{"id":10441,"date":"2025-03-17T19:46:32","date_gmt":"2025-03-17T19:46:32","guid":{"rendered":"https:\/\/fusionmindlabs.com\/blogs\/google-ai-removes-image-watermarks-2\/"},"modified":"2025-03-17T19:46:32","modified_gmt":"2025-03-17T19:46:32","slug":"google-ai-removes-image-watermarks-2","status":"publish","type":"post","link":"https:\/\/fusionmindlabs.com\/blogs\/google-ai-removes-image-watermarks-2\/","title":{"rendered":"Google AI Removes Image Watermarks"},"content":{"rendered":"\n<\/p>\n
\n

Google has introduced its new Gemini Flash 2.0 image generation tool, which has been found by social media users to be capable of removing watermarks from copyrighted images<\/a>. Currently, the model is in its experimental phase and is only accessible to developers.<\/p>\n

A range of apps already exist that can remove objects from photos and fill in the gaps, and newer generative AI models have further enhanced their capabilities. Companies like Adobe have been updating their photo editors with advanced object-removal tools. Additionally, Apple’s Intelligence feature includes a “Clean up” tool on iOS and macOS that enables this functionality for supported devices.<\/p>\n

The following post on X demonstrates the effectiveness of Gemini Flash 2.0 in removing objects.<\/p>\n

\n

New skill unlocked: Gemini 2 Flash model is really awesome at removing watermarks in images! pic.twitter.com\/6QIk0FlfCv<\/a><\/p>\n

\u2014 Deedy (@deedydas) March 15, 2025<\/a><\/p>\n<\/blockquote>\n

Major closed models, such as those from Google and OpenAI, typically have significant restrictions in place to avoid legal issues. For instance, OpenAI’s Dall-E image model will not generate images of copyrighted characters. Microsoft recently sued a group of individuals<\/a> who managed to trick its image models into generating explicit content. However, it is likely that Google will attempt to restrict watermark removals, so \/r\/photoshoprequest<\/a> should not be too concerned.<\/p>\n

Some have noted that it may be impossible to completely prevent the misuse of this technology, as open models can have their safety guardrails disabled. Nevertheless, by demonstrating its efforts to prevent misuse, Google can protect itself from legal liability. Even open models can come with license agreements, and law enforcement and courts can intervene to stop abuses.<\/p>\n

Interestingly, in the case of watermark removal, Google’s model adds its own watermark<\/em> to images modified or generated by AI, making it clear to viewers that the image is not genuine. This can be seen as an allegory of AI itself: taking something owned by someone else, removing proof of ownership, and then adding one’s own identifying mark.<\/p>\n<\/p><\/div>\n