Gemini Omni 1.2 Flash: 15-Second AI Video at $0.07/s (Pricing, How to Use, Limits)

Google DeepMind has released Gemini Omni 1.2 Flash, the new version of its video model, and it is a very practical upgrade: clips now run up to 15 seconds (up from 10), the 720p price drops 30% to $0.07 per second, and you can edit a full 15-second clip by chatting with it instead of regenerating. It went live on 5 November 2026 in the Gemini API, AI Studio, Vertex AI, Flow, Google Vids and the Gemini app for AI Ultra subscribers. In this guide I cover exactly what changed from 1.1, every price per resolution, how to use it step by step (no-code and Python), the limits and region rules, how it compares with Veo 3.1, Kling, Seedance and Wan, and fixes for the errors people are already hitting.

What Is Gemini Omni 1.2 Flash?

Gemini Omni Flash is Google’s video generation and editing model inside the Gemini family. It first appeared as gemini-omni-flash-preview on 30 June 2026, became generally available as Omni 1.1 Flash on 27 August, and now gets its second GA update. It has effectively replaced Veo in the Gemini API: the Veo 3.1 preview models there were shut down on 22 October 2026, and Google’s docs now recommend Omni Flash for all video generation and editing in the API (GA Veo 3.1 stays available on Google Cloud’s Gemini Enterprise Agent Platform).

What makes Omni different from a classic text-to-video model is that it works like a conversation. You generate a clip, then say “make it night”, “swap the red bag for the blue one” or “change to 9:16”, and the model edits the existing clip rather than starting again. Version 1.2 extends that to the full 15-second length. Every output carries an invisible SynthID watermark.

What Changed From Omni 1.1

The quality claims (less morphing, better lip-sync, steadier cameras) are Google’s own and have not been independently benchmarked yet. Google also admits it is still working on fast sports action, rapid script changes and complex physics.

Pricing Per Second and Per Clip

Google bills Omni by tokens. Input (text, image, video, audio) is $1.50 per 1M tokens, text output $9.00 per 1M and video output $12.00 per 1M. Each second of video is a fixed number of tokens, which gives these per-second prices:

Clip costs are my arithmetic from the per-second rates and exclude the small input-token charge for your prompt and reference images. There is no free tier for Omni in the Gemini API. For how this fits with the rest of Google’s API pricing (and the Gemini price rises coming on 1 December and 1 January), see my Gemini API pricing guide.

Where You Can Use It (Free and Paid)

Not sure which Google plan you have? My Google AI Plus vs Pro vs Ultra comparison lists what each tier includes.

How to Use It Without Code

Option A: Google AI Studio (pay as you go)

1. Open aistudio.google.com, sign in, and set up billing on a Google Cloud project (Omni has no free API quota).
2. Pick Gemini Omni 1.2 Flash (gemini-omni-1.2-flash) from the model list.
3. Choose aspect ratio (16:9 or 9:16), resolution (start with 360p) and duration (3-15 s).
4. Optional: attach up to 6 reference images (a character, a product, a style) or a motion clip of up to 3 seconds.
5. Write the prompt: subject, action, camera move, lighting, sound and any dialogue in quotes.
6. Generate, then type follow-up edits in the same chat (“make it rain”, “slower camera”).
7. When you’re happy, re-render or upscale to 720p, 1080p or 4K and download the MP4.

Option B: Google Vids (no extra cost)

1. Open vids.google.com and create a new video.
2. Use the AI video generation option, describe your shot and generate a clip of up to 15 seconds.
3. Drop it on the timeline and add text, voiceover or other scenes.

Option C: Gemini app (AI Ultra)

In the Gemini app, choose the video tool and describe your clip. Rollout is gradual, so it may take a few days to show up on your account.

How to Use It in the Gemini API (Python)

Omni uses the Interactions API in the google-genai SDK. Install it and set your key:

pip install -U google-genai
export GEMINI_API_KEY="your-key"

Text to video

import base64
from google import genai

client = genai.Client()
interaction = client.interactions.create(
    model="gemini-omni-1.2-flash",
    input="A golden retriever runs along a beach at sunrise, slow tracking shot, waves and seagulls audible.",
    response_format={"type": "video", "resolution": "360p", "aspect_ratio": "16:9"},
)
with open("beach.mp4", "wb") as f:
    f.write(base64.b64decode(interaction.output_video.data))

Edit the same clip by chatting

edit = client.interactions.create(
    model="gemini-omni-1.2-flash",
    previous_interaction_id=interaction.id,
    input="Same shot, but at dusk with warm orange light.",
)

Image to video and extending a clip

# image to video
img = client.files.upload(file="product.png")
i2v = client.interactions.create(
    model="gemini-omni-1.2-flash",
    input=[{"type": "image", "uri": img.uri},
           {"type": "text", "text": "The bottle slowly rotates 360 degrees on a marble table."}],
    generation_config={"video_config": {"task": "image_to_video"}},
)

# extend an existing clip
clip = client.files.upload(file="my_video.mp4")
ext = client.interactions.create(
    model="gemini-omni-1.2-flash",
    input=[{"type": "video", "uri": clip.uri},
           {"type": "text", "text": "Continue the scene."}],
)

Useful settings: response_format.aspect_ratio takes "16:9" (default) or "9:16"; response_format.resolution takes 360p, 720p (default), 1080p or 4k; generation_config.video_config.task can be text_to_video, image_to_video, reference_to_video, edit or extend. For files over 4 MB, add "delivery": "uri" to response_format, poll client.files.get() until the state is ACTIVE, then call client.files.download(). Temperature, top_p, system instructions and negative-prompt fields are not supported, so write “no text on screen” and similar directly in the prompt. Code above follows Google’s official Omni guide examples; only the model ID changes from 1.1.

If you’re still on gemini-omni-1.1-flash, you don’t have to do anything: from 3 December 2026 Google routes those calls to 1.2 automatically. Switch now if you want the lower price and longer clips immediately.

The Cheap Draft-Then-Upscale Workflow

Google’s own advice is to iterate cheaply and only pay for quality at the end:

1. Draft at 360p ($0.023/s). Try 3-4 prompt variations of a 10 s clip for under $1 total.
2. Fix details with follow-up edits on the best draft (lighting, background, a product swap) instead of new generations.
3. Upscale only the winner to 1080p or 4K.

Example budget: four 10 s drafts at 360p ($0.93), two edits at 360p ($0.46) and one final 10 s at 1080p ($1.04) comes to about $2.43, against $2.80 for four straight 720p attempts that you can’t upscale cleanly.

Gemini Omni vs Veo 3.1 vs Kling vs Seedance vs Wan

Which to pick: early hands-on tests published by Segmind (which hosts the model) found Omni strongest at explainers, kinetic typography, product callouts, stylised looks and single-change edits to real footage, while Seedance 2.0 still looked clearly ahead for photoreal people, cinematic brand films and physics-heavy action. Omni is now the cheapest big-lab option with audio at 720p and far cheaper than Veo 3.1 on Vertex. If you want zero per-second cost and have a decent GPU, open models are the alternative: see my Wan 2.2 on 16 GB guide and LTX-2.5 GGUF setup. For phone-first editing, CapCut’s Seedance tools are the easiest route.

Limits, Regions and Safety Rules

• Regions: editing or extending uploaded videos is unavailable in the EEA, Switzerland and the UK (extending model-generated clips works). Uploading or editing images of minors is not supported there either.
• People: images of certain recognisable people are refused, and early testers report celebrity names are blocked. Real brand logos in source footage triggered a copyright block in Segmind’s tests.
• Uploaded clips: must be 10 seconds or less for editing or extension, and uploaded extension clips can’t add spoken dialogue.
• Extension only appends to the end; you can’t prepend or insert mid-clip.
• Audio: no uploaded audio inputs and no voice editing; audio in reference videos is ignored. Audio references are allowlisted on Google Cloud only.
• Language: only English is fully supported.
• Storage: setting store=false means you can’t edit that clip later with previous_interaction_id.
• Accuracy: Segmind’s tests saw about one instruction in four skipped (often a background change or last camera move), duplicated labels on a map, and chart points at the wrong heights. Check text and data carefully.
• Speed: Segmind measured 20-30 s for 360p text-to-video and 40-70 s at 720p; OpenRouter shows a median of about 48 s end to end.
• Provenance: every clip has a SynthID watermark, and Google’s policies expect AI video to be disclosed where required (for example on YouTube).

Troubleshooting

The response has no video URI, only base64

Google’s docs note that GET /v1beta/interactions/{id} returns inline base64 even if you asked for URI delivery. Take the URI from the original creation response (or the SSE stream) instead of re-fetching the interaction.

Large videos fail or time out when returned inline

Clips over 4 MB need "delivery": "uri" in response_format. Poll files.get until ACTIVE; if it shows FAILED, retry the request.

Follow-up edit fails with previous_interaction_id

The first request was probably made with store=false, so there is nothing to edit. Generate again with storage on.

“Not supported in your region” when editing my own video

Editing and extending uploaded videos is blocked in the EEA, Switzerland and UK. You can still generate new clips and extend clips the model made.

Refine button says “not available yet”

Segmind reported Refine was switched off at launch. Use a conversational edit instead.

The model ignores part of my prompt

Make one change per edit turn, put the most important instruction first, and describe camera moves explicitly (“slow dolly-in for the last 3 seconds”). Since there is no negative-prompt field, write exclusions in plain words.

Request blocked by safety filters

Remove real people’s names, recognisable faces, children and brand logos from prompts and references. Filters vary by region.

My Take

This is a smart, unglamorous release. Longer clips and a 30% price cut matter more to most creators than a new benchmark record, and $0.07 a second with audio undercuts Veo 3.1 on Vertex by more than five times. The 360p draft tier at about two cents a second is the real gift: you can explore ideas almost for free and only pay for the shot you keep. Conversational editing across 15 seconds is where Omni feels genuinely different from Kling or Seedance, and the free 15-second clips in Google Vids put it in front of a huge audience. The weak spots are clear too. Early tests say it skips instructions, struggles with photoreal people and fast physics, and European users lose the editing features that make it special. My advice: use Omni for explainers, product clips and social shorts, keep Seedance or Kling for cinematic human scenes, and move your 1.1 code to 1.2 now rather than waiting for the 3 December switch.

Frequently Asked Questions

How much does Gemini Omni 1.2 Flash cost?

About $0.023 per second at 360p, $0.07 at 720p, $0.104 at 1080p and $0.209 at 4K. A 15-second 720p clip is about $1.04.

Is Gemini Omni 1.2 Flash free?

Not in the API. You can make 15-second clips in Google Vids at no extra cost, and AI Ultra subscribers get it in the Gemini app.

How long can Gemini Omni videos be?

One generation runs 3 to 15 seconds. You can extend a clip by 3-10 seconds per turn, adding up to 30 seconds in total.

Does Gemini Omni 1.2 Flash make audio?

Yes. It generates synced audio with the video, including dialogue with lip-sync and multiple speakers. You can’t upload your own audio track as input.

What happens to Gemini Omni 1.1 Flash?

From 3 December 2026, requests to gemini-omni-1.1-flash are routed to 1.2 automatically.

Is Gemini Omni better than Veo 3.1?

In the Gemini API it has replaced Veo, and it costs far less ($0.07/s vs $0.40/s at 720p). Google hasn’t published a head-to-head quality benchmark.

Can I use Gemini Omni in Europe?

Yes for generating new clips, but editing or extending uploaded videos and uploading images of minors are not available in the EEA, Switzerland or the UK.

Sources

• Google: Gemini Omni 1.2 Flash, fast and cost-effective with longer clips
• Gemini API docs: Generate and edit videos with Gemini Omni Flash
• The Decoder: Gemini Omni 1.2 Flash stretches clips to 15 seconds
• OrcaRouter: Gemini Omni 1.2 Flash vs 1.1 Flash (token pricing and limits)
• OpenRouter: Gemini Omni 1.2 Flash pricing and latency
• Segmind: Gemini Omni 1.2 Flash tested (20 tests)
• KreatorFlow: AI video API price index (Veo, Kling, Seedance, Wan)