OpenAI May Watermark ChatGPT Images For Free Users

The buzz surrounding ChatGPT’s latest image-generation capability continues to grow. As usual, tech enthusiasts and users have been exploring the app and recently uncovered references to a watermark feature intended for the images produced by the AI.

Notably highlighted by X user Tibor Blaho, a snippet of code named image_gen_watermark_for_free hints that this feature may apply only to images generated by users utilizing the free service. This setup could motivate users to consider upgrading to a paid plan.

This isn’t the first instance of OpenAI venturing into watermarking concepts. In the past year, reports indicated that the organization had explored a similar tool aimed at watermarking AI-generated text, although it never made it to public release.

The decision to hold back on implementing watermarking for text received critique, as it seemed to prioritize profit over ethical considerations. Implementing watermarks could help prevent AI-generated content from being misused, yet restricting access might lead to a decrease in user engagement.

On the flip side, applying watermarks to images created for free users could benefit the company financially if executed properly. Although it’s still uncertain if this feature will be rolled out, if it does, the feedback from users will likely be vocal.

The visual impact of these watermarks is of utmost importance. The term “watermark” usually evokes images of text or logos superimposed on photographs, but there are alternative methods. For example, Google’s watermarking for AI images employs a technique that subtly alters a small number of pixels, creating a pattern that can be detected by specialized tools without being apparent to the human eye.

This strategy holds multiple advantages; it doesn’t detract from the image’s quality for viewers and makes it more challenging to remove the watermark through standard editing methods.

Such a watermarking approach would likely enhance the user experience for those on the free tier. Conversely, the absence of watermarks for paid users might present a peculiar situation. If the images are indistinguishable, the only advantage of the watermark-free images for subscribers would be the potential to misrepresent AI-generated images as genuine or human-created. This raises ethical concerns.

However, this remains speculative, as concrete details about this potential update are yet to be revealed, and it’s possible OpenAI will decide against implementing it altogether. If enough individuals inquire about it with CEO Sam Altman on X, he might offer some insights.