Quick Verdict
- Best for
- Most realistic output (via its successor, Gemini Omni 1.2 Flash)
- Starting price
- ≈$0.07/s at 720p via the Gemini API
- Free tier
- No free tier
- Price verified
- Nov 2026
Why consider it
- Most realistic video generation
- Native audio with dialogue
- Google ecosystem integration
Limitations
- 15-second clips max (was 8s on Veo 3.1)
- No free tier
- Paid API tier only
Google shut down the Veo 3.1 preview API endpoints on 2026-10-22. The replacement is Gemini Omni: the current model is gemini-omni-1.2-flash, generally available since November 2026, generating up to 15 seconds per clip with native audio at about $0.07 per second (720p). The old Veo 3.1 three-tier per-second pricing is gone; the rates below are the current Omni token rates. Prices as of Nov 2026; check ai.google.dev for current API rates.
What Happened to Veo 3.1
Veo 3.1 was Google DeepMind's video model, and version 3.1 made native audio the default across all three quality tiers. Dialogue, environmental sound, and effects generated together with the visuals rather than as a separate pass. That audio capability also extended to features that were silent in 3.0, including Ingredients to Video, Frames to Video, and clip extension.
On 2026-10-22 Google shut down the three Veo 3.1 preview API endpoints (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview). The recommended replacement was the Gemini Omni endpoint gemini-omni-1.1-flash, itself succeeded by gemini-omni-1.2-flash. Effective December 3, 2026, all API requests to gemini-omni-1.1-flash automatically route to gemini-omni-1.2-flash and are billed at the lower 1.2 rates — so for new work, 1.2 is the model to target.
What carried over from Veo: reference-image consistency (now up to 6 reference images), first-and-last-frame interpolation (3–10 seconds), and extension (per-turn video extension up to 30 seconds total). What changed: clips run up to 15 seconds instead of 8, and editing is conversational — you refine and extend videos through natural-language turns via the Interactions API, rather than chaining fixed 8-second shots in Flow alone.
Omni 1.2 is available through the Gemini API (paid tier) and Google AI Studio. The Gemini app and Flow remain the consumer surfaces for Google video generation.
Pricing and Access
Developers pay per token on the Gemini API (verified on Google's official pricing page, Nov 2026):
- Input: $1.50 per 1M tokens (text / image / video)
- Output: $9.00 per 1M tokens (text); $12.00 per 1M tokens (video)
- Video output is billed at 5,792 tokens per second of 720p video, which the official page converts to about $0.07/sec at 720p (~$0.70 for a 10-second clip, ~$1.04 for a 15-second clip); 360p $0.023/s, 1080p $0.104/s (upsampled), 4K $0.209/s (upsampled)
There is no free API tier; the official pricing page lists free tier as not available for the Omni video models. A 15-second 720p clip costs about $1.04; the same length at 1080p costs about $1.56.
Price-change notice (Nov 2026): Google's Gemini API paid prices rise on a schedule — the flagship Standard tier doubles Dec 1, 2026; Flash and text-to-speech double Jan 1, 2027 (old TTS prices run through Dec 31, 2026). The Omni video rates above are the current 1.2 rates and carry no doubling schedule on the official page; still, re-check the official pricing page before budgeting.
Consumers access Google video generation through Google AI Pro at $19.99/mo and Google AI Ultra at $249.99/mo, usable in the Gemini app and Flow (subscription details verified Oct 2026). Reports of free consumer credits exist but the numbers conflict across sources and Google publishes no uniform figure, so do not count on a specific free allowance.
Archive — Veo 3.1 preview API rates, no longer available (official page, checked 2026-10-05): Fast $0.10/s (720p), $0.12/s (1080p), $0.30/s (4K); Standard $0.40/s (720/1080p), $0.60/s (4K); Lite $0.05/s (720p), $0.08/s (1080p). The old Standard tier cost about $3.20 for an 8-second 1080p clip; the same shot on Omni 1.2 costs about $0.83 at 1080p.
Output Quality
Reviewers consistently rated Veo 3.1's fluid physics simulation as the best in class: water, smoke, and fabric move convincingly. Photorealism led the field, especially for natural scenes and human faces at rest. The synced native audio was the real differentiator; a street scene arrived with matching street noise, and dialogue roughly matched lip movement. Omni 1.2 inherits that lineage — it is Google's current video model, built on the same approach with longer clips and conversational editing — but treat the quality claims as inherited from Veo 3.1's track record, not as fresh hands-on tests of 1.2.
Limits to know: the 15-second cap is strict per generation, so narrative work still means planning in shots and extending. Lip-sync on dialogue still needs human review. Omni 1.2 is generally available rather than preview-labeled, but Google replaced its video model twice in four months (Veo 3.1 → Omni 1.1 → Omni 1.2), so production users should still build fallbacks. And while per-second costs fell sharply versus Veo Standard, retries still multiply the bill — drafting at low resolution stays the cheapest habit.
Who It Is For
Anyone whose top priority is realism: commercials, establishing shots, product visuals. Developers building video features who want a real API with documentation. Google ecosystem users who already pay for AI Pro or Ultra. Filmmakers comfortable storyboarding in 15-second shots and extending through the Interactions API.
Who Should Skip It
Long-form creators who do not want to stitch clips should look at Kling 4.0 (30 seconds) or script-to-video tools. Budget experimenters get no free API tier, so testing costs money from the first second — though at $0.023/s for 360p drafts, a test is cents, not dollars. Teams that needed a stable, non-preview API can use Omni 1.2 (generally available), but should still build redundancy: Google's video lineup turned over twice in four months.
Compared With Alternatives
Against Kling 4.0, Omni wins on realism and audio quality but loses on clip length (15s vs 30s) and free access. Against Runway, Omni wins on visual fidelity while Runway wins on editing control and clip manipulation — and Runway now also routes third-party models including Kling 3.0 and Seedance 2.5 in the same subscription. Against Pika, Omni is the quality flagship and Pika is the fast, fun alternative.
Pricing Deep Dive: Draft Cheap, Render Once
Omni's per-resolution rates exist for a workflow: draft at 360p ($0.023/sec) for rapid prompt tests, preview at 720p ($0.07/sec), render finals at 1080p ($0.104/sec) or 4K ($0.209/sec). A creator who drafts ten 8-second tests at 360p spends $1.84; the same tests at 4K cost $16.72. The discipline of drafting low and rendering high is the difference between Omni being affordable and Omni being shocking.
For consumers, the math runs through Google subscriptions. AI Pro at $19.99/mo includes video access in the Gemini app and Flow with a credit pool; Ultra at $249.99/mo multiplies that pool roughly 25x. Neither Google nor third parties publish exact video credit counts, so heavy users should test consumption in the first month before assuming a tier covers their output.
The 15-second cap interacts with pricing in a useful way: every generation is a bounded cost. A 15-second 720p clip is always about $1.04. There are no surprise 60-second renders. Budgeting is simple arithmetic once you know your shot count.
Real-World Workflow
Omni projects are planned in shots. Storyboard the sequence, generate each shot (up to 15 seconds), then extend or stitch: per-turn video extension runs up to 30 seconds total, and first/last-frame interpolation bridges specific start and end frames you design. For character consistency across shots, use up to 6 reference images.
Google replaced its video model twice in four months, so build fallback logic (retry on failure, degrade resolution) rather than assuming 100% stability. Reviewers also flag that dialogue lip-sync needs human checking; do not ship client dialogue without watching it.
For developers, Vertex AI offers an enterprise path with separate SKUs, including video-only options without audio at lower rates. That matters for pipelines where you add your own sound design anyway.
Choosing a Resolution
360p is genuinely useful for composition and motion tests; the resolution is the limit, not the model's understanding. 720p at about $0.07/sec is the sweet spot for most preview work. 1080p earns its premium on final renders where fine detail (fabric texture, facial nuance, small text) must hold up. 4K output is upsampled, so reserve it for hero shots rather than every take.
Bottom Line
Veo 3.1 is gone; its successor Gemini Omni 1.2 Flash is the current quality ceiling of Google video generation, and it costs less per second than Veo Standard did (about $0.07/s at 720p versus $0.40/s). If your shots fit in 15 seconds and realism is the priority, use Omni 1.2. If you need length or volume, Kling 4.0 or script-to-video tools will serve you better for less money.
Access Paths Compared
The Gemini app is the simplest entry: prompt in chat, get video. Flow is Google's filmmaking interface and the only consumer surface with Scene Extension, reference management, and proper project organization; anyone making multi-shot work should learn Flow rather than fighting the chat interface. The Gemini API suits developers who need programmatic generation with token billing (about $0.07/s at 720p). Vertex AI is the enterprise route with compliance features and volume arrangements.
Each path bills differently, so a team using both Flow (subscription credits) and the API (token charges) should track the two pools separately. The most common beginner mistake is prototyping in the app, then being surprised by API costs when moving to production, or vice versa.
Bottom Line for Buyers
Omni 1.2 asks you to accept two constraints (15 seconds, no free API tier) in exchange for the best-looking Google video available. Creators who storyboard in shots will barely notice the constraints. Everyone else should weigh whether the realism premium beats Kling's longer, cheaper clips for their specific output.
Revisit the decision yearly; this market reprices and remodels faster than any other software category right now.
