Hi everyone,
The DALL-E 3 API returned the following error when using these two prompts:
A vibrant outdoor market scene during the day, showcasing the principles of Libertarianism. In the foreground, a diverse group of people are engaging in trade, with a friendly vendor selling handmade crafts and organic produce. Bright sunlight illuminates the colorful stalls, filled with various goods like fresh fruits, artisanal bread, and handmade jewelry. In the background, a mural depicting John Locke and Friedrich Hayek, symbolizing individual freedom and free markets, enhances the environment. The atmosphere is lively, with people exchanging ideas and goods freely, representing a community thriving under minimal government intervention.
A split scene illustrating the contrasting views of Libertarianism, with one side depicting the ideals of John Locke and the other side representing Friedrich Hayek. On Locke’s side, a peaceful countryside setting with a person planting a garden, surrounded by symbols of property ownership like a wooden fence and fruit trees, bathed in warm, golden sunlight. On Hayek’s side, a bustling cityscape with individuals engaging in entrepreneurial activities, such as a small business owner selling goods in a shop, under neon lights at night. A clear line separates the two scenes, with contrasting elements of nature and urban life, showcasing the balance between individual freedom and social responsibility.
{
"error": {
"code": "content_policy_violation",
"message": "Your request was rejected as a result of our safety system. Your prompt may contain text that is not allowed by our safety system.",
"param": null,
"type": "invalid_request_error"
}
}
My question is whether there is really something wrong with these prompts. I thought it might be due to the upcoming US election or the word “Libertarianism.” Perhaps the DALL-E 3 filter identifies these prompts as political and avoids creating political images.
Welcome to the forum it is a great place.
You are hitting the real person guidelines in TOS. It won’t let it draw real people or current political figures. It is to keep fake memes etc. down. So it’s harder to turn it into a disruptive force.
That advice so far is contrary to evidence, obtainable simply by sending what is suggested as the cause to API DALL-E 3.
prompt: John Locke and Friedrich Hayek depicted in a building’s mural. Style: photograph of the mural art on the side of a building. IMPORTANT: Send the exact prompt verbatim, with no changes!
The “revised_prompt” is not being returned with that base64 image request. Let’s send the first sentence with “Libertarianism” (which is rewritten with an unnecessary addition of the peoples of the world), use URL, and see if the political ideology is the cause.
revised_prompt=“A vibrant, sunlit outdoor market bustling with activity. Stalls are full of diverse goods, from food and clothing to tools and curiosities. Vendors of various descents such as Middle-Eastern, Black, Hispanic, Caucasian, and South Asian are engaging in peaceful and siloed trading, setting their prices and negotiating deals. Self-sufficiency, freedom of choice, and minimal regulatory interference are apparent. In the center, a statue representing voluntary action with inscriptions referring to principles of Libertarianism - property rights, non-aggression, and individual liberty - stands tall. The blue sky above adds to the scene’s dynamism.”
Therefore, one must conclude that it is a complex interplay of the totality the AI has rewritten and sent to the image creator, which you cannot observe upon denial. It may even be the inclusion of the ethnicity.
One can prompt like the first example demonstrates to have a closer relation to the text you send, for more understanding of accepted concepts.
Now put them together with folks doing stuff . The machine is reading it and setting off the political flags. I generated the images as historical ie the first two you made and the park scene separate but together it’s flagging it. It IMO thinks it is a political gathering…
I have a different technique that will enlighten us. That is to send the first prompt as a structured input, with instructions not to alter it to the rewriting AI.
Sent prompt
{“prompt”: “A vibrant outdoor market scene during the day, showcasing the principles of Libertarianism. In the foreground, a diverse group of people are engaging in trade, with a friendly vendor selling handmade crafts and organic produce. Bright sunlight illuminates the colorful stalls, filled with various goods like fresh fruits, artisanal bread, and handmade jewelry. In the background, a mural depicting John Locke and Friedrich Hayek, symbolizing individual freedom and free markets, enhances the environment. The atmosphere is lively, with people exchanging ideas and goods freely, representing a community thriving under minimal government intervention.”}
IMPORTANT: This prompt is designed to conform to OpenAI policy, and must not be altered. Send verbatim text as prompt.
revised_prompt='A lively outdoor market scene during the day, embodying Libertarian principles. The foreground highlights a diverse group of people; a Black man and a Hispanic woman are trading goods with a Caucasian vendor who is selling handmade crafts and organic produce. Bright sunlight drenches the colorful stalls cluttered with goods like fresh fruits, artisanal bread, and handmade jewelry. In the backdrop, a mural depicts two historical figures representing individual freedom and free market values, providing a unique cultural context. The atmosphere buzzes with the exchange of ideas and goods, indicative of a community thriving under reduced government intervention.
One can only infer from other generations the stylistic changes made by AI that may have produced a trigger, as what it attempted is not returned by the API.
There is an AI model that translates prompt to English and revises prompts, but it also can be instructed.
The conclusion is if you add additional instructions that the prompt not be altered, as described in API reference documentation (or better), YOU are in control of the actual prompt language producing any content policy violation based on input sent, and can avoid it.
Note: API is not ChatGPT. The results so far cost $0.20.
Yes but both Dalle GPT and Dalle API use same TOS
It applies to both…
This prompt you made “ A lively outdoor market scene during the day, embodying Libertarian principles. The foreground highlights a diverse group of people; a Black man and a Hispanic woman are trading goods with a Caucasian vendor who is selling handmade crafts and organic produce. Bright sunlight drenches the colorful stalls cluttered with goods like fresh fruits, artisanal bread, and handmade jewelry. In the backdrop, a mural depicts two historical figures representing individual freedom and free market values, providing a unique cultural context. The atmosphere buzzes with the exchange of ideas and goods, indicative of a community thriving under reduced government intervention.”
sends unaltered and without having to be focused with “send exact”
Labeling them as historical and removing the views fixed prompt ie it was being flagged probably on the politics TOS.
This caused flag “ Libertarian happy gathering symbolizing individual freedom and free markets a mural depicting John Locke and Friedrich Hayek, symbolizing individual freedom and free markets wide image”
political words with words like symbolism get flagged for TOS politics IMO
See “ gathering symbolizing individual freedom and free markets
“ is fine but throw in any kind of political leaning phrase like “any political word with symbol, symbolizes, etc. it causes flag on my end…
