We’re excited to launch 𝗦𝘆𝗻𝘁𝗵𝗜𝗗 today with GoogleCloud: a digital tool to watermark and identify AI-generated images. 🖼️ It will be available on Imagen, one of Google’s latest text-to-image models. Here’s how it works. 🧵 #GoogleCloudNext

💧 Watermarking SynthID can embed a digital watermark directly into the pixels of AI-generated images, making it imperceptible to the human eye. It won’t compromise the quality and can withstand common editing techniques like cropping or applying filters.

☑️ Verification By scanning an image for the watermark, SynthID can assess the likelihood of it being created on Imagen. It provides three confidence levels for interpreting its results, aiming to help users make more informed decisions around AI-made content.

🔍 Developing ways to help tell when content is created by AI is more important than ever. While SynthID is not a silver bullet for addressing the problem of misinformation, it is an early and promising technical solution to a pressing AI safety issue:

🔮 What’s next? As we continue to gather feedback on SynthID, we’re exploring its potential to expand to other models and be integrated into more @Google products - empowering people and organizations to work with AI-produced content responsibly.

@googlecloud @Google Very excited about this release! For those who may be interested in knowing more about related algorithms, I take this opportunity to advertise the work we did at Meta regarding the same techniques:

@googlecloud @Google Impressed by the new watermarking tool, but have you considered attacks like regeneration? Our research proves these can effectively remove any pixel-based invisible watermarks, regenerating images closely resembling the original.

@googlecloud @Google Why on earth would you not give it commercial use MIT license, stick it on @github and push the community to proliferate it. It would be 1000x more useful if it could be integrated into non-Google products, such as @StabilityAI cc @EMostaque

@googlecloud @Google Not sure this is going to work. Just Encode and Decode the images using a different model and should remove the water mark?

@googlecloud @Google You can’t to be a super intelligence without QPU

@googlecloud @Google This is super cool.