A generated title card under the eyebrow '20 November 2026 - consumer rollout', titled 'Gemini Omni 1.2 Flash Is Now a GIF Button', with the subtitle 'Subscriber-only, eight countries, app-only, and not one line about it in Google's Gemini API changelog.' Three numbered rows below read: the model went generally available on 5 November 2026, sixteen days before the GIF post; a three-second 720p render costs about $0.21 in output tokens at $12.00 per million; and there is no GIF response format in the API, so the conversion happens in the app layer. The OrcaRouter logo is composited bottom-right.
Guides & Insights

Gemini Omni 1.2 Flash Makes GIFs Now: Google's Docs Don't Say So

Author

jinhao.song

Date Published

Latest models · 20View all models →
Benchmarks: Artificial Analysis · updated daily
Back to all posts

On the evening of 20 November 2026, the @testingcatalog account on X posted a screenshot of the G​emini app's composer with a third creation card in it. Beside "Restyle" and "Make a picture" sat "Create GIF: Pelican riding a bicycle." The post said G​oogle AI subscribers can now generate GIFs in G​emini, that the feature is powered by Gemini Omni 1.2 Flash, and that it is live in the U.S., Japan, Korea, Australia, Canada, the U.K., Germany and France. The model is not the news here. Gemini Omni 1.2 Flash went generally available on 5 November 2026, sixteen days before that post. What is new is that G​oogle has put a video model in front of consumers as a GIF button, behind a subscription, in eight countries, and has documented none of it on any page we could find.

That gap is worth taking seriously rather than treating as a formality. A model release has a changelog entry, a model card and a pricing row. A consumer feature bolted onto a model has whatever the app ships, and the documentation catches up later or never. Anyone planning around Gemini Omni 1.2 Flash — for a product, a pipeline, a budget — needs to know which of those two things just happened. As of 21 November, it is the second.

What shipped on 20 November

The evidence is a single third-party post and the screenshot inside it. That is thin, and it is what exists. The image shows the standard G​emini app composer with its prompt-suggestion cards, one of which is a GIF prompt, which tells you the feature is a creation mode in the consumer product rather than something you call. It does not show a duration limit, a resolution, a file size, an export path, or a cost. Those details are not in the post and not on G​oogle's site.

What the post does establish, and what the country list corroborates, is the shape of the rollout: subscriber-gated, geo-gated, and app-only. Three gates at once is the signature of a staged consumer experiment, not a platform launch. If G​oogle had shipped GIF as a capability of the model, the natural place to say so would be the G​emini API changelog — the same page that carried the 5 November entry for Gemini Omni 1.2 Flash. That entry is still the most recent word on the model, and it does not mention GIF.

Why a video model is behind a GIF button

The pairing is less odd than it sounds. Gemini Omni 1.2 Flash generates three to fifteen seconds of video at any whole second, defaults to ten, and produces native synchronized audio alongside the picture. A GIF is a short, silent, looping animation, and the first two of those properties — short and animated — are exactly what the model already makes. Turning a three-second render into a GIF is a transcode: decode, drop the audio track, quantise to a 256-colour palette, loop it. Nothing in that pipeline needs a new model, which is why a GIF feature can appear on top of an existing video endpoint without the endpoint changing.

The same reasoning explains why GIF is not an obvious API output type. The API takes a response modality — video, with an optional URI for files over 4MB — and hands back a container. A GIF is a container choice, not a modality, and G​oogle's guide documents no GIF response format for the model. If the app is producing GIFs on top of Omni 1.2 Flash, the conversion is happening in the app layer, after the model returns.

It also explains the transcription of the model's headline numbers into this context. G​oogle's 5 November changelog described "up to 15-second video generation and conversational editing," a 30% cut in video output pricing to $12.00 per million tokens, doubled rate limits on paid tiers, and support for up to six reference images for character and subject consistency. In a GIF button, most of that is irrelevant. Fifteen seconds is far longer than a GIF should be, conversational editing is not exposed in a one-shot creation card, and reference images are not part of the evidence.

The eight countries, and the paywall

The country list is the part of the post that reads like a real rollout rather than a demo, because it is oddly specific. Eight markets, and the composition is telling:

• North America — the U.S. and Canada, both included.

• Asia-Pacific — Japan, Korea and Australia, with India and Indonesia absent despite being large G​emini markets.

• Europe — the U.K., Germany and France only, which is three of the four largest EU-plus-UK markets and none of the smaller ones.

• Absent entirely — the rest of the EEA, Switzerland, India, Brazil, and everything in Latin America outside Canada and the U.S.

Compare that with a restriction G​oogle does document for this model: editing and extending uploaded video is unavailable in the EEA, Switzerland and the U.K. The GIF list puts the U.K. in and the wider EEA out, so if the app feature follows the same regulatory logic the U.K. position differs between the API and the app. That is a real inconsistency and worth flagging rather than resolving — nothing published explains it.

The subscriber gate is the simpler half. Gemini Omni 1.2 Flash has no free tier on the API at all; every row of the pricing page reads "Not available" in the free column. So a subscription wall in front of a GIF button is consistent with how G​oogle already prices the model. The question a reader in one of the eight countries will actually have is not whether it costs anything but whether a GIF counts against an app quota or is free once you are paying. Neither the post nor G​oogle's pages answer that.

Screenshot of the Gemini Omni 1.2 Flash pricing block on Google's Gemini API pricing page, showing the Standard tier table with 'Input price (text / image / video) - Not available on the Free Tier, $1.50 on the Paid Tier per 1M tokens' and 'Output price including thinking tokens - Not available on the Free Tier, $9.00 (text) / $12.00 (video) on the Paid Tier'.

What a three-second GIF costs to render

Subscribers do not see a bill, so this arithmetic is about what the underlying render is worth, not what anyone is charged. G​oogle bills video output by the second, at a fixed token rate per resolution, and publishes the constants. At 720p the model emits 5,792 video output tokens per second of footage, and video output is $12.00 per million tokens.

• Three seconds at 720p — 17,376 tokens, about $0.21 of output.

• Five seconds at 720p — 28,960 tokens, about $0.35.

• Ten seconds at 720p — 57,920 tokens, about $0.70, which matches G​oogle's own "~$0.70 for a 10s clip" line and confirms the constant.

• Three seconds at 360p — 1,931 tokens per second, so 5,793 tokens, about $0.07.

• Three seconds at 1080p — 8,688 tokens per second, so 26,064 tokens, about $0.31, though 1080p and 4K are both upsampled from a lower render rather than native.

Two observations follow. First, the per-GIF numbers are small enough that a subscription comfortably covers them, which is presumably why the feature is gated on a subscription rather than metered. Second, resolution is the only real lever: the same three seconds costs three times as much at 1080p as at 720p, and a GIF's colour quantisation discards most of what that extra resolution bought. Whoever built the app feature had an obvious reason to render low and quantise.

A GIF throws away the model's best parts

This is the honest limitation of the whole framing. Two things make Gemini Omni 1.2 Flash interesting as a video model — native synchronized audio and a fifteen-second ceiling — and the GIF format discards both by definition. GIF has no audio track and no standard for one. A looping animation is also the one video form where length is a liability, since every extra second adds file size to something people expect to load instantly.

So the capability a subscriber gets is the model's visual coherence at a length and format that strips its audio and its duration advantage. That is not a criticism of the feature; a pelican on a bicycle is exactly what a three-second looping GIF is for. It is a caution against reading the feature as evidence that Omni 1.2 Flash's video quality has arrived in the consumer app. Three seconds of quantised, silent, looping animation is a narrow window onto a model whose specification is fifteen seconds with dialogue, lip-sync and multi-speaker conditioning.

There is also a provenance point worth knowing. Every output from this model carries a SynthID watermark, per G​oogle's documentation. Whether a SynthID mark survives GIF re-encoding and palette quantisation is not something G​oogle has published, and it is the kind of question that matters more for an easily-shared looping file than for a video. Treat it as open. The safe assumption is that a GIF made this way is still G​oogle-generated content, but do not assume the watermark is intact in the file you downloaded.

A generated single-column scoreboard titled 'Gemini Omni 1.2 Flash - the scoreboard', with rows reading 'Generally available: 5 November 2026', 'Duration: 3 to 15 seconds, 10 by default', 'Video output: $12.00 per million tokens', '720p cost: about $0.07 per second', 'Free tier: none' and 'GIF in the API: no documented format'. The footer reads 'Model, duration and pricing figures per Google's Gemini API pages, read 21 November 2026; the GIF button is per a single X post and is undocumented by Google.' The OrcaRouter logo is composited bottom-right.

Not in the changelog, not in the model card

The documentation gap is specific enough to list, because each absence is a place a reader would reasonably look first:

• The G​emini API changelog — the 5 November 2026 entry for Gemini Omni 1.2 Flash covers duration, conversational editing, extension to 30 seconds, reference images and the price cut. No GIF mention, and no later entry.

• The G​emini app release notes — no GIF entry.

• The model card — describes output as video generation and editing up to 15 seconds with native audio, keyframe interpolation and video extension. No image-sequence or GIF output type.

• The pricing page — three rows for this model: input at $1.50 per million tokens, text output at $9.00, video output at $12.00. No per-image or per-GIF line.

Absence of documentation is not absence of a feature, and this is the case where the distinction bites. The screenshot is not fabricated: the composer UI, the prompt cards and the wording are consistent with the app as it ships today. What is missing is any G​oogle statement of limits, retention, cost or availability window. A feature that exists only in a client build can be pulled in a week, and staged consumer experiments frequently are.

If you need this from an API, you don't have it

That is the practical line for anyone building rather than tapping. There is no GIF endpoint to call — not on G​oogle's API, and therefore not through any gateway, including ours. OrcaRouter does not serve Gemini Omni 1.2 Flash; we checked the catalogue, and the Omni endpoints are not in it. If Omni is the model you need, G​oogle's own API is the path, and saying so is more useful than implying otherwise.

What a router is good for here is the layer around a video pipeline, not the render itself. Prompt expansion, shot-list generation, captioning, moderation and post-hoc auditing are all text jobs, and they are the ones that get pinned to whatever model happened to be up when the code was written. One key across more than 200 models with automatic failover moves that risk off your critical path, and because OrcaRouter bills at 0% markup with the provider list price passed straight through, a vendor price change on the models you do call lands on our side the same day it lands on theirs. The 1.1-to-1.2 video price cut is the example: $17.50 down to $12.00 per million output tokens, no repricing on our end to wait for.

For the GIF question specifically, the honest answer for the next few weeks is to watch the API changelog rather than the app. If GIF output appears as a documented response format, it becomes routable and the arithmetic in this piece becomes a budgeting exercise. Until then it is a consumer feature with a country list.

Screenshot of Google's Gemini API changelog showing the 'Gemini Omni 1.2 Flash generally available (GA)' entry, naming the model id gemini-omni-1.2-flash, up to 15-second video generation, 3 to 10 second per-turn extension up to 30 seconds total, up to 6 reference images, a 30% lower video output price at $12.00 per 1M tokens and doubled rate limits, with the following notice that gemini-omni-1.1-flash requests route to 1.2 from 3 December 2026.

What to watch

Three signals would settle this. A line in the G​emini API changelog describing an animated-image or GIF output format would mean the capability reached the platform, and that is the one that changes what a developer can build. A G​emini app release note naming the feature, with duration and size limits, would confirm it is more than a staged experiment. And a change in the eight-country list — expansion into the rest of the EEA, or a contraction — would tell you whether G​oogle is testing a market or a feature.

None of those exist as of 21 November 2026. What exists is a screenshot, a date, a country list, a paywall, and a sixteen-day-old model that has quietly acquired a second face in the consumer app. Gemini Omni 1.2 Flash is still, on paper, a fifteen-second video model at $12.00 per million output tokens with no free tier. The GIF button does not change that specification. It changes who can reach the model without writing a request, which is a different kind of news and worth naming as such.