No Face Variability for DALL-E 3? Who is this woman?

Yes, it was late, and I had a long night. I wanted to reply to someone about the use of paid services and assistants and such. And then I emptied my soul.

This post:
"…So I wonder if in ChatGPT Plus, where you payers are, you start a new conversation (so it doesn’t have the same ability to keep building variations by image name on chat history of what you produced before), if you can get a different direction with additional descriptions of a different scene or style description, framing, etc. Not just for the new setting, but just to stir up the prompt.

Perhaps this new tuning is to address those that couldn’t get AI to create any consistent style for book illustrations, for example.

(Sorry if I don’t have any great ideas of what to make an AI produce…)"

So, I had the initial post, but I asked the bot to make the text somewhat less negative, resulting in two attempts of “mmm, no, not good.”

But apparently, the last version was submitted by me. It does describe, at a high level, my findings and problems with OpenAI’s API, policies, prompt altering, and whatnot.
If you want the original post, no problem, and if you think I am wrong, please tell me where and with what?

  • API = expensive
  • ChatGPT+ subscription = confusing
    OpenAI is allowing less copyright infringement, but they decide before the copyright is infringed that they will not give the user the choice to decide if the outcome is too identical.
  • Prompts are rewritten with now all kinds of extra inclusive elements.
  • Prompts are altered, and the new one is not better but uses different words, I guess to avoid people getting triggered.
  • I use the assistant, but you cannot use it with one thread as it will send your entire history of a different conversation. It should not happen, says OpenAI, but it does… i am flooded with old messages…

I think now I have explained better what some of the caveats are.


Let me know if there’s anything else you’d like to adjust or any other assistance you need! ( Yes, i asked the Bot again to spellcjheck )

There is no “ChatGPT Plus API” - it’s either ChatGPT Plus, where you pay the monthly subscription and talk to the chatbot about making images, or “OpenAI API” without the name ChatGPT, where you can send requests directly to the API image models (although DALL-E-3 still has an AI that rewrites prompts).

The symptom in Bing is likely because they used the same AI rewrite and embellishment to generate all four images in the later version, while for a short time, they might have been sending the prompts directly to DALL-E-2 and 3 and you get a wide variety from ambiguous short inputs.

Now with techniques to overcome the rewriting and have proof the text was sent to the image model as desired, we can again explore the variations that come about from short ambiguous prompts, like the “When cats invade!” prompt that previously gave all the same style cartoon.

and again see enjoyable diversity, and in ChatGPT, the ability to re-use the random seed of the most art-looking one (although with such a simple prompt, a transformation “When dogs invade” doesn’t bring about the style)


To wrap up the thread, let’s send the initial “who is this woman” prompt, but to DALL-E-3 on API, using enough prompt to have it passed unaltered, using natural API parameter, requesting a tall image (which is not so reliable without changing the prompt to reinforce), and also using the “HD” option for another AI sharpening pass.

Result 1: Pay $0.12 for an API timeout.

Result 2: Enjoy your natural woman that I had to rotate

Kudoos ,

Using the ‘cheapest’ settings you got a woman that has indeed the specifications that were requested…

i will actually use your settings in my app. let me check what happens

The setting was
size : 1024*1024
style: natural
quality: hd

but afteer that it reset to the well known girl face. tried it via copilot but only too perfect.

The setting was
size : 1024*1024
style: natural
quality: hd

The setting was
size : 1024*1024
style: vivid
quality: hd

and

The setting was
size : 1024*1024
style: vivid
quality: hd

then

The setting was
size : 1024*1024
style: natural
quality: standard

tried Dalle-2 just to check

The setting was
size : 1024*1024
and tried 3 images

so i was alsoi able to get closer to the prompt