OpenAI to watermark AI-generated images using DALL-E 3

DALL-E 3, powered by ChatGPT allows users to generate images using text prompts.

An image of a shipwreck generated by DALL-E 3 (Image credit: OpenAI)

OpenAI has started incorporating the Coalition for Content Provenance and Authenticity (C2PA) specifications for images generated using DALL-E 3. The company has announced that it will soon add watermarks to images generated using DALL-E 3, which includes a visual watermark and updated metadata, allowing anyone to verify the source of the image via C2PA.

The aim of adding a watermark to an image is to help users quickly identify if the image has been generated using AI or created by humans, and brands like Adobe, and even camera makers have been following the C2PA standard for a while. OpenAI has confirmed that there will be a visible watermark on DALL-E 3 generated images via the web, API, and mobile app.

Images generated in ChatGPT and our API now include metadata using C2PA specifications.

This allows anyone (including social platforms and content distributors) to see that an image was generated by our products. https://t.co/kRv3mFnQFI pic.twitter.com/ftHqECS8SB

— OpenAI (@OpenAI) February 6, 2024

The watermark will include details like the image generation date along with the logo of the C2PA on the top left corner of the image. Currently, the DALL-E 3 image generator is available for ChatGPT Plus subscribers, which costs $20 a month.

Story continues below this ad

Also read | ChatGPT now supports Dall-E 3: Top 10 techniques for creating stellar images

OpenAI claims that adding a watermark to an AI-generated image will not affect the performance such as image quality or latency in any way. However, the image size does go up by three to five per cent when generated through API, and it goes up by 32 per cent when an image is generated using the ChatGPT platform.

However, one can still tamper with the image by physically cropping the image and manipulating the metadata. When one takes a screenshot of an AI-generated image or uploads it to a social media platform, the metadata gets removed.

Microsoft also includes a watermark for images generated using the Bing Image Creator, which is also powered by GPT. Samsung adds a watermark and updates metadata when an image is manipulated using the built-in AI image editing tools on the Galaxy S24 series, which will soon be available on the Galaxy S23 series. And Meta also recently introduced a feature that adds invisible watermarks to AI-generated content.

Why is it crucial to watermark AI-generated images?

Also read | Taylor Swift AI ‘deepfakes’: What happened and where did the images come from?

AI-generated or AI-manipulated images have been creating issues for a while, including impersonation, deepfakes of celebrities, and a lot more. With an easily visible watermark, it will make it easy for non-tech-savvy audiences to identify if it is an original image or if it has been generated using AI to help prevent the spread of misinformation.