ChatGPT’s image generation tool is now available to free users—and with that, OpenAI may be taking steps to clearly label what’s AI-made. According to early tests spotted by researchers, a new chatGPT image watermark may be rolling out to mark images created using ChatGPT’s image generation model.
The watermark, which appears subtly on AI-generated images from the free version of ChatGPT, is currently in testing and hasn’t been officially announced. If it does happen, it would be a deviation from some strange responses OpenAI Ceo Sam Altman has made on X about ChatGPT infringing on copyrights with its image generation tools.
However, AI researcher Tibor Blaho and others have reported seeing the watermark mentioned in the code for ChatGPT’s image generation model. It only mentions free generation, though, so it is possible the watermark won’t apply to premium ChatGPT subscriptions.
This change could be significant if it goes through, especially given how viral ChatGPT’s ImageGen model has become in recent days. We’ve seen some crazy images generated by ChatGPT since the update dropped in late March.
The image watermark is likely OpenAI’s response to concerns about authenticity and user-generated content from ChatGPT. As more users create and share AI-generated art online, the ability to distinguish between real and synthetic content becomes critical—not only for artists and developers but also for the general public. Watermarking is a simple way to indicate that AI generated the image without interfering too much with the end result.
Under the hood, OpenAI says its image generator is trained on a massive dataset of online images and text. This dual training helps the model produce visuals that are both relevant to the prompts and surprisingly coherent—we’ve seen so many Studio Ghibli-styled images in the past week alone.
The company has also confirmed it’s developing an API for the image generation tool, which would make it easier for third-party developers to build their own tools around it. Whether the image watermark will extend to the API, or only to ChatGPT users is unclear, though.
