Words by
Jemma
Something is happening with Google's Veo series, and it's worth paying attention. On January 18, 2026, a well-known Google internals leaker posted a screenshot showing Veo 3.2 appearing in backend service configs, with a note that it was being added to Google Workspace. No official announcement. No blog post. Just a system reference that set the AI community off.
I've been tracking the Veo series closely since Veo 3 landed in May 2025 and changed what everyone thought AI video could do. If the leaked codenamed models are any indication, 3.2 could be a bigger jump than the 3.0 to 3.1 transition. For brands running video ads, that matters.
How We Got Here: The Veo Timeline
Google DeepMind's Veo series has moved fast. The original Veo was announced at Google I/O 2024 with claims of 1080p output and minute-long video generation. Veo 2 arrived in December 2024 with 4K support and better physics simulation. Then Veo 3 in May 2025 did something genuinely significant: it added native audio generation. Synchronized dialogue, sound effects, ambient noise generated from a text prompt alongside the video. Google DeepMind CEO Demis Hassabis called it the moment AI video left the silent film era.
Veo 3.1 shipped in October 2025 and was a production-focused update. Vertical format support (9:16) for Shorts and TikTok. An 'Ingredients to Video' feature letting you feed in up to three reference images to anchor character and scene consistency. Scene extension pushing clips beyond 60 seconds. First and last frame control for guiding how a sequence starts and ends. These weren't flashy features. They were the kind of things you need when you're actually trying to make something for a real campaign, not just a demo clip.
The Veo 3.2 Leak: What We Know
The leak that started the current round of speculation came from @bedros_p on January 18, 2026. The screenshot showed Veo 3.2 appearing in what appeared to be backend service configuration or internal dashboard data, with a reference to it being added to Workspace. The post spread quickly, and within hours the AI community was pulling apart what it might mean.
More interesting are the codenamed model checkpoints that have surfaced on third-party AI evaluation platforms like Artificial Analysis since late December 2025. Models labeled 'Sicily', 'Sisyphus', and most recently 'Artemis' have shown up in benchmark data. Artemis in particular has drawn attention for smoother motion, stronger adherence to complex prompts, and hints of extended clip durations. These checkpoints map closely to how Google typically stages model rollouts: internal testing, benchmark appearances, selective leaks, then a staged public release.
The Workspace angle is the one I keep coming back to. If Veo 3.2 lands natively in Google Vids, Docs, or Slides, it fundamentally changes the enterprise content creation pipeline. Marketing teams that already live in Google's ecosystem would have AI video generation built into their existing workflow, no separate tool, no API budget to manage.
What Veo 3.2 Might Actually Bring
Based on what we know about the Artemis checkpoints and the direction the 3.x series has been moving, there are a few areas where 3.2 is likely to improve. Better multi-shot consistency is the clearest gap in 3.1. Long scenes with the same character still drift in ways that require post-production cleanup. Getting that to hold reliably across cuts would open up full ad productions rather than single clip generation.
Extended native clip duration beyond the current 8-second base generation is another area to watch. The scene extension tool in 3.1 can push clips longer, but native generation at 30 seconds would change what you can produce in a single pass. That's the difference between a clip you insert somewhere and a complete ad unit.
Audio has also been improving. Veo 3's audio generation was impressive as a proof of concept but inconsistent for production use. Better multi-speaker dialogue, more reliable sound effect sync, and potentially music generation within the same prompt would make Veo genuinely usable for finished ad creative rather than a first draft that still needs a sound session.
The Workspace integration, if it materializes the way the leak suggests, is probably the biggest practical development. Not because the model is better inside Workspace, but because it removes the friction for teams that aren't AI-native. When video generation is three clicks away inside a tool people already use every day, adoption stops being a decision and starts being the default.
What This Means for Ecommerce Video Ads
For ecommerce teams specifically, the Veo trajectory is worth following more closely than most AI video developments. The 'Ingredients to Video' feature in 3.1 is directly relevant to product-based creative. You can feed in a product image as a reference, describe the scene, and get a clip where the product looks consistent across the shot. That's the core use case for brands running paid social: product in context, multiple scenes, no reshoots.
If 3.2 tightens up multi-shot consistency and extends native clip length, the practical outcome is AI-generated video ads that can actually run without significant post-production. Right now, AI video for ecommerce is a strong starting point that still needs finishing. The gap is closing faster than most brands realize.
The pricing context matters here too. Veo 3.1 on Vertex AI runs around $0.75 per second, which adds up fast for volume creative production. If 3.2 comes with a faster, lower-cost generation mode (which has been speculated based on the Workspace integration requirements), the economics shift considerably. Enterprise Workspace access suggests Google wants this in the hands of marketing teams that don't have a $10k/month API budget.
What to Watch
The most reliable signal for a 3.2 release will come from the Gemini app and Google Flow. Google tends to push new model capability there first before making a formal announcement. If you're using Flow regularly and a new model option appears, that's usually the first sign.
Keep an eye on Vertex AI changelogs and the Gemini API model list. Version bumps tend to show up there quietly before any public communication. Trusted leakers in the Google ecosystem, particularly those watching Workspace backend configs, have been accurate on timing before.
If Veo 3.2 delivers on what the Artemis benchmarks suggest, 2026 is going to be the year AI video stops being a novelty in ecommerce creative and becomes a standard line item in production budgets. I'll be running tests the moment it's publicly accessible.
