Most companies that are involved with AI that generate images have made it so that such images are watermarked to indicate as much. But an unexpected use of such image generators is also to remove existing watermarks, such as those by stock image repositories. And the Google Gemini 2.0 Flash is apparently pretty good at that.
Examples of this are posted on Reddit as well as X, where watermarks of all sorts are removed, while Google Gemini 2.0 Flash fills in any gaps from their removal. Of course, being an AI-generated image, the tool then inserts its own watermark on the lower left corner. Which, ironically, is pretty emblematic of AI image generation as a whole.
According to TechCrunch though, this specific feature is labelled as “experimental” and “not for production use”. As such, it’s only available to developer-facing tools like AI Studio. Though the fact that it’s allowed in any version of Google Gemini 2.0 Flash to begin with may cause some legal issues, as the report notes rivals like OpenAI’s GPT-4o will simply refuse to remove watermarks.
Speaking of lacking guardrails in Google Gemini 2.0 Flash, the report also pointed to posts on X where the model was made to create images depicting celebrities and copyrighted characters. These were apparently done with little resistance, which is unexpected considering that a prominent artist was subjected to NSFW image generations last year.
