The problem with AI-created images is that, well, AI is creating images. That means a simple prompt can create a relativity realistic image of anything. It is not difficult to imagine the issues that could arise from that fact.
Google announced on Tuesday a feature called SynthID that is aimed at combatting deepfakes and misuse of AI-generated images. Google DeepMind launched a beta version of the tool, which, the company said, "embeds a digital watermark directly into the pixels of an image, making it imperceptible to the human eye, but detectable for identification."
In other words, on its surface, the AI-generated image will look normal on the surface but have an easily identifiable marker to let folks know it's not an authentic image. Right now SynthID has been released to a "limited number" of customers using Imagen, Google Deepmind's model that uses text promtps to create photorealistic images.
Google said SynthID should be more effective that traditional watermarks, which are more easily removed, cropped out, or are an eyesore on an image. The SynthID watermark should remain visible even after filters are added, colors are changed, or an image is resized. The SynthID tool should be able to give users a reading on whether or not Imagen was used to create an image using a three tier, confidence level system — likely, unlikely, and possibly detected.
Moving forward, Google said it might expand the the tool to be able to identify images from other models and that it might integrate the SynthID tool into other products it offers.
