Gemini Omni 1.1 Flash

Who Should Read

Developers building creative tools or media editing services on top of a generative video API

Core Mechanics

• Scene extension sees 10 seconds of context
  • Earlier models referenced only the final second; Omni 1.1 analyzes up to 10 seconds of prior video.
  • You extend in 10-second increments up to 40 seconds total.
• First and last frames can be specified
  • The model generates continuous video between two keyframes.
  • Suggested uses are camera orbits, zoom transitions and seamless loops.
• 360p drafts are faster and cheaper
  • Up to 60% faster and a third of the cost versus Omni 1.1's standard 720p.
  • The 60% figure is based on system throughput of 360p versus 720p.
• Final output goes up to 1080p or 4K
  • The workflow is to settle on a direction with drafts, then render at 1080p or 4K.
• Video references can join multimodal input
  • You can reference up to three seconds of video to keep visual context and character consistency.

Community Reactions

• Draft and final results differ
  • One commenter liked a 360p draft, reran the same prompt at standard resolution and got a different result.
  • They attribute it to the model being non-deterministic.
• No sync to existing audio
  • A commenter says neither the last Omni nor this one can sync generated video to pre-existing audio.
  • They use Minimax H3 locally on a 12GB RTX 4070 for lip sync instead.
• Long videos are still hard
  • A commenter says getting even 30 seconds of an AI avatar narrating is difficult because videos drift.
  • They asked for real examples of extension and interpolation that work.
• Curiosity about even lower draft resolutions
  • One commenter wonders whether 240p or 144p would be usable, upscaled by a fast upscaling-only network.
• Doubts about practical use
  • One comment sees ads and pre-production as plausible but struggles with other applications.

How to Apply

• Generate 3-4 360p drafts in AI Studio today
  • Change one variable at a time on the same scene and generate in 360p.
  • Compare side by side and render high resolution only for the winner.
• Measure draft versus final differences first
  • Run the same prompt at 360p and 720p and record how much composition changes.
  • If it differs a lot, budget regeneration cost into your approval flow.
• Add scene extension to your pipeline
  • Pass the prior interaction id as previous_interaction_id to extend by 10 seconds.
  • Cap the number of extensions in code so you stay within the 40-second limit.
• Use keyframes for loops and camera moves
  • Feed the same image as first and last frame to try a seamless loop.
  • Put One continuous shot, no jump cuts in the prompt.

Try It

# Run in Python (based on the article's example; needs a prior video interaction)
# pip install google-genai

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-omni-1.1-flash",
    previous_interaction_id=previous_video_interaction.id,
    input=[{"type": "text", "text": "Continue the scene."}],
    response_format={"resolution": "360p"},
)

# Next: repeat this call 3 times to reach the 40s cap, then raise resolution only for the last render

Terminology

Related Papers

• Mercury 3: Diffusion at scale

  Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality

• Timestral: Mistral's time-series foundation models

  A 209M time-series model ranks first on three benchmarks and runs on CPU.

• DeepSeek V4.1 Flash 1017 (Max)

  DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.

• If you don't run your own LLM proxy, how do you know whats going on?

  Running your own LLM proxy shows which keys, models, and cache misses drive your usage.

• DeepSeek: DeepSeek v4.1 Flash 1017

  DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents

• DeepSeek-v4.1-Flash-1017

  DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length

• Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware

Related Resources

• https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/