Since 7 October anyone can upload an image, a video or an audio file to Google’s SynthID Detector and see whether it carries the watermark from Google’s AI or a partner’s. Google says it has marked more than 180 billion images and videos. For a brand that ships AI work, disclosure stops being only your decision — your audience can now check.
Google opens its SynthID detector to everyone
In a post dated 7 October, Google DeepMind’s Pushmeet Kohli said the SynthID Detector is available globally in English from that day (Google, 7 Oct). It checks images, video and audio for the SynthID watermark from Google’s models and from partners OpenAI, NVIDIA and Kakao, with Apple “soon”. Google counts more than 180 billion watermarked images and videos and 240,000 years of audio. The limit is in the design: it finds its own watermark and nothing else. A clean result says nothing about tools that do not use it.
My read: The images for the Shirin Asal Christmas greeting came from Nano Banana before I built each sequence in Kling. Google’s API docs say every image its model generates carries SynthID (Gemini API docs). The question for a brand team is not whether the film can be checked. It can. The question is what you said before someone checked it. Decide the disclosure line at the brief, put it in the caption, and never let “AI-made” be the thing a commenter discovers.
Omnicom will put brands inside shows after they are shot
Digiday reported on 7 October that Omnicom Media’s Content Collective has partnered with Rembrand, whose AI system VISTA scans finished programmes for scenes that can carry a brand. The example: a packaged goods client inserted into an episode of a cooking reality show after it was produced and shot. Omnicom’s research with Rembrand found that these placements, paired with regular video ads, gave a 5.5 times impact on message recall and a 4x lift in purchase intent. Omnicom called it a nascent capability, with rates that differ by network (Digiday, 7 Oct).
My read: I solved the same problem the other way round in Festival Toffee. The sets and characters were painted with Kling, the packs stood in real 3D in front of them, and the AI never redrew a pack. That is the line I would hold with any insertion vendor: the pack in the scene is your approved artwork, lit to match the scene, not a regenerated lookalike. Get it in writing. And read the 5.5x for what it is — research by the two companies selling the product.
Figma’s agent now reads your brand rules from a file
Figma made its agent generally available on 6 October. Teams can add guidelines to design libraries as markdown files, pick the image model for on-canvas edits, and use generative plugins, now also generally available. From the same day the agent uses AI credits (Figma release notes, 6 Oct).
My read: Rules are what make a system hold. In One Group. Every Shelf. eight product categories share one grid, one timing and one tile movement, so fifty seconds read as one brand, not eight posts. I fixed those rules before I made a single product image. A guidelines file does that job for an agent, but only if someone writes real rules: margins, type sizes, what never touches the logo. A 60-page PDF brand book is not a rule. And budget the credits; the agent is now metered.
What I’d do today
Take your three most recent AI-made deliverables and run them through the SynthID Detector. Then agree one disclosure line with legal and the brand lead this week, before a journalist or a retailer runs the same check. If you want the production side, here is how I work.
Sources
- Google DeepMind (Pushmeet Kohli) — SynthID Detector expands globally, 7 Oct 2026 — blog.google
- Google — Gemini API image generation docs (SynthID on all generated images), checked 8 Oct 2026 — ai.google.dev
- Digiday (Michael Bürgi) — Omnicom Media and Rembrand partner on in-content ad placements, 7 Oct 2026 — digiday.com
- Figma — Release notes: the Figma agent is generally available, 6 Oct 2026 — figma.com