Google DeepMind's Gemini Omni 1.2 Flash is now generally available on the Gemini API and Google Cloud, and the headline for anyone budgeting video generation is the price. Google says it cut the cost 30% from Gemini Omni 1.1, to $0.07 per second for 720p video. A full-length 15-second clip at 720p is therefore about $1.05 (my arithmetic, not a Google figure). Per-second rates for 1080p and 4K are not given in the announcement; it points to the pricing page.

What changed from Omni 1.1

Item Omni 1.1 Omni 1.2 Flash
Max native clip length 10 seconds 15 seconds
Duration control not stated 3 to 15 seconds, 1-second steps
720p price not stated (1.2 is 30% lower) $0.07 per second
Availability not stated GA on Gemini API and Google Cloud
Editing not stated multi-turn conversational edits on full 15-second clips

Google also says speech is more natural: clearer lip-sync, plus "distinct multi-speaker conditioning". Camera work is more controllable, with "significantly fewer unwanted jump cuts" on long continuous moves such as FPV flythroughs, orbits and zoom-outs.

The cost workflow Google is steering you toward

The lower rate is only half the story. Google describes a pipeline built to avoid paying for full regenerations:

  1. Draft at 360p or 720p.
  2. Fix a detail with a follow-up edit instead of re-generating the whole clip.
  3. Upscale the keepers to 1080p or 4K.

I would build around that from day one. Spend is driven by how many whole-clip regenerations your pipeline allows, not by the per-second rate. If your current approach re-rolls a clip whenever the product colour or background is wrong, multi-turn editing is the feature to test first. The announcement's examples include swapping a background turn by turn while keeping the characters' performances and camera framing intact.

Reference inputs and consistency

For ad and catalog work, the useful part is reference conditioning. You can combine:

  • character, product or style images
  • short motion clips, up to three seconds
  • audio cues

Google says faces, garments, packaging typography and brand colours stay consistent across camera angles. It also claims less drift or morphing between the first and last frame when interpolating between keyframes or extending a clip.

Treat those claims as vendor-selected examples. The 1.1 versus 1.2 comparisons in the post are curated. Before committing, run your own brand assets through it, especially text on packaging, which is the first thing video models tend to mangle.

Production details that matter

  • GA status. The post says that "for the first time, Gemini Omni is now generally available", backed by Google Cloud's enterprise infrastructure and security.
  • Capacity. Provisioned Throughput is available. If you need predictable latency for a campaign burst, that is the lever.
  • Provenance. Outputs carry an imperceptible SynthID watermark. Check that against any downstream compliance or platform requirements you have.
  • Where to try it. Google AI Studio for prototyping, the Google Cloud console for enterprise use. Flow, Google Vids and Google Ads also get it. Vids offers 15-second clips at no additional cost with a Google account or Workspace subscription. The Gemini app rollout is for Google AI Ultra subscribers.

What Google admits is not solved

Google states that development is ongoing on "fast sports action, rapid script changes and complex physics". If your content is any of those, do not assume the consistency gains carry over. I would test those cases first rather than last.

What I would do this week

For teams already on Omni 1.1, the migration case is simple: the same workload gets cheaper and gains five more seconds per clip. Re-benchmark one real campaign end to end: draft at 360p, edit by conversation, upscale. Count how many full regenerations you actually needed against your current process. That number, multiplied by $1.05 for a 15-second 720p clip, is your real unit cost. Then price the 1080p and 4K upscale step from the pricing page, since that is the part the announcement leaves open.

Sources: