Photo: Google.
Not being able to tell what is or isn’t AI generated content is becoming a key inhibitor to promoting trust in gen AI. Now, this doesn’t have to be an issue with the release of SynthID from Google into opensource and the addition of Notes to tell if an image in Google Photos has been edited by AI.
Working with SynthID
Currently SynthID can be used to watermark and identify text generated by the Gemini app and web.
SynthID works in two ways. First, SynthID embeds an invisible digital watermark directly into AI-generated content whether images, audio, video, and text as they are generated, without compromising the original content. Second, SynthID can scan images, audio, text or video for digital watermarks, helping users determine if content, or part of it, was generated by Google’s AI tools.
According to Google, SynthID works by altering the probability score of some of the words that are generated to the user in a way that’s invisible to the human-eye, but clear to a SynthID detector:
- An LLM generates text one token at a time. These tokens can represent a single character, word or part of a phrase. To create a sequence of coherent text, the model predicts the next most likely token to generate. These predictions are based on the preceding words and the probability scores assigned to each potential token.
Photo: Google.
- For example, with the phrase “My favourite tropical fruits are __.” The LLM might start completing the sentence with the tokens “mango,” “lychee,” “papaya,” or “durian,” and each token is given a probability score. When there’s a range of different tokens to choose from, SynthID can adjust the probability score of each predicted token, in cases where it won’t compromise the quality, accuracy and creativity of the output.
- This process is repeated throughout the generated text, so a single sentence might contain ten or more adjusted probability scores, and a page could contain hundreds. The final pattern of scores for both the model’s word choices combined with the adjusted probability scores are considered the watermark. This technique can be used for as few as three sentences. And as the text increases in length, SynthID’s robustness and accuracy increases.
Photo: Google.
According to Google, SynthID has been designed, “so it doesn’t compromise image or video quality, and allows the watermark to remain detectable — even after modifications like cropping, adding filters, changing colours, changing frame rates and saving with various lossy compression schemes”.
And while not a silver bullet for addressing problems such as misinformation or misattribution, Google says that SynthID is step towards a suite of promising technical solutions to address this pressing AI issue.
SynthID is available through the Google Responsible Generative AI Toolkit, which provides guidance and essential tools for creating safer AI applications. Additionally, Google has been working with Hugging Face to make the technology available on their platform, so developers can build with this technology and incorporate it into their models.
Keeping track of AI edits
According to a blog post, starting from next week, Google Photos will note when a photo has been edited with Google AI right in the Photos app to "further improve transparency" the company said. When it arrives and you click on a photo in Google Photos, you will see if its been “Edited with Google AI” when you scroll to the bottom of the “Details” section.
Photo: Google.
Google said that it will also use IPTC metadata to indicate when an image is composed of elements from different photos using non-generative features. For example, Best Take on Pixel 8 and Pixel 9, and Add Me on Pixel 9 use images captured close together in time to create a blended image to help you capture great group photos.
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.