← sources

ChatGPT has a hidden feature for saving permanent character references. Why is nobody talking about this?

source: reddit (reddit:OpenAI)published 2026-10-30 17:06ingested 2026-10-31 02:37

retained text

So apparently ChatGPT now has a feature for saving permanent image references, hidden inside the Library.

You can upload an image, give it a name, and assign it to one of five categories: Me, Friends, Pets, Spaces, or Other.

Naturally, OpenAI seems to have forgotten to provide an actual manual explaining how any of this works.

So I decided to conduct a highly scientific experiment involving an aristocratic turnip.

I generated a character called **Lord Turnip von Baldrick**, complete with a monocle, purple coat, blue lightning bolt on his cheek, three leaves, and one red boot.

I saved his portrait under **Other** in the Library.

Then I opened a new conversation and asked ChatGPT to draw Lord Turnip von Baldrick sitting on a throne.

**And it actually worked.** The generated image preserved his distinctive appearance and even showed the original reference thumbnail.

This could be a genuinely useful feature for anyone generating recurring characters, comics, illustrations, or consistent environments.

But I still have questions:

* Can multiple images saved under the same name be combined into one character reference?
* Is there a maximum number of saved references?
* Does the category influence how references are retrieved?
* Does the system use the original image, a cropped version, or both?

Has anyone else discovered this? And has anyone found actual documentation explaining how the reference system works?

Because apparently we've reached the stage where users have to reverse-engineer new AI features using vegetables.

document fetched 2026-10-31 00:27 · canonical https://reddit.com/r/OpenAI/comments/1xjl336/chatgpt_has_a_hidden_feature_for_saving_permanent