Model Card｜Gemini Omni 1.1 Flash

🌏 中文版

Model Information

Highlights

• Scene extension jumps from the predecessor's "reads only the last second" to analyzing up to 10 seconds of prior context, meaningfully improving visual consistency, and chains in 10-second increments up to a cumulative 40 seconds
• New first/last-frame interpolation: specify two keyframes and the model generates the continuous motion between them — useful for camera orbits, zoom transitions, and seamless loops
• A new 360p draft mode runs up to 60% faster than standard 720p and costs one-third as much, letting teams iterate cheaply before committing to a final render
• Ranks #1 on the Artificial Analysis Text-to-Video Arena (without audio) at 1322 Elo, ahead of second-place MiniMax H3 at 1302

Benchmark Results

⚠️ Arena Elo scores are live blind-vote figures from the Artificial Analysis leaderboard (snapshotted late August 2026) that shift weekly as new votes come in — they are not vendor self-reported numbers. Scene-extension, resolution, and pricing specs come from Google's official announcement and documentation.

Comparison with Predecessor/Competitors

Compared with its own preview, Omni 1.1's biggest advance isn't resolution but controllability: the scene-extension context window grows from 1 second to 10 seconds, making multi-segment extensions far more visually and narratively coherent, and first/last-frame interpolation lets developers pin down exactly where a camera move starts and ends instead of hoping a text prompt gets it right.

Against competitors, Omni 1.1 tops the "without audio" T2V leaderboard (1322 Elo) but slips to #2 on the "with audio" board (1237), trailing Wan3.0 at 1241 — its strength is concentrated in visual quality and camera control rather than synchronized audio-video generation. On pricing, the 720p rate holds steady at $0.10/sec, unchanged despite the added capabilities; but the $0.30/sec 4K rate matches Google's own Veo 3.1 Fast exactly, so it isn't undercutting on price.

The tiered-by-resolution pricing model — 360p at a third of the 720p rate — mirrors Alibaba's Wan3.0 pricing logic, a sign that mainstream video models are converging on the same billing pattern: cheap drafts, pricier finals.

Implications for Agent Development

previous_interaction_id turns "generate → review → tweak one thing in plain language" into a multi-turn conversation within a single session, rather than requiring the full scene to be re-described every time. For agents running content-generation pipelines, that means the state-management burden of tracking "previous video description plus diff" moves to the API side.

• If you're building a marketing/social content automation agent: use 360p draft mode to have the agent generate candidates in bulk, then upscale only the selected ones to 4K after a human or scoring agent picks winners — cutting average generation cost substantially
• If you're building a long-form narrative video pipeline: the 40-second cumulative extension cap combined with first/last-frame interpolation supports a "generate keyframes, let the model fill in the motion" storyboard workflow that replaces manual keyframe animation
• Not a fit: scenarios needing precise audio-video synchronization (it isn't the strongest on the with-audio arena — Wan3.0 and MiniMax H3 fit better), or scenarios needing offline/local deployment or fine-tuning, since it's fully closed-source and API-only

Today's Insight

I assumed "4K output" meant an across-the-board bump in native rendering quality, but Omni 1.1's 4K is an upscale, not a native render — native output tops out at 720p. That's a reminder to distinguish "native resolution" from "upscaled resolution" when reading a video model's spec sheet, since the two carry very different implications for detail fidelity; the per-second price gap ($0.10 vs $0.30) reflects that distinction too — upscaling is a comparatively cheap compute task, not triple the cost of native rendering.

References

• Google Blog: Gemini Omni 1.1 Flash lets you build with more control
• Gemini API docs: Generate and edit videos with Gemini Omni Flash
• Gemini API Release notes: Gemini Omni Flash in public preview (2026-06-30)
• Google DeepMind Model Card: Gemini Omni Flash
• Artificial Analysis: Text to Video Leaderboard
• the-decoder: Google's Gemini Omni 1.1 Flash makes AI video generation cheaper and more flexible
• Dataconomy: Gemini Omni 1.1 Flash Adds 4K Video Upscaling
• Hedra: Best AI Video Models in 2026