要点
- Native generation and editing up to 15 seconds (from 10s on Omni 1.1); duration selectable in 1-second steps from 3 to 15.
- Price cut of 30% vs Omni 1.1: $0.07 per second for 720p; draft at 360p/720p, upscale winners to 1080p or 4K.
- Multi-turn conversational editing across full 15-second clips: swap product, background, lighting, aspect ratio without restarting.
- Stronger reference adherence for character, product, style images, short motion clips (up to 3s), and audio cues; SynthID watermark; Provisioned Throughput on Google Cloud.
- GA on Gemini API and Google Cloud; available in Flow, Google Vids, Gemini app (AI Ultra rollout), and Google Ads; DeepMind still flags weak spots in fast sports, rapid script changes, and complex physics.
On 5 November 2026, Google DeepMind published "Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips." The headline numbers are easy to quote: clips stretch from 10 seconds on Omni 1.1 to 15 seconds, and list price falls 30% to $0.07 per second at 720p. The sharper move is workflow. Multi-turn conversational edits let teams change a product, background, or aspect ratio on the same clip instead of burning another full generation. Length and price matter; the durable claim is that regeneration waste, not one more second of raw footage, is the cost center.
[1]What shipped
Omni 1.2 Flash is generally available on the Gemini API and Gemini for business / Google Cloud. Creators can set duration in one-second steps from 3 to 15 seconds. Reference inputs combine character, product, or style images, short motion clips up to three seconds, and audio cues. DeepMind says faces, garments, packaging type, and brand colors hold across camera angles, aimed at virtual try-on, catalog spins, and branded ads.
Editing is framed as conversation: follow-up turns can swap backgrounds, weather, lighting, or framing while keeping performances. Speech claims include more natural dialogue, clearer lip-sync, and distinct multi-speaker conditioning. Camera control demos stress continuous FPV-style moves with fewer unwanted cuts. Outputs carry an imperceptible SynthID watermark. Provisioned Throughput is available for enterprise. Consumer paths include Flow, Google Vids (15-second clips at no additional cost with a Google account or Workspace), the Gemini app for AI Ultra subscribers (rolling out), and Google Ads campaign tooling.
[1]Evidence on the table
Checkable commercial facts are the price and duration: 30% below Omni 1.1, $0.07/s at 720p, native 15-second generation and edit. The recommended cost path is explicit: draft at 360p or 720p, fix details with follow-up edits, then upscale chosen takes to 1080p or 4K. That is a statement about where money leaks in current pipelines.
Consistency claims are demo-backed, not third-party benchmarked in the post. Side-by-side notes say 1.2 follows a layered exhibition-poster prompt more closely than 1.1, and that first-to-last frame interpolation keeps more scene details. Multi-turn examples keep character armor and framing while swapping pavilion, forest, stable, fair, and castle backgrounds. A try-on sequence combines one model photo with front-and-back outfit shots into a 15-second 360-degree turn. These are controlled demos from the vendor.
DeepMind itself lists remaining gaps: fast sports action, rapid script changes, and complex physics are still under improvement. That admission is useful evidence about the product surface the launch is optimizing: branded, explained, edited clips at scale, not sports broadcast replacement.
[1]Steelman and gaps
Steelman. This is a DeepMind product blog with vendor demos and vendor pricing. "Stronger consistency" and "more cinematic" are qualitative. SynthID is a provenance aid, not a proof that edited ads are truthful. GA across Cloud does not tell you latency, failure rate, or brand-safety reject rates in production. Rivals can cut price and extend clip length without matching the multi-turn story.
Fair. The falsifiable near-term test is economic and operational: do production teams cut spend per approved asset because conversational edits replace regenerations, and can competitors offer 15-second editable clips at similar unit cost with equal reference hold? If users still regenerate full clips for every variant, the 15-second headline outruns the thesis.
[1]Who should care, and what to watch
Ad and ecommerce teams already stitching short clips should model cost as (draft seconds + edit turns + upscales), not only sticker price per second. API developers should compare Omni against other video models on reference hold across multi-turn edits, not only first-frame FID-style scores. Policy and brand-safety desks should treat SynthID as necessary but incomplete when conversational edits can swap products and scenes inside one session.
Next six months: whether $0.07/s and 15-second edits become the default buyer checklist; whether sports and physics gaps close enough for broader categories; and whether Google Vids / Ads distribution turns Omni into a default production path or stays a Cloud API feature.
[1]Close
Omni 1.2 Flash is easiest to misread as a length bump. Six months out, either approved-asset cost curves bend because teams edit in place, or the market treats fifteen seconds and a 30% cut as another launch card while regeneration waste stays the real bill.
[1]