Back to Discussions
Help

R evaluation

一一般人
Translated to English·

I don't understand the criteria for being classified as R15 after publishing.
I am aware that I upload borderline images, so I accept the reviews and R15 classifications, but I often find myself wondering, "Why is this one designated as R15, while that similar image over there is left untouched?"
Is the R15 rating determined by AI or the administration? Or is it because someone is submitting "reports," and the images that managed to escape their notice are left as they are?

If AI is capable of detecting sensitive content and making moderation judgments, couldn't the system be designed so that R18 or R15 images can still be generated, but an R18 or R15 mark is placed in the corner of the image, and images with that mark cannot be uploaded to the gallery? That way, it would eliminate the hassle of manual moderation reviews, and if the R-rating is applied at the moment of generation, it would be much easier for us to understand "what is wrong with it?" and "what needs to be fixed to clear the guidelines."

2

6 replies

S
SNoW·

The R15 determination is done manually. In most cases, the process involves humans later finding and manually marking content that slipped through the AI's automatic review and got published.

Since it is done entirely by hand, it is inevitably limited by our capacity. The ones we haven't found yet are simply a matter of us not keeping up, which unfortunately leads to situations like "Why is this one marked R15, but that similar one over there is left alone?"

If you spot an image that you feel might be a bit too borderline to be public, it would be a huge help if you could hit the "Report" button.
Honestly, putting the "R15 judgment criteria" into 100% accurate words is quite difficult. This is because, in the end, it largely comes down to "how the person looking at the image feels." If you get a vaguely suggestive vibe, it is already close to being considered R15. That is why it is hard to achieve high accuracy even with AI—because AI doesn’t have that "sense" yet.

Common patterns include (though of course not limited to these):

・Specific poses/expressions/angles
・Close-ups of specific body parts
・Overly emphasized cleavage, or cleavage being the main focus of the image
・Bikinis that don't fit the scene

If you feel "this might be a bit risqué" at the time of posting, choosing to hold off on publishing it is also a great option :)

👍1😄1❤️1
た
たぬー·

Thank you for your hard work on the manual tasks. Thank you also for listing the common patterns. Since anime and game fans tend to have a slightly broader range of tolerance, I was worried if I might be off-track.

S
SNoW·

And one more thing, we would appreciate it if you could refrain from repeatedly posting borderline images or trying to test the limits of the AI moderation.

A site featuring diverse, safe, and high-quality content gains advertising value. In turn, this serves as the sole financial foundation that allows us to continue offering the free model, so everyone can easily try out the AI. This is meant to be a win-win situation for both parties.

While testing the limits might indeed be a bit of fun, in reality, it wastes the site's resources (for example, the cost of reviewing a single image's classification can sometimes be more than ten times the cost of generating it) and also risks putting the site itself in jeopardy.

👍1😄1❤️2
一
一般人OP·

Thank you for the detailed explanation. I understand now. I will be careful in the future.

"If you feel 'this might be a bit borderline' at the time of posting, go ahead and hold off on publishing."
"It could also lead to putting the site itself at risk."
How about displaying these two sentences at the top of the gallery, or on the page before posting to the gallery? Since it depends on individual judgment, I don't think it will eliminate the issue entirely, but I do think it will reduce the "might be a bit borderline" cases.

👍2
や
やまびこ | yamabico·

I found it very interesting to read about what the administration thinks regarding the sensitive filters!
I don't know how many posts there are in a day, but you review them all manually... I understand the feeling that you can't just leave it up to user reports... Thank you for your hard work every day!
Since I personally use the site frequently and add to the load, I want to contribute by posting "diverse, safe, and high-quality content" every day. 🙏

A site with diverse, safe, and high-quality content

While I think, "That makes sense if you want to increase the site's value!", at the same time, because of the review process itself, "diversity" might get chipped away as innocent content gets swept up alongside explicit content. I imagine finding the right balance for the reviews must be really difficult.
If users worry too much about the filter threshold and become overly hesitant to post even safe content out of concern that they might "add to your burden," that probably isn't the desired outcome either... Hmm.
Even so, I think that if guidelines on "how posts are currently being reviewed" (like the content of this post) were made somewhat explicit on the site, it might reduce friction. (Though if you make it explicit, some people might try to find loopholes around it... which would lead to a cat-and-mouse game.)

By the way, as a personal "common occurrence related to reviews":
① Post something -> it goes under review -> after some time, the review is finished before I know it, but I can't tell whether it passed or failed.
-> This happens to me a lot, so I would really appreciate it if there was something like an "Unpublished (Failed Review)" category in the generation history tab of My Profile. That way, I could look back later and easily realize, "Ah, so this one wasn't allowed." (It might also make us more aware of the rejected threshold). Or maybe send a notification for failed reviews? But some people might take that as a provocation...
② Are there times when posts are sent to review automatically because of prompts?
-> This happened to me recently: I added the word "sexual" to the prompt as a bit of spice. The generated output itself wasn't sexual at all, but when I tried to post it, it was sent to review immediately, although it ultimately passed.
I ended up feeling like, "Sorry for making you do extra work just because I used the word 'sexual' haha." But I guess that means there are checks on both the image and the prompt itself, right?

👍4
G
Giangi·
@やまびこ | yamabico

Completely agree! I have already asked for a 'rejected' (or 'not approved') section, simply because i cannot remember which picture has been refused publication and there is often no way of telling.
One funny thing is that sometimes it's the prompt generated by the AI that is labeled as unappropriate by the same AI! So one could argue that it's the filter's being too 'tight' that causes the need for so much revision and thus more work for the moderator(s)...
I too suspect that the wording of the prompt is taken into account also *after *the image's generation. Is it so?

👍1🚀1

Leave a reply

Join the community to share your thoughts and ideas.

Sign in to join the discussion