Hi there! I am trying to make an Agent that creates storyboards (images) from a text prompt, by splitting this prompt on \n and creating an image for each paragraph
I have tested this pipeline in a notebook and it works fine.
I am trying to integrate it into an Agent and getting the error below on the 2nd call. Is there a block on making Dall-e-3 calls through an agent tool?
I have pasted the prompt below and I am pretty sure the content is ok.
ERROR
openai.BadRequestError: Error code: 400 - {‘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’: None, ‘type’: ‘invalid_request_error’}}
PROMPT
The heaviest rains during the wettest 90 days get worse with an addition of 32mm rainfall, affecting the city’s famously spectacular outdoor sumo tournaments and the massive seasonal parades. The already hot, humid summers will have 10 more scorching, dry days which could impact the outdoor tea ceremonies Nagoya residents relish.
I tried typing your paragraph directly into DALL-E and it came back with following response.
It seems like you’re describing a scenario where climate change is impacting traditional events and activities in a specific city. The increase in rainfall and the number of scorching, dry days could significantly affect outdoor cultural practices like sumo tournaments, parades, and tea ceremonies. These changes emphasize the importance of considering and adapting to the effects of climate change in maintaining cultural heritage and traditions. If you need further information or assistance on this topic, feel free to ask!
I believe the error message you received is a general message that GPT throws up when it does know what to do with your prompt. You will need to experiment but I found that by adding the following phrase to your prompt will likely get a better response. Good luck.
Create a single storyboard image that best represents the following…[your paragraph]
The API doesn’t talk to you. ChatGPT also doesn’t get the ultimate reason for content denial - it just hallucinates if pressed.
For the OP, the exact prompt language shown needs the rewriting of the AI. It is likely what the AI wrote for you that was being flagged content violation, which can be a denial as diverse as named depiction of races or inclusion of trademark - or can be the visuals within.
The prompt that is a “story” doesn’t describe an image. Here’s what is typical if you send the language for unaltered prompting:
However, we can prompt the DALL-E 3 API, knowing that we are talking to an intermediary language AI that can do what we want:
Create a vibrant description of imagery that will accompany this segment of a story.
Scene = [“The heaviest rains during the wettest 90 days get worse with an addition of 32mm rainfall, affecting the city’s famously spectacular outdoor sumo tournaments and the massive seasonal parades. The already hot, humid summers will have 10 more scorching, dry days which could impact the outdoor tea ceremonies Nagoya residents relish.”]
The AI wrote that for us as:
Depict a drenched city, the excessive rain making the scenery colorful and vibrant, impacting the grandeur of an outdoor sumo tournament. Showcase enlarged sumo wrestlers engaged in a fierce fight, with spectators huddled under raincoats and umbrellas. Also, incorporate a massive parade in the backdrop with floats and performers that additional showers are affecting. Transition the scene towards a disarmingly hot and dry summer. Place Nagoya residents in the shade under the lush trees partaking in an outdoor tea ceremony, their sweat glistening, and their expressions showing displeasure at the unexpected heat.
Oh hey! thanks for weighing in everybody. The solution was better pre-prompting so that Dall E knew what to do each time to return an image completion. The suggestions in this thread were helpful and appreciated. The error has been resolved.
