Google's Gemini Omni 1.2 Flash Slashes Video Generation Costs by 30%

Google has released Gemini Omni 1.2 Flash in AI Studio and the Gemini API. According to the company, the model costs 30% less than its predecessor, generates clips up to 15 seconds, and preserves the current shot across conversational edits.

Google introduced the Omni series at Google I/O in May 2026. Version 1.1 added scene extension and first-and-last-frame controls, which let creators pin the opening and closing composition. Version 1.2 focuses on cost, clip length, latency, audio, and prompt adherence.

Cheaper seconds, longer shots

Google lists six changes that affect production workflows and budgets:

Using the previous rate of about $0.10 per output second, a 30% reduction produces an estimated rate of $0.07 per second. A 15-second clip would cost about $1.05 under that baseline. Additional generations, revisions, and variants increase the effective cost, so teams should confirm current billing terms before forecasting production spend.

Google describes generation as faster but does not provide comparative latency figures in the announcement. Production evaluations should measure median and tail latency with representative prompts, resolutions, audio settings, and concurrency levels.

Edits keep the shot in context

Conversational editing now keeps a generated clip in context while follow-up prompts alter specific elements. A creator could request a slower camera push, a red jacket, or different dialogue while retaining the broader composition and timing.

A typical editing session follows three steps:

1. Generate the initial clip from text, images, or both.
2. Submit a targeted revision through the Interactions API.
3. Review the revised output and continue the same conversation as needed.

The workflow can reduce prompt reconstruction and random variation between attempts. Google’s summary does not specify whether each conversational revision receives the same billing treatment as a fresh generation, making that detail important for cost models built around repeated edits.

From prompt to API job

Google exposes the model through several products, each aimed at a different workflow:

The Python SDK starts video generation as a long-running operation, so application code must poll for completion or use the equivalent asynchronous method. With credentials configured, a minimal call has this shape:

import time
from google import genai

client = genai.Client()

operation = client.models.generate_videos(
    model="gemini-omni-1.2-flash",
    prompt=(
        "A slow dolly-in on a rain-soaked neon street, "
        "cinematic lighting"
    ),
    config={"duration_seconds": 15},
)

while not operation.done:
    time.sleep(10)
    operation = client.operations.get(operation)

generated_video = operation.response.generated_videos[0]
client.files.download(file=generated_video.video)
generated_video.video.save("neon-street.mp4")

Follow-up edits use the Interactions API and the references returned by the initial generation. Exact field names can vary by SDK version, so implementations should follow the current API documentation and preserve operation state for retries, logging, and asset tracking.

Every generated video includes SynthID, Google’s embedded provenance watermark. Pipelines that distribute or transform outputs should retain that requirement in their publishing and compliance checks.

Workloads that fit the Flash tier

The lower-cost Flash model suits workloads that favor iteration volume and turnaround time:

• Advertising variants: Generate multiple versions of a short spot with different products, settings, or camera directions.
• Animatics: Turn storyboard beats into clips long enough to evaluate pacing and shot order.
• Agent-driven production: Let a planning model generate, inspect, and revise clips through the Interactions API.
• B-roll and social video: Produce short assets where 15-second outputs and SynthID meet publishing requirements.

Teams evaluating the model should test character continuity, object persistence, speech quality, camera compliance, and revision stability. The 15-second ceiling still requires sequencing and post-production for longer scenes, while project quotas and regional availability may shape throughput.

No Veo deprecation date accompanies this release. Existing Veo integrations should be compared through parallel evaluations that cover request schemas, supported controls, latency, quotas, output quality, and total cost before any migration.

The cost curve changes production planning

Lower output prices make generation budgets easier to forecast, while longer clips reduce the number of assets needed for short sequences. Conversational revisions can also simplify orchestration because each requested change stays attached to the current shot.

The production decision depends on measured latency, edit consistency, quota, and effective cost after retries. Omni 1.2 Flash gives developers a cheaper, longer-running option for iterative video pipelines, with the largest gains likely in systems that generate and revise many short clips.