Errors occur too frequently in DALLE3

Prompting
1

Entering a PROMPT results in a fairly high frequency of errors. Approximately six to seven times out of ten.
It is not a content privacy error, but seems to be purely an internal problem.
This error is so frequent that it is not possible to generate a proper image.
The errors are of two types

「I’m sorry, but I encountered issues while trying to generate images based on your request. Please let me know if you’d like to proceed with a different request or if there’s any other way I can assist you.」
「I apologize for the inconvenience, but there were issues generating images based on your request. Please let me know if you would like to proceed differently or if there’s any other way I can assist you.」

It is not possible to generate images properly.
I don’t know the details of the error and I don’t know what to do.

At least be able to see what the error code is.

3 Likes
2

Welcome to the forum.

Are you using BING or ChatGPT Plus interface?

What is your prompt? Can you share an example? You can’t use brand names, etc. Bing is a lot more restrictive as it’s free…

3

I use ChatGPT PLUS
It is not a specific Prompt, but occurs frequently in a variety of Prompts.
It is not uncommon for a prompt that worked fine in horizontal resolution to produce errors in square or portrait resolution.

4

Gotcha. It’s still in beta, and they’re working out the problems.

If you can give an example of one that doesn’t work, let us know. Both your prompt and whatever ChatGPT changes it into (it’ll show once you click on the image…)

Sometimes it’s just the network having problems. I’ve found using during “off-hours” makes it a lot smoother.

If you search the forums, we have a few dalle3 threads…

Good luck! Hope you stick around. We’ve got a great community growing here.

5

I’m having the same issue plus it ignores specific prompts

6

Are you using Bing or ChatGPT Plus interface? What’s the exact prompt you’re trying?

7

ChatGPT plus. I can’t find the prompt but it was as daft as draw a circle but it drew a square

8

That might be too short of a prompt. What style? Animation? Painting? Photograph? Verbose is better.

Can you take a screenshot of it failing “draw a circle” as I’ve not seen it stumble too much personally.

1 Like
9

スクリーンショット 2023-10-25 114212
スクリーンショット 2023-10-25 114212919×865 66.8 KB

This is a screenshot of a failed generation.
It is a very simple prompt and simply a anime-style scene with a girl reading a book.
This prompt always fails in portrait resolution. However, it succeeds in square size.

This is just one example.
Sometimes only squares fail, sometimes all of them do not.
It is not at all certain what causes the error.

This is the problem that is occurring now.

10

You’ve got a lot of “noise” around your prompt - ie the JSON … I’d strip that and just use the prompt.

11

This is not noise.
The Prompt is formed by ChatGPT into JSON data containing size, Prompt and seed, and passed to DALLE3.
This is a way to deliver the Prompt, Seed, and Size directly to DALLE3 without the intermediary of ChatGPT.
The format of the JSON data can be obtained from ChatGPT.

12

The other thing is likely the mandarin (?) characters instead of English.

Those extra JSON characters like { and " etc will effect the prompt/output, though. Try it without and see…

13

JSON data is created in English only.
However, my ChatGPT UI is set to my country’s language settings, so ChatGPT replies basically in my native language.
The languages other than English on this screen are from ChatGPT.

14

He is likely using a trick mentioned in another post where he can set seeds and similar.

However, this method is not official and maybe that’s why he sees intermittent errors.

15

It’s perfectly valid JSON, I also write my prompts like this (and check them before in JSON lint).

image
image648×872 31 KB

This way you can perfectly address the image generator what you want and expect.

It’s a multi-dimensional JSON array, that’s why the [ and { are there. Also a numeric is formatted different (no single quotes) than non-digits (encapsulated by quotes).

It also comes in handy when doing iterations of existing images (once you did save and validate the JSON data).

{
  "size": "pixels x pixels",
  "prompts": [
    "Text 1",
    "Text 2",
    "Text 3",
    "Text 4"
  ],
  "seeds": [1, 2, 3, 4]
}

Basically, it’s the recipe for every image you feed and fetch.

When you create a .git repository for those JSON-data, you can not only track every change of an image (at pixel level) but also the commands that forced those changes.