Google DeepMind's video generation and editing model, GA on November 5, 2026 — this site's first video-model topic: native 3–15s video with synchronized audio (whole-second steps; 1.1 capped at 10s), stateful conversational editing and 3–10s extensions per turn (30s total), up to 6 reference images. Billed per token: video output $12/1M tokens, 720p ~$0.07/s (15s ~$1.04), about 30% cheaper than 1.1. No free tier on the API, but Google Vids gives any Google account 3 free 15-second videos per month and Flow gives 50 credits a day. From Dec 3, 1.1 API requests auto-route to 1.2 at 1.2 prices. Independent measurement: AA text-to-video Elo 1154 (#6, tied with Seedance 2.5), editing Elo 1164 (#2), but image-to-video Elo 1086 (#14), well behind Seedance 2.5; Google published no benchmark scores of its own, so all quality claims are vendor claims. This page collects official specs, pricing, access channels and first tests, and marks what is still unverified.
The confirmed picture first: 1.2 is the second GA generation of the Gemini Omni family; the core changes are duration up from 10s to 15s, prices down ~30%, and the Dec 3 auto-routing. Its editing is strong, its image-to-video is weak, and all quality claims are still unverified by third parties.
previous_interaction_id; the model remembers the earlier video context and preserves unmentioned elements; each round produces a new videoThe Omni family went from its May I/O announcement to 1.2 GA in under six months; 1.1 lived only three months after its August GA before being auto-routed to 1.2 on Dec 3.
Google I/O announces the Gemini Omni family ("any-to-any", starting with video, image and audio output to follow); live the same day in the Gemini app, Flow and YouTube Shorts.
Gemini Omni Flash enters public preview on the Gemini API as gemini-omni-flash-preview.
Gemini Omni 1.1 Flash GA, adding scene extension (40s cumulative), first/last-frame control, a 360p draft tier and 1080p/4K upscaled output; the preview retires Sep 30.
Veo 3.1's Gemini API preview endpoints retire; the official migration target is Omni.
Gemini Omni 1.2 Flash GA (paid Gemini API tier + Google Cloud); official blog published; Google Vids switches to the 1.2 engine and opens free quotas the same day; Flow retires Veo Extend the same day.
Google Cloud blog published (by John Nack).
Artificial Analysis video board gains 1.2 scores (it wasn't listed on Nov 9).
1.1 API requests auto-route to 1.2 at 1.2 prices.
Flow retires Veo 3.1 Fast, Veo 3.1 Quality and Omni Flash 1.1.
This topic page goes live.
Per the model page and developer guide (read directly Nov 14, 2026). Note: 1080p and 4K are upscaled outputs, not native generation; aspect ratios are only 16:9 and 9:16; the API supports the Interactions API only.
previous_interaction_id; can change background, lighting, weather, aspect ratio, swap products and more; the model remembers the earlier video context and preserves unmentioned elements; each round produces a new video and consumes a generation + output tokensPOST https://generativelanguage.googleapis.com/v1beta/interactions), no generateContent; SDK client.interactions.create; large videos (>4MB, ~above 720p) should use delivery="uri"interactions.create call that consumes one request quotatop_p, stop sequences or negative prompts (you can write "don't do X" in the prompt); English is the only fully supported languageThe API has no free tier, but there are two free ways to try it: Vids gives 3 free 15-second videos a month, Flow gives 50 credits a day. Developers go via the API or AI Studio; for 15s and extension, Flow needs a paid subscription.
Model ID gemini-omni-1.2-flash, Interactions API only; API keys need a billing project. Google AI Studio can prototype it (the "Try it in Google AI Studio" button points at this model). Free tier not supported.
Available in Gemini for business and the console, supports Provisioned Throughput. Note: the Cloud blog says in-chat video generation inside Gemini for business is "Coming soon" — the GA covers the API and Cloud, not the in-chat feature.
Running 1.2 since Nov 5; the official blog says any Google account or Workspace subscriber gets up to 3 free 15-second videos per month; desktop. Note the conflicting figures: the Vids limits page says personal accounts get 60 seconds of AI video per month — 3 × 15s (45s) vs 60s are both Google pages, and Google doesn't explain the conversion; this page lists both and flags the conflict.
Omni 1.2 available to all eligible accounts (free tier 50 credits/day); without a subscription, generation/editing is capped at 10s; 15s, Extend and upscaling need Google AI Plus, Pro or Ultra; 4K upscaling is Ultra only. Plans (US): Plus $4.99/mo (200 credits/mo), Pro $19.99/mo or $199.99/yr (1,000 credits/mo), Ultra 5x $99.99/mo (10,000 credits/mo), Ultra 20x $199.99/mo (25,000 credits/mo); each tier also gets 50 daily credits. Flow requires age 18+ and a supported region.
15s video is "rolling out to Google AI Ultra subscribers" (official blog wording); Plus/Pro users are still at 10s; as of Nov 13 no announcement the rollout completed.
Higgsfield (listed as "Gemini Omni Flash 1.2"), Segmind (6 video APIs), OpenRouter (google/gemini-omni-1.2-flash), Runway, Picsart, Artlist, FLORA, invideo, Krea, Photoroom, QuickFrame AI; TeamoRouter hadn't added 1.2 as of Nov 8. Google Ads can scale video campaigns (the official blog mentions it in one sentence, with no further details).
Paid tier only, free tier "Not available". Video output $12/1M tokens, down from 1.1's $17.50 — about 30% cheaper (exactly 31.4%). Billed per output token, not per second — the pricing page footnote gives the conversion: 720p = 5,792 tokens/second.
| Item | Omni 1.2 Flash | Omni 1.1 Flash |
|---|---|---|
| Input / 1M tokens (text/image/video same price) | $1.50 | $1.50 |
| Text output / 1M tokens (incl. thinking tokens) | $9.00 | $9.00 |
| Video output / 1M tokens | $12.00 | $17.50 |
| Free tier | None | None |
| Batch pricing | None (Standard only) | None |
| Resolution | Tokens/s | 1.2 per second | 1.1 per second | 15s at 1.2 |
|---|---|---|---|---|
| 360p | 1,931 | $0.023 | $0.034 | ~$0.35 |
| 720p | 5,792 | $0.07 | $0.101 | ~$1.04 |
| 1080p (upscaled) | 8,688 | $0.104 | $0.152 | ~$1.56 |
| 4K (upscaled) | 17,376 | $0.209 | $0.304 | ~$3.13 |
The exact 15s 720p figure is $1.04256 (15 × 5,792 × $12/1M); the rounded $0.07 × 15 gives $1.05 — a different rounding, so this page writes $1.04 and states the basis. RanThis measured $1.05 including thinking tokens. The 1.1 1080p/4K per-second rates ($0.152/$0.304) are this site's calculation from official token rates, not Google-published per-second prices. Audio has no separate charge (native audio generated by default); input and text output are billed separately.
360p 1 credit/s, 720p 3 credits/s (first 10s); seconds 11–15 at 360p 2 credits/s, 720p 6 credits/s; 15s 720p = 60 credits, 15s 360p = 20 credits; extensions 360p 1 credit/s, 720p 3 credits/s; 720p/1080p upscaling free on paid tiers for clips ≤10s, 1 credit/s beyond 10s; 4K upscaling Ultra only, 50 credits for ≤10s, then 5 credits/s. Free tier: 50 credits/day.
Official pages: Google official blog · Gemini API pricing · Developer guide
In the week after launch, four self-funded tests agreed: instruction-following is strong, chart animation and single-point edits on real footage are strongest; but a 360p draft and the same-prompt 720p are completely different scenes — 360p cannot be a cheap preview; and conversational editing failed on every attempt via an aggregator.
15s 720p measured $1.05 (86,880 video output tokens = exactly 15×5,792, plus ~$0.01 thinking tokens); generation 59–86s; English–Vietnamese three-line dialogue verbatim accurate, Vietnamese signage (with stacked tone marks) spelled perfectly; but the Vietnamese clip inexplicably carried Vietnam national TV (VTV)'s logo; the 360p draft and same-prompt 720p are completely different scenes (different actors, room, whiteboard) — 360p cannot be a cheap preview of 720p; the correct workflow is "upscale the draft you like", not "preview in 360p then re-run the prompt at 720p".
The conclusion: "the most instruction-following video model tested" — narration, typography, chart animation and single-point edits on real footage strongest; cinematic feel and realistic people weaker; Segmind platform prices ~$0.026/s (360p), $0.075/s (720p), $0.083/s (720p editing), slightly above Google's official prices; chart animation needs the first+last-frame method (prompt-only numbers misplace/dropped labels); the copyright filter rejects outright when source material contains real brand logos even if the prompt doesn't mention the brand.
15s uncut FPV long take succeeded; 6s clips 45–55s, 15s clips 81s; prompts involving children rejected and not charged; Polish dialogue works; but conversational editing failed 5/5 via the aggregator (aggregator limitation, not necessarily the model).
10s clip + extension = 20s video, seam consistent, but an unwanted dissolve appeared ~3s into the extension.
Six in-depth articles from this site (Chinese + English), one question each, rewritten from public sources with a reference link.
Google published no benchmark scores of its own — all quality claims are vendor claims. Independent measurement comes from AA and LMArena, two blind-vote boards — but their image-to-video conclusions are opposite, so they must be cited separately, never merged into one ranking.
| Board | Omni 1.2 Flash | Seedance 2.5 | Notes |
|---|---|---|---|
| Text-to-video | 1154 ±14 (#6) | 1155 ±13 | Statistically tied; 1.1 was 1127 (#12) — 1.2 rose from #12 to #6 |
| Image-to-video | 1086 ±15 (#14) | 1161 ±15 (#1) | A 75-point gap; 1.1 was #17 |
| Video editing | 1164 ±12 (#2) | 1147 ±11 | Top is Wan 3.0 (1192); 1.2 is #1 in both complex-editing and VFX-editing sub-items |
Omni 1.2 Flash is #1 in text-to-video (1540 ±16, 1,714 votes) and #1 in image-to-video (1503) — contrasting sharply with AA's image-to-video #14. The two boards differ in sample size and method; this page cites them separately and never merges them into a single "image-to-video ranking". Note the timeline: 1.2 wasn't on the AA board on Nov 9 and appeared around Nov 10; AA recalibrates Elo periodically, so scores from different dates aren't directly comparable.
Google says 1.2 generates faster, follows references better, keeps characters/objects/scenes more consistent, produces more natural dialogue, steadier camera motion and fewer jump cuts — AccessAllGPT explicitly says it didn't reproduce these quality claims. Officially admitted weak spots (blog): fast sports action, fast script changes, complex physics.
Omni 1.2 is positioned for volume and editing: ~5.7x cheaper than Veo 3.1 with longer clips; ~3.3x cheaper than Seedance 2.5 with stronger editing, but 75 Elo points behind on image-to-video.
| Dimension | Omni 1.2 Flash | Veo 3.1 Standard | Seedance 2.5 |
|---|---|---|---|
| Single clip | 3–15s (extend to 30s) | 4/6/8s | Up to 30s (two extensions) |
| Resolution | 360p–4K (1080p/4K upscaled) | 720p/1080p/2160p | Up to 1080p (4K via Runway/Picsart) |
| 720p per second | $0.07 | $0.40 (with audio) | $0.231 |
| 10s 720p | $0.70 | — (8s 1080p $3.20) | $2.32 |
| AA T2V | 1154 (#6) | Nov 9 read: ~164 points below Omni 1.1 | 1155 (statistically tied) |
| AA I2V | 1086 (#14) | — | 1161 (#1) |
| AA editing | 1164 (#2) | — | 1147 |
| Reference assets | Up to 6 images / 3s video | Reference images (forces 8s) | Up to 50 (30 images + 10 videos + 10 audio) |
| Conversational editing | Yes (Interactions API) | No | — |
| Free tier | API: none; Vids/Flow have free quotas | None | — |
Veo 3.1 (Google's own flagship): public preview Oct 15, 2025; 4/6/8s durations, 24 FPS, 720p/1080p/2160p, 16:9 and 9:16; 1080p, 4K, reference images and extension all force 8s. Pricing (Google Cloud): standard with audio $0.40/s (720p/1080p), 4K $0.60/s, silent $0.20/s; Fast $0.10/s (720p)/$0.12/s (1080p)/$0.30/s (4K); Lite $0.05/s (720p)/$0.08/s (1080p). Runway Dev lists Fast at $0.15/s, conflicting with Cloud's $0.10/$0.12 — this page uses Cloud's and flags the difference. On Oct 22, 2026, Veo 3.1's Gemini API preview endpoints all retired, with Omni as the official migration target; Veo 3.1 is still usable via Google Cloud and Flow (Flow keeps Lite only after Dec 5). Google hasn't confirmed whether Omni reuses Veo architecture — any such claim is unverified.
Seedance 2.5 (ByteDance Seed): released Jul 31, 2026; single generation up to 30s; up to 50 reference assets per generation (30 images + 10 videos + 10 audio); joint audio-video generation; max 1080p (BytePlus API; 4K from Nov 11 via Runway and Picsart). Pricing (BytePlus ModelArk official, read Nov 13, 16:9): 480p $0.103/s, 720p $0.231/s, 1080p $0.569/s; 10s 720p = $2.32, ~3.3x Omni 1.2's $0.70. Seedance 2.5 Flash: announced Nov 12, API available Nov 15 14:00 (UTC+8); 720p $0.15/s, no 1080p; model ID dreamina-seedance-2-5-flash-261115; no AA score as of Nov 12. See this site's Omni vs Veo 3.1 article for detail.
The language model in the same Gemini family is on this site's Gemini 4 Carbon topic (Google's strongest model, GA Nov 10).
Nine days after launch, these are still unknown: some await a Google statement, others need more hands-on testing.
model field; after Dec 3 the docs offer no way to keep getting 1.1 outputConfusion around 1.2 comes mostly from four places: the two December dates, the two Vids quota figures, the role of the 360p draft, and the scope of the Gemini for business GA. One by one.
They're not. Dec 3: 1.1 Gemini API requests auto-route to 1.2 at 1.2 prices. Dec 5: Flow retires Veo 3.1 Fast, Veo 3.1 Quality and Omni Flash 1.1 (kept at standard credit prices until then). Two different dates, two different platforms — don't confuse them. Also note: Dec 3 routing applies only to the Gemini API ID; on Google Cloud, 1.1's ID is gemini-omni-1.1-flash-preview, marked deprecated with 1.2 recommended, but no date given.
3 × 15s is 45s, not 60s. The official blog says 3 free 15-second videos per month; the Vids limits page says personal accounts get 60 seconds of AI video per month. Both are Google pages, and Google doesn't explain the conversion; this page lists both and flags the conflict. A Workspace Updates post reportedly says the 15s feature rolls out to Pro/Ultra personal accounts and some Workspace editions, in tension with the blog's "any Google account" wording (the original post wasn't read).
The Google blog suggests drafting in 360p and upscaling, but RanThis's test proves 360p is an independent generation — the same prompt at 720p gives a completely different scene (different actors, room, whiteboard). The correct workflow is "upscale the draft you like", not "preview in 360p then re-run the prompt at 720p".
The GA covers the Gemini API and Google Cloud; the Cloud blog says in-chat video generation inside Gemini for business is "Coming soon" — not live yet. An early brief overstated the GA scope; this page corrects it.
Felo's claim, sharply inconsistent with AA's 1154 (#6) on Nov 12 — and Felo itself notes "Scores as published by the model provider". This page doesn't use it.
Same model: the official name is Gemini Omni 1.2 Flash, the family name is Gemini Omni Flash, and the in-product form is Gemini Omni Flash 1.2 (used by Flow help pages and third-party platforms like Higgsfield). The Google Cloud blog once miswrote it "Gemini 1.2 Omni Flash" — an official typo, not used here.
Google says 30%; the exact figure is 31.4% ($12.00 ÷ $17.50). The 720p per-second rate fell from $0.101 to $0.07, a 30.7% cut. This page writes "about 30% (exactly 31.4%)".
A single forum rumor (an HN user, relayed by turangga), unverified. The official Flow credit price is 3 credits/s for 720p in the first 10s, so a 10s clip is 30 credits — matching the rumored "rose to 30" — but the 15-credit figure for 1.1 has no official source.
November 5, 2026 GA (paid Gemini API tier and Google Cloud, available the same day; official blog Nov 5, Google Cloud blog Nov 6). It's the second GA generation of the Gemini Omni family: 1.1 went GA on August 27, 2026, and the preview launched June 30. See the section 02 timeline.
Yes, three names for the same model: the official name is Gemini Omni 1.2 Flash; the family name is Gemini Omni Flash (the model-page and developer-guide title, and the highest-search-volume form); the in-product form is Gemini Omni Flash 1.2 (used by Flow help pages and third-party platforms like Higgsfield). The Google Cloud blog once miswrote it "Gemini 1.2 Omni Flash" — an official typo, not used here.
About $1.04: billed per output token, video output $12/1M tokens, 720p = 5,792 tokens/second, so 15 seconds is 86,880 tokens, exactly $1.04256. Note the rounded $0.07 × 15 gives $1.05 — a different rounding; RanThis measured $1.05 including thinking tokens. 15s at 360p ~$0.35, 1080p ~$1.56, 4K ~$3.13. See this site's pricing guide.
No free tier on the API (free tier "Not available"). Two free ways to try it: Google Vids — the official blog says any Google account or Workspace subscriber gets up to 3 free 15-second videos per month; Google Flow — the free tier gives 50 credits a day (generation/editing capped at 10s without a subscription). Note the Vids limits page also gives a 60-seconds-per-month figure, conflicting with 3 × 15s (45s); this page lists both and flags the conflict. See this site's access guide.
1.1 API requests auto-route to 1.2: all requests to gemini-omni-1.1-flash are served by 1.2 at 1.2 prices — no code change needed, no error, but output comes from a different model. Three notes: routing applies only to the Gemini API ID (Cloud's 1.1 ID and schedule differ); Google hasn't said whether routed responses report 1.1 or 1.2 in the model field; after Dec 3 the docs offer no way to keep getting 1.1 output. Dec 5 is Flow's 1.1 retirement date — don't confuse the two. See this site's migration guide.
Video output fell from $17.50 to $12.00/1M tokens — exactly 31.4%, Google says 30%; the 720p per-second rate fell from $0.101 to $0.07. Duration expanded from 10s to 15s, and paid rate limits doubled. But the total extension cap fell from 1.1's 40s to 1.2's 30s — an easy-to-miss downgrade at migration. See this site's migration guide.
Veo 3.1 is the quality flagship: single 8s clips, standard with audio $0.40/s (720p/1080p), ~5.7x Omni's 720p price; Omni 1.2 is volume/editing-oriented: 15s, conversational editing, cheap. Pick Veo 3.1 for the highest-quality single 8s shot; pick Omni 1.2 for volume, longer clips and multi-round editing. Note Veo 3.1's Gemini API preview endpoints retired on Oct 22, with Omni as the official migration target; Veo 3.1 is still usable via Google Cloud and Flow. See this site's comparison article.
On AA: text-to-video tied (1154 vs 1155), image-to-video Omni behind by 75 points (1086 vs 1161), editing Omni ahead by 17 (1164 vs 1147). Price: 10s 720p costs $0.70 on Omni vs $2.32 on Seedance 2.5 (~3.3x). Seedance generates up to 30s in one go with up to 50 reference assets; Omni caps at 15s per clip (extend to 30s) with up to 6 reference images, but has conversational multi-round editing and a lower price.
No. The Google blog suggests drafting in 360p and upscaling, but RanThis's test proves 360p is an independent generation — the same prompt at 720p gives a completely different scene (different actors, room, whiteboard). The correct workflow is "upscale the draft you like", not "preview in 360p then re-run the prompt at 720p".
No. Users in the EEA, Switzerland and the UK currently cannot edit or extend uploaded video (editing/extending model-generated video still works there); the same regions also don't support uploading/editing images containing minors; images with certain identifiable people are unsupported for upload/edit. From the developer guide's Limitations section, read directly Nov 14, 2026.
Interactions API only: POST https://generativelanguage.googleapis.com/v1beta/interactions, no generateContent; SDK client.interactions.create; large videos (>4MB, ~above 720p) should use delivery="uri". Conversational editing references the previous round with previous_interaction_id. Rate limits: Tier 1 8 RPM / 40 RPD; Tier 2 16 RPM / 200 RPD; Tier 3 40 RPM / 2,000 RPD. See this site's access guide.
Same Gemini brand, different jobs: Omni 1.2 Flash is a video generation and editing model, Gemini 4 Carbon is a language model (Google's strongest, GA Nov 10, built for long-horizon agents and cybersecurity). Google Trends shows cross-interest between the two. See this site's Gemini 4 Carbon topic.
Every fact on this page comes from the public sources below (checked Nov 14, 2026).