Back to Discussions
Help

Tightening regulations on generation?

ウウドンエ
Translated to English·

I feel like the detection of sensitive images has become stricter since around yesterday. Has the judgment become more severe due to an AI version update?
It might be partly because my favorite character is lightly dressed, but even the original image that I finally managed to generate has now been flagged as sensitive, and I can no longer generate it.
I was hoping to use this service for a long time, so I find this a bit disappointing.

14 replies

さ
さな_全人類ケモナー化計画·

It seems like even a bikini gets flagged as sensitive. Since they already claimed that we can generate up to R15-level content, I really wish they would keep their promise on that.

👍3
S
SNoW·

Let me clarify the situation. Are we talking about image generation or video generation?

Over the past few days, we have indeed made some adjustments to video generation, changing some of the pre-generation review models to unify them with the ones used for image generation. As for image generation, I don't think we have really touched anything recently.

さ
さな_全人類ケモナー化計画·

This is about image generation. I've only checked WAI, but I feel like they've become extremely strict about skin exposure compared to before.

ウ
ウドンエOP·

I'm on the video generation side.
I'm disappointed that images I used to be able to generate are now being flagged as sensitive.
I suppose the issue is that the characters wear revealing outfits, but even swimsuits and dancer-like costumes seem to get flagged when they are anime characters.
Also, characters with very large breasts—as is common in anime—mostly ended up being rejected, even when they were fully clothed.

ウ
ウドンエOP·

P.S.
When I hand-drew tears onto a successfully generated image to create a crying-laughing expression, it was flagged as sensitive.
If even expression variations made through manual retouching are deemed unacceptable, that is honestly too harsh.
I won’t go so far as to ask to revert to the previous version, but I do wish the restrictions were relaxed a bit more.

S
SNoW·

Thank you for the feedback! It might be that the AI is prone to false positives with anime-style images. I'll check into it and try to make some adjustments.

ウ
ウドンエOP·
@SNoW

There are anime characters, like women in swimsuits or those with somewhat disheveled clothing showing skin that you often see in image posts, but when you try to turn them into videos, they get flagged as sensitive at the initial judgment stage.
Common recent trends, such as women with large breasts or female anime characters in eccentric outfits, are almost always flagged as sensitive, meaning nothing can be generated. The range of what we can do has decreased dramatically over the past few days.
Generating here was truly fun and wonderful, so I sincerely ask for a better review of this.

香
香音·

I guess AI is still in its developmental stage right now. Naturally, it is sad that it cannot create with the consciousness of it being "art," but I suppose that can't be helped. I would be happy if the regulations were a bit more relaxed for the paid versions, but it doesn't seem to be that simple of an issue, does it?

こ
こばと🕊️_ダヴ(DOVE)·

If you look at the gallery, you'll see that a video of a completely ordinary woman just walking has been sent for review. Yet, while simply putting a cardboard box over the woman's face gets it instantly displayed in the gallery, an image of Little Red Riding Hood and the grandmother is sent for review. How do you explain this? Is it violent? Is it obscene? This is proof that the AI is not working. It is completely useless.

た
たかはしがに·

It depends on whether you want to post to the gallery or just enjoy it privately, but here are a few little tips.
I don't really want to teach this so openly right in front of the admin (since they might patch it), but...

Even if it gets flagged as sensitive once, don't give up—if you try a few more times (without changing the prompt), you might succeed.
There is probably something like random number generation that causes high and low rolls. Also, the order of the prompts is important, and the results differ depending on whether they are in English, Kanji, Hiragana, or Katakana, or even if they are separated by "、 (Japanese comma)" or ", (comma)". It is a remarkably deep system.

Even if it doesn't pass the gallery review, you can still generate images with wardrobe malfunctions depending on the prompt. You can assume that 99% of dirty words in English won't go through. Japanese is incredibly convenient.
While "oppai" is completely blocked, "o ppai (with a space)" might go through. For Kanji like "胸 (breasts)" and "乳 (boobs)", the prompt order is crucial, and they get blocked sometimes and go through other times. But since that's a hassle, using hiragana like "pai" or "chichi" will work. For size, "big chichi" is probably a must. If you want to see a slip, first you should... (self-censored). Japanese really is wonderful.

I don't really want to show off these kinds of techniques since they might go against public order and morals, but...
As a fellow comrade fighting against this site's sensitive filters, I wanted to help you out a bit. It used to be unfiltered, though...

👍2👎1👀1
ウ
ウドンエOP·

Advice? Thank you.
Well, nowadays, outfits unique to anime characters and breasts as big as their faces all get caught, so it's a problem even before trying...

た
たぬー·

I can’t help but think it might also depend on the video model. If the progress bar completes up to the "base image check -> prompt check" stage, I feel like it's probably failing at the frame generation step. I wonder if all the generation errors have the exact same wording.

·

I've returned after a few months, but the images that used to be possible to create can now barely be created at all. Even for completely private, personal use, it is not allowed. I understand that it can't be helped. Please do your best, management, so that the number of users doesn't decrease.

日
日本國を応援する会·

In conclusion, it's better not to leave everything to AI, right?

Leave a reply

Join the community to share your thoughts and ideas.

Sign in to join the discussion