False positive in deepfake detection | OpenAI | Page 1

Reported by @high mist

Bug Report: False positive in deepfake detection

`Steps to Reproduce`

First, I request a realistic image of a fictional person, and it works perfectly. It's convincing and looks real, even though it's clearly fabricated.
Then, I request more images from the same GPT (nothing shady, just a studio image from a promotional cover photoshoot) and send it the first image it generated (the one that looks real), but it can't do it.
It says that even though it (customGPT) generated that image, it can't use it for the new image because the system sees it as too realistic

`Expected Result`

The expected result is that it uses the provided image as a reference because it's not real (it was generated by the GPT).

`Actual Result`

It says that even though it (customGPT) generated that image, it can't use it for the new image because the system sees it as too realistic and triggers the deepfake protection. It's as if I were asking it to use a real person's image as a base, but that's not the case.

`Environment`

windows 11, chrome navigator

Additional Information

Please provide relevant details to help resolve the issue, such as:

ChatGPT Shared Link (if applicable).
Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

#False positive in deepfake detection