How AI ruins the internet

OpenAI's new AI images are basically undetectable, and that's a huge deal because it's coming from OpenAI. There have been other great AI image generators, but none miss powerful. And when Google released Nano Banana Pro in November, they released Synth ID along with it. So even if it generated a really good AI image, there was always a backup. OpenAI claims they have internal detection tools beyond C2PA. They are not releasing them publicly. There's no such system here. Was this image generated by OpenAI's AI image generator? Does it have any invisible watermarks? Short answer: I can't confirm that this was generated by OpenAI's image model, and I don't see clear evidence that it is. Longer, more useful answer: 1) Does it look AI-generated? It actually leans photographic rather than synthetic. Natural skin texture and asymmetry (especially around the eyes and mouth) Coherent lighting and shadows across both subjects Realistic depth of field and lens behavior No obvious artifacts (hands, edges, background geometry all check out) If this were AI, it's from a very high-end model or heavily post-processed. Nothing here screams "GPT images 2" specifically. 2) Could it still be AI? Yes. Modern models can produce images at this level. But there are no strong tells in this frame alone. 3) Invisible watermarks (the part you care about) I cannot detect invisible watermarks from just the image. It is not perfect, but it is almost there in the best case scenario. "Best case" or "worst case" - depends who you ask. I was able to generate this image of Obama, this image of Harris. I was not able to generate a certain New York financier, lol. I just not use the right words. You can't use them by name for any of these, but you can just describe (notable) people and it will generate them. It is not perfect. I'm not claiming that these are perfect or entirely undetectable, but they don't need to be. There are probably mistakes still being made here, but it can do really accurate text generation. It's very customizable. Look, if you're excited about some cheap diagrams or funny photos or cheap advertisements, I understand, but that comes at the cost of no one being able to trust any new photo they see on the internet anymore. Software not come to save us anytime soon. Our systems aren't ready for this. Photographic evidence isn't dead, but that does not scale. Chain of custody doesn't scale on social media. This isn't some random open-source image generator. OpenAI is accessible & influential. It's a very sad moment and it's an incredibly cynical release from OpenAI.