Google is making video generation more controllable. On August 29, 2026, the company released Gemini Omni 1.1 Flash, a production update to its native multimodal video generation and editing model. The upgrade moves Omni from a capable generator to a directable one — a useful distinction when “make it cinematic” stops being enough.
The largest change affects scene extension. Omni 1.1 analyzes up to 10 seconds of prior context when continuing a clip, replacing the previous approach of referencing only the final second. Extensions run in 10-second increments and can stack to a cumulative 40 seconds, while some final frames from the input are edited to make the seam continuous.
Uploaded input videos must be 10 seconds or shorter unless users extend a model-generated video in multi-turn. Spoken dialogue works in multi-turn extension through previous_interaction_id, but users cannot add new dialogue when extending an uploaded video where someone is speaking.
More Control Over Frames, Characters, and Resolution
Omni 1.1 also adds first-frame and last-frame control for interpolation between defined points. Prompts bind media to specific roles with <FIRST_FRAME>, <LAST_FRAME>, <IMAGE_REF_N>, and <VIDEO_REF_N> tags, giving creators shot-level control over camera movement and character continuity.
Video clips can serve as references for character consistency, with a maximum of three clips capped at three seconds each. Audio inside a video reference is ignored, so the model can borrow visual identity without treating the reference as an audio source. Voice editing and audio references remain unsupported, and YouTube URLs cannot be used as sources.
Google’s recommended workflow is blunt: draft in 360p, ship in 4K. The resolution parameter in response_format accepts 360p, 720p, 1080p, and 4k, while 1080p and 4K are upscaled outputs. Google reports that 360p previews generate up to 60% faster and cost a third as much as 720p.
Every generated video carries SynthID watermarking. The model does not support system instructions, temperature, top_p, stop sequences, or negative prompts, which makes the control surface focused on media and framing rather than the usual pile of generation settings.
Production Access Comes With a Meter
Omni 1.1 Flash is live in Google Flow for AI Plus, Pro, and Ultra subscribers, with scene extension available in the Gemini app. The model is paid-tier only, with no free tier. Google names Adobe’s Firefly, Figma Weave, GMI Cloud, and Runway as customers already running Omni Flash in production.
The pricing explains why 360p drafts matter. Input costs $1.50 per 1M tokens, while output costs $9.00 per 1M text tokens and $17.50 per 1M video tokens. Video billing runs at 5,792 tokens per second of 720p video, producing an effective cost of about $0.10 per second — cheap enough for iteration, but not free enough to ignore.
Alongside the release, Google employees are testing Gemini 3.8 Flash Preview, a new AI model made available on the company’s internal coding platform, Jetski. Some staff have begun using the preview version, and one employee said, “It already felt noticeably better than 3.7 Flash, but emphasized it was too early for a full review.”
That second development is still an internal test, not a public launch. The public Omni update is easier to judge: longer scene memory, first-and-last-frame interpolation, reference clips, 40-second extensions, and 4K outputs turn video generation into a more structured editing workflow. The model still has sharp limits around audio, prompt controls, and access — because apparently even video models need a paywall and a rulebook.
Based on
