Google DeepMind's incremental 2026 video update โ 60-second native 4K clips, scene-level consistency, and audio that stays in sync across cuts.
Veo 3.2 is DeepMind's late-2026 video model update. It extends clip length from 30s to 60s, adds a "scene lock" that keeps characters consistent across cuts, and finally delivers audio that doesn't drift out of sync past the 15-second mark. Available via Vertex AI and YouTube Shorts for creators.
Who it's for: Ad agencies, YouTubers with high output quotas, and enterprise marketing teams already paying for Google Cloud credits.
Doubles the previous max. Most TikToks and YouTube Shorts now generate in one shot, no stitching.
Reference a character and the model holds their look across multiple cuts โ previously the #1 complaint.
Dialogue, SFX, and ambience stay aligned through the full clip, not just the first 15 seconds.
$0.50 buys 1080p; $1.20 buys 4K with detail that survives projection.
Prompt "35mm shallow DOF" or "dolly push" and get cinematically correct results, not mush.
Queue 100 renders and walk away. Rate limits are generous on the Vertex SKU.
Veo 3.2 is the model Google should have shipped at Summer Showcase. Scene lock alone closes the character consistency gap that plagued every 2026 video AI, and synced audio removes the last blocker to professional delivery. Sora 3 is still better for stylized anime; Veo 3.2 is the pick for photoreal ads.