Gemini Omni 1.1 Flash: Google's directable video model with 40-second scenes and 4K upscaling

Google released Gemini Omni 1.1 Flash, a production update that turns its native multimodal video model into a directable tool for AI builders. The key additions: scene extension that reads up to 10 seconds of prior context, first and last frame pinning for camera control, video references for character consistency, and a draft-then-upscale workflow with 360p previews and 4K output. For teams building video generation into products, this changes the economics and control surface.

Scene extension with real context

Omni 1.1 analyzes up to 10 seconds of prior context when continuing a clip, compared to previous models that referenced only the final second. Extensions run in 10-second increments to a cumulative 40 seconds, generating 3-10 seconds per call. Some final frames of the input are edited to make the seam continuous. The extension appends to the end of a clip only; no prepending or mid-clip insertion is supported. Uploaded input videos must be 10 seconds or shorter, unless you are extending a model-generated video in multi-turn.

Frame pinning and video references

You can now supply a first and last frame and have the model generate the continuous video between them, enabling orbits, dolly-zooms, and seamless loops. Prompts bind media to roles with tags: <FIRST_FRAME>, <LAST_FRAME>, <IMAGE_REF_N>, and <VIDEO_REF_N>. Video references accept a maximum of three clips, up to three seconds each, and work best for likenesses. Audio inside a video reference is ignored. Reasoning across multiple videos is not supported and may degrade output.

Editing is stateful: you pass previous_interaction_id and the model applies your change while preserving what you did not mention, without re-uploading the prior video. This makes iterative editing practical for production pipelines.

The draft-then-upscale workflow

The resolution parameter in response_format takes 360p, 720p (default), 1080p, and 4K, with the top two upscaled. Google reports 360p previews generate up to 60% faster and at a third of the cost of 720p, based on system throughput. That makes the draft-then-upscale loop the intended production pattern: iterate cheaply, render once.

Pricing is $1.50 per 1M input tokens (text, image, video, audio). Output is $9.00 per 1M text tokens and $17.50 per 1M video tokens. Video billing runs at 5,792 tokens per second of 720p, an effective ~$0.10 per second under standard pricing. Every generated video carries SynthID watermarking, invisible to viewers but programmatically detectable for provenance.

Availability and production users

Omni 1.1 Flash is available through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. Google names Adobe (Firefly), Figma Weave, GMI Cloud, and Runway as customers already running Omni Flash in production. The model is also live in Google Flow for AI Plus, Pro, and Ultra subscribers, with scene extension in the Gemini app.

What's missing

Notable gaps: no system instructions, temperature, topp, stop sequences, or negative prompts (negatives go in the prompt text). Voice editing is unsupported. Audio references are unsupported. YouTube URLs cannot be used as a source. English is fully supported; other languages are unevaluated. You cannot add new dialogue when extending an uploaded video where someone is speaking; spoken dialogue is supported in multi-turn extension via `previousinteraction_id. For outputs above 4MB, use delivery="uri"` and poll the Files API until the file is ACTIVE.

For builders, the practical takeaway is clear: Omni 1.1 Flash gives you a stateful, directable video generation API with a cheap iteration loop and 4K output, but the constraints around extension direction, dialogue, and audio references mean you need to design your workflow around what the model can and cannot do.