrelevant since the reference image doesn't show it and I'm not prompting to outpaint.
The point is that shoulders alone should not qualify as inappropriate any more than a face or a an arm would.
The content filter is also pretty inconsistent in certain circumstances. Same reference image as mentioned above redone as Anime is fine but pixel art is a violation, but also sometimes it's acceptable it's really a coin-toss. Oil painting is fine, but as an action figure 𝘩𝘰𝘸 𝘥𝘢𝘳𝘦 𝘺𝘰𝘶.
(The reference image is just a previously rendered AI image from this same website so it's not any actual living person whos privacy could be violated by the content. I'm just using it with some others to experiment with V2's capabilities.)
I guess what I'm suggesting is to narrow the scope of the content filter or refrain from charging so many credits for erroneous content restrictions until the filter you're using is more reliable.
And also to let us know what we theoretically violated and to leave the notification active long enough for anyone who is not a supercomputer to read it, to reiterate the original post.