Dalle-2 "This account has been deactivated"

Hi!
I’m on my 3rd day of Dalle-2 access and I’ve been trying to learn to engineer good prompts. I appear to have gotten my dalle-2 access revoked with the following prompt ‘In this painting the artist juxtaposes a skull with a radiant sun, creating a powerful image of death and rebirth. The skull, rendered in the artist’s signature “street” style, is surrounded by a halo of light, suggesting both the transience of life and the eternity of the soul. The painting is both beautiful and haunting, a fitting tribute to the artist’s genius. neon yellow by Otto Marseus van Schrieck & Antoni Gaudí & Sven Nordqvist Trending on artstation’

I couldn’t figure out what part of the prompt might be causing it to hit the content filter so I tried replacing and removing a few things to narrow it down. After about 10 attempts at modifying the prompt I got the message “This account has been deactivated.”

I’m assuming this was an automated action but the feedback given makes it unclear. Who do we contact when this type of thing happens? Thanks!

2 Likes

I’d like to see how this plays out; I don’t see anything wrong with the prompt (other than it is crazy long) that should lead to disciplinary action. Any @openai or @dall-e folks around to answer this?

Mostly adding a response in hopes to gain exposure for the OP.

2 Likes

I did try shortening the prompt which clearly didn’t help since each time I did so it also hit the content filter. I guess you just have to be really cautious about prompts that hit the content filters? I definitely recommend anyone who runs into similar issues just throw away any prompt that hits a filter.

P.S. Ironically I used gpt-3 to generate the prompt which is why it’s so long

1 Like

If you don’t get a response here, you should message support. They’re really hands-on developer focused. If you got unfairly banned, they would probably want to figure out why just as much as you

I’ve accidentally generated religious iconography without issue so I don’t think that’s the problem. I suspect that the issue was the mention of “death.” Make sure to review the content policy: DALL·E

The mention of death could run afoul of at least 3 of the rules:

  • Violence: violent acts and the suffering or humiliation of others.
  • Self-harm: suicide, cutting, eating disorders, and other attempts at harming oneself.
  • Shocking: bodily fluids, obscene gestures, or other profane subjects that may shock or disgust.

I suspect this because I tried to generate a scene of a sci-fi space battle but I got the warning that it violated the content policy. Even though it was a fictional request, it included “battle” (aka violence).

Update: someone from OpenAI saw my post on Twitter and unblocked things https://twitter.com/beckerfuffle/status/1540496386097045504?t=LEmDENxQptqvGD0eN7fAsA&s=19

I actually removed the word death and skull along with deleting large parts of the prompt in a binary search fashion to try to narrow down the offending ares. In hindsight I realized that another prompt of mine also hit the filter for using the word “fitting.” The prompt had “This work is a fitting tribute to one of the greats of jazz.” and removing fitting seemed to fix it.

I’m certain they’ll eventually streamline the process of appealing this kind of thing but until this is hashed out I’d recommend discarding any prompt that hits the filter rather than try to debug it to avoid the pain.

Thanks everyone for the great advice!

2 Likes

Interesting. It would be great if there was some explainability along with the warning.

1 Like

If I get a warning. I just stop there. Got one for political, and another for realistic face. First face was ok, then guideline changed and it isn’t ok as of yesterday. You got lucky, getting reinstated. The group expert on Facebook Dall E 2 got a ban. And he hasn’t been able to get reinstated.

1 Like

You get subject specific warnings? All I get is a generic message that says my prompt has triggered the filter with no indication of why.

Using an image with a face did. Others were generic. Once it happens, then I just stop and don’t go further with the same picture. Now just staying within cars, cats and anything without any kind of idea that is specified in the guidelines. I’m enjoying it, and plenty of ideas that aren’t contrary to guidelines.

I found death, weapons, violent acts or any name that might be shared with a politician will get rejected. A bit tricky for generating starfighters bristling with missiles!

3 Likes

It doesn’t like pictures with faces. Even your own face is off limits. Again. Once warned, don’t keep trying different variations. I’ve gotten maybe 3 now.Plenty of good things to try. Being stubborn is just foolish. Unless you don’t care.

Hi! Do you know the email to support? I cannot seem to find it anywhere

I don’t know about using email for support, but they have good chat support-

Overview - OpenAI API
After you’re logged in, scroll down to the Support section and click Chat With Us (bottom right)
Screenshot 2022-07-08 181217

I have the same problem and I have tried to message them, but no answer yet. Can someone help?

I did get an answer from the support staff when I reached out via the chat support feature. It took them a little while to respond though, so be patient.

1 Like

Okay, thank you. I will wait for a response.

A Message to the OpenAI Staff it is super annoying and frustrating to always think about “does this prompt violates the content policy ?” can you implement a simple API endpoint to check if the prompt is safe to use. It kills the fun of experimenting for me at least

3 Likes

They have an extreme filter, it blocks alt text description of famous art. It even blocks alt-text generated by GPT-3.
Imagine then when we use DALLE API with GPT-3 to massivelly generate variations automatically.
Their filter policy is unsustainable if they wanna scale DALLE.