Could GPT action access URL of Dall-E generated image?

GPT builders
1

Hey folks,

I am building a GPT that includes Dall-e image generation .

Once you generated an image I want to pass this image to one of our action, so it gets uploaded to our servers.

For that I need to get access to the full absolute URL.
But I can’t make this work.

First it’s passing the URL as relative temporary URL

"params": {
    "url": "/mnt/data/{filename describing the picture}.png"
  }

Then I ask to use an absolute URL and it hallucinates

Or when prompted to use the files.openaiusercontent.com

it gives
https://files.openaiusercontent.com/file-0t1Oe7IwL4iCbyIJlz8NC0Mt/A_friendly_panda_living_in_Manhattan,_strolling_do.png

Which does not exist.

So my question is simple: what is the way to communicate between Dall-e and action in a GPT to get access to the URL of a generate image.
Thank you.

2

ChatGPT does not receive any image file name or location information in the DALL-E generation metadata which it receives as a report of success. There is therefore no followup method for image access and analysis except by user upload.

(edit) It does have a mount point file that python can operate on.

3

thank you @_j for your answer.

That confirms what I have found.
I also asked to provide the instruction it had and it answered back

The current guidelines I follow restrict me from providing direct URLs to images generated through Dall-E. When an image is created, it is displayed within our conversation interface, and you can download it directly by clicking on the image. This approach ensures a seamless and integrated user experience within the chat, without the need for external links. If you have any other requests or need assistance with different tasks, feel free to let me know!

That’s so unfortunate and limiting of all the things custom GPT could potentially achieve.

4

Even programming the API would take careful workings around expected patterns. Given:

  • the image API can return either URL or base64 image;
  • image URL expires after an hour;
  • the chat API can only accept images for vision as user role message attachments, either URL or base64 image.

The best vision results would be preprocessing and controlling the image inputs to chat completions API, by resizing and vision parameter selection.

That basically leaves programming a chat that would have several calls required to make for a request “generate an image, then analyze the contents”. We must work within model training.

provide:

  • image generation function, returning database ID
  • image analysis function, invoked by database ID query, connected to custom pattern
  • database function, polling for non-conversation directory contents
  • database that contains ID, description, prompt, rewritten prompt, image, …

method:

  • user input
  • image generation function emitted
  • return value with database item number for image, base64 stored
  • image displayed
  • image analysis function emitted
  • tool call return confirming user will supply image
  • user message injection “here’s the image to analyze as discussed” with base64
  • AI response (if it doesn’t want more functions)
5

were you ever able to figure this out? i have the same question: https://community.openai.com/t/how-do-i-get-a-public-link-for-a-dall-e-generated-image/

with the code-interpreter, could you pass the image file via a POST request? i’d obviously prefer for my server to take in an image URL instead of an image file, but at this point i’d be willing to consider either.

(i have my own api that is doing further processing on the dall-e generated image.)

6

This approach might help maybe? Generated Files/Images and GPT Actions - #12 by coderjun