Google says advertisers generated over 1.5 billion campaign assets with Gemini in the third quarter alone, up more than seven times in a year. Making variations is no longer the cost. Choosing them is, and so is deciding what a machine is never allowed to change about your product.
Google: 1.5 billion Gemini-made ad assets in one quarter
On Alphabet’s Q3 call on 28 October, chief business officer Philipp Schindler said “Gemini models helped advertisers generate over 1.5 billion campaign assets, up over 7x year-over-year”, that Google has “integrated our AI creative tools directly into AI Max”, and that ads in AI Mode are expanding to international markets, with multi-turn ads in AI Mode in the US (call transcript, 28 Oct). Google Search & other revenue grew 12% to $63.35 billion, and the Gemini app has reached one billion users (Alphabet Q3 release, 28 Oct). The international markets were not named on the call (TechCrunch, 28 Oct).
My read: Volume was never the hard part. On one launch I cut ten opening hooks from the same film; the data picked the winner, and it beat my own favourite. The skill now is deciding what is allowed to vary. For a full group range, that list is short and strict: pack shape, logo colour, the product claim. Everything else can be tested. If nobody writes that list before AI Max starts generating, the platform decides it for you, a billion times.
Genie 4: a world you can direct, with sound
Google DeepMind’s Genie 4 generates an interactive world in real time, now with native audio (ambient sound and dialogue), more natural human motion and physics. A text prompt can change the weather or the camera mid-run. DeepMind says testers in creative fields use it as a “draft mode” for previsualisation, running several takes and changing camera, light and action. It also lists the limits: detail and object permanence drift over longer runs, and legible text is still unsolved. Access starts with trusted testers, then Project Genie for Google AI Ultra subscribers and early-access Gemini API partners (Google DeepMind, 27 Oct).
My read: This lands in previs, not in the final film. On films like Mooo’s back-to-school spot I drew the location sheets before a single shot existed; walking a camera through a rough world first would save rounds of “move it left”. But a product film lives on the pack, and Genie still cannot hold legible text. Use it to agree space and blocking, then build hero shots the slow way.
ByteDance’s Jimeng opens a short-drama studio
Jimeng AI, ByteDance’s creative tool, launched a short-drama studio on 26 October, according to DoNews. It runs script generation with a “hit potential” check and parsing of texts up to 100,000 characters (model DreamScript), storyboards with a 3D director’s stage for camera blocking (DreamShot), single-shot reshoots, voice separation and re-dubbing, and a remake mode that recognises an uploaded drama’s characters, scenes, props and shot structure and regenerates it for a target region’s style and format (DoNews, 26 Oct; TechNode, 27 Oct). Neither report gives a price or availability outside China.
My read: The remake mode is the one brand teams will ask about. One hero film, many markets is real work for me: one concept, several language and format versions for a regional rollout. What I learned there is that localising is not regenerating. You change voice, text, pacing and sometimes casting. You do not let a model redraw the product. If a tool re-renders every frame “for the region”, check the pack in every frame before anyone signs off.
What I’d do today
Write the three things no AI tool may change in your ads: the pack, the logo colour, the claim. Put that line at the top of every brief and every automated-asset setting before you switch them on. If you want a film where those three hold in every frame, I direct it shot by shot.
Sources
- Alphabet — Q3 2026 results press release, 28 Oct 2026 — https://s206.q4cdn.com/479360582/files/doc_news/2026/Oct/28/attachments/2026q3-alphabet-earnings-release.pdf
- Investing.com — Alphabet Q3 2026 earnings call transcript (Philipp Schindler remarks), 28 Oct 2026 — https://www.investing.com/news/transcripts/earnings-call-transcript-alphabet-q3-2026-profit-miss-sparks-10-afterhours-drop-93CH-4981037
- TechCrunch — Google expands AI search ads globally as core ad growth slows, 28 Oct 2026 — https://techcrunch.com/2026/10/28/google-expands-ai-search-ads-globally-as-core-ad-growth-slows
- Google DeepMind — Genie 4: a new generation of world simulation, 27 Oct 2026 — https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
- DoNews — Jimeng AI short-drama studio launches (in Chinese), 26 Oct 2026 — https://www.donews.com/news/detail/1/6754064.html
- TechNode — ByteDance launches AI studio for short drama production, 27 Oct 2026 — https://technode.com/2026/10/27/bytedance-launches-ai-studio-for-short-drama-production/