[meander_water]

Not foolproof, but a couple of easy ways to verify if images were AI generated:

- OpenAI uses the C2PA standard [0] to add provenance metadata to images, which you can check [1]

- Gemini uses SynthId [2] and adds a watermark to the image. The watermark can be removed, but SynthId cannot as it is part of the image. SynthId is used to watermark text as well, and code is open-source [3]

[0] https://help.openai.com/en/articles/8912793-c2pa-in-chatgpt-...

[1] https://verify.contentauthenticity.org/

[2] https://deepmind.google/models/synthid/

[3] https://github.com/google-deepmind/synthid-text

[adrian17]

I just went to a random OpenAI blog post ("The new ChatGPT Images is here"), right-click saved one of the images (the one from "Text rendering" section), and pasted it to your [1] link - no metadata.

I know the metadata is probably easy to strip, maybe even accidentally, but their own promotional content not having it doesn't inspire confidence.

[meander_water]

Yeah that's been my experience as well. I think most uploads strip the metadata unfortunately

[danielbln]

Synth id can be removed, run it through an image 2 image model with a reasonably high denoising value or add artificial noise and use another model to denoise and voila. It's effort that probably most aren't doing, but it's certainly possible.

[cubefox]

> The watermark can be removed, but SynthId cannot as it is part of the image.

That's not quite right. SynthID is a digital watermark, so it's hard to remove, while metadata can be easily removed.

[hayinneedles]

Reminder that provenance exists to prove something as REAL, not to prove something is fake.

AI content outnumbers Real content. We are not going to decide if every single thing is real or not. C2PA is about labeling the gold in a way the dirt can't fake. A Photo with it can be considered real and used in an encyclopedia or sent court without people doubting it.