Updated November 2026
How to compare AI video models
Compare AI video models on seven things: the longest single clip, resolution, native audio, image-to-video and frame control, reference inputs, price per second, and whether you can actually reach the model. The tables below line up current models on each, mostly from the makers’ own documentation, with press and reseller figures labelled where a maker hasn’t published one.
Try Veo 3.1 Lite, Fast and Quality in one place
Text-to-video, image-to-video and first-and-last-frame, 4 to 8 seconds, up to 4K. Packs start at $9.99 with no subscription.
Sign in with Google to create an account; buy credits only when you need them.
Key takeaways
- No model leads on every criterion. Decide which two or three matter for your project first.
- The longest single clip now ranges from 8 seconds (Veo 3.1) to 30 seconds (Kling 4.0, Seedance 2.5, Wan3.0).
- Runway’s API has no audio setting and Luma Ray3.2 generates no audio; Veo 3.1 generates sound natively.
- Sora 2 is discontinued: OpenAI removed its API on September 24, 2026.
The seven criteria that matter
Longest single clip
How many seconds one generation returns. Longer single takes keep a camera move and a scene continuous; shorter clips are fine for edited content made of cuts. Check whether “extend” is offered on the route you’ll actually use, because it often differs between a maker’s app and its API.
Resolution, native or upscaled
Some models generate 1080p or 4K directly; others generate lower and upsample. The label on the download can be the same, so look for the word “upsampled” or “upscaled” in the maker’s docs.
Native audio
Whether the model generates dialogue, sound effects and ambience with the picture. If it doesn’t, budget time for sound design. If it does, check that audio can be switched off, because silent clips are usually cheaper.
Image-to-video and frame control
Can you give a starting image, and can you also fix the last frame or several keyframes? Frame control is the most reliable way to keep a character or product consistent across shots.
References
How many reference images, videos or audio clips the model accepts for subjects and style. This varies the most between models, from none to dozens, and it changes often.
Price per second
Normalise everything to dollars per second at the resolution you need, with or without audio. Compare the maker’s own API price against any reseller or subscription you’re considering, and remember that subscriptions only work out cheap if you use the credits.
Availability and API status
Is the model generally available, in preview, app-only, or discontinued? Models are being retired quickly: Sora 2’s API closed, Veo left the Gemini API, and Google Flow is dropping Veo 3.1 Fast and Quality. A model you can’t reach next quarter is a poor foundation.
Current models at a glance
Most cells come from the maker’s own pages, listed under Sources; a few launch dates come from the press, and Seedance’s price is a reseller’s, and those are labelled in the cell. “Not published” means we could not find the figure on an official page, not that the feature is missing. Prices are approximate API rates per second of output, cheapest to dearest setting. Some makers bill in credits, tokens or Google’s “count” unit, which we convert; Omni Flash figures are Google’s own approximations of a per-token price. Sora 2 is included for reference only.
| Model | Longest single clip | Top resolution | Native audio | Image-to-video | First and last frame |
|---|---|---|---|---|---|
| Veo 3.1 (Quality) | 8 s (extend to 37 s on Google Cloud) | 4K | Yes (can be switched off) | Yes | Yes |
| Veo 3.1 Fast | 8 s | 1080p on Google’s model page; 4K is on Google’s price list and in our generator | Yes | Yes | Yes |
| Veo 3.1 Lite | 8 s | 1080p | Yes | Yes (Google’s Cloud page contradicts itself: one row says image input is not supported) | Yes |
| Gemini Omni Flash 1.2 | 15 s (extend to 30 s) | 720p native; 1080p and 4K by upsampling | Yes (no audio inputs) | Yes | Yes (clips up to 10 s) |
| Kling 4.0 | 30 s in one pass (Flash: 20 s) | 1080p; Kling’s own page both claims “up to 4K” and says 4K is “coming soon” | Yes | Yes | Yes (up to 10 keyframes) |
| Seedance 2.5 | 30 s | 1080p | Yes | Yes | Yes |
| Wan3.0 | 30 s | 1080p | Yes | Yes | Yes |
| Grok Imagine Video 1.5 | 15 s | 1080p | Yes | Yes | Yes (plus up to 4 keyframes) |
| MiniMax H3 | 15 s | 2K | Yes | Yes | Yes |
| Runway Gen-4.5 | 10 s | 1280×720-class output sizes in the API | No audio parameter in the official API | Yes | First frame only in the API |
| Luma Ray3.2 | 10 s (Modify Video: 20 s) | 1080p, with native 16-bit HDR | No | Yes | Yes |
| Pika H3 Flex | 15 s | 768p native (up to 4K by upscaling) | Yes | Yes | Yes |
| Sora 2 / Sora 2 Pro | 12 s per the API reference (the guide also lists 16 and 20 s) | 1080p (Pro) per OpenAI’s model page; the API’s size list had no 1080p option | Yes | Yes (first frame) | First frame only |
| Model | Reference inputs | Extend | Maker’s API price, about per second | Status (November 2026) | Available on imaginary.app |
|---|---|---|---|---|---|
| Veo 3.1 (Quality) | Up to 3 asset images or 1 style image (Google Cloud API) | Google Cloud only (+7 s per extension) | $0.20/s to $0.60/s | Generally available on Google Cloud (supported until “May 4, 2027 or later”); Gemini API preview shut down October 22, 2026; leaves Flow December 5, 2026 | Yes |
| Veo 3.1 Fast | Up to 3 asset images (Google Cloud API) | Google Cloud only | $0.08/s to $0.30/s | Generally available on Google Cloud (supported until “May 4, 2027 or later”); Gemini API preview shut down October 22, 2026; leaves Flow December 5, 2026 | Yes |
| Veo 3.1 Lite | No | Google Cloud only | $0.03/s to $0.08/s | Preview on Google Cloud (no published support date); Gemini API preview shut down October 22, 2026; stays in Flow | Yes |
| Gemini Omni Flash 1.2 | Up to 6 images plus short reference videos | Yes, to 30 s total | ≈ $0.023/s (360p) to ≈ $0.209/s (4K upsampled); Google prices Omni per token | Live on the Gemini API and in the Gemini app; Flow’s default model | Not offered today |
| Kling 4.0 | Up to 10 images, 5 videos and 7 subject elements | Described as coming soon | Not published for 4.0 (official Kling 3.0 API: $0.084/s to $0.42/s) | Live in Kling’s app; official 4.0 API not yet open | Not offered today |
| Seedance 2.5 | Up to 30 images, 10 videos and 10 audio clips | Yes | Token-priced; fal.ai’s resale works out to about $0.47/s at 720p | Live on BytePlus (ByteDance’s Seed blog still calls the API “coming soon”) | Not offered today |
| Wan3.0 | Up to 10 images, 5 videos and 5 audio clips | Yes | $0.05/s (480p) to $0.28/s (Prime, 1080p) | Live (docs still say preview) | Not offered today |
| Grok Imagine Video 1.5 | Yes (xAI’s docs and news post give different maximums) | Classic model only | $0.02/s (1.5 Lite) to $0.08/s (1.5) | Live | Not offered today |
| MiniMax H3 | Up to 9 images, 3 videos and 3 audio clips | No (API) | $0.025/s (H3 Lite) to $0.13/s (H3, 2K) | Live | Not offered today |
| Runway Gen-4.5 | Not published | No (edits go through Aleph 2.0) | $0.05/s (Gen-4 Turbo) to $0.12/s (Gen-4.5) | Live (all paid plans and the API) | Not offered today |
| Luma Ray3.2 | Not published | Yes (SDR only) | $0.06/s (720p) to $0.24/s (1080p) at Luma’s default list rate; HDR costs double | Live | Not offered today |
| Pika H3 Flex | Up to 9 images, 3 videos and 3 audio clips | Not published | $0.02/s (480p) to $0.16/s (4K upscaled), before launch promos | Live | Not offered today |
| Sora 2 / Sora 2 Pro | Up to 2 “characters” | Was available | $0.10/s to $0.70/s (historical) | Discontinued: API removed September 24, 2026 | Not offered today |
The prices are the makers’ list prices, not ours. What Veo 3.1 costs on imaginary.app is on each comparison page and on the Veo 3.1 cost calculator; it is higher than Google Cloud’s list price, in exchange for no Cloud setup and one-time packs. For details on each Veo model, see the models hub.
Our comparisons
We only publish a comparison when we can source both sides and say honestly when the other model is the better choice. All were updated in November 2026.
- Veo 3.1 Lite vs Fast vs Quality
Same model family, three price points. Lite for drafts and volume, Fast for most finished clips, Quality for the most detail. Includes our full credit table and real gallery clips.
- Veo 3.1 vs Gemini Omni Flash
Google’s newer model makes longer clips and edits by conversation; Veo 3.1 generates 1080p and 4K natively, and Veo 3.1 and Fast are generally available on Google Cloud.
- Veo 3.1 vs Kling 4.0
Kling wins on clip length and references; Veo has an official API today and native 4K on its Quality model. Kling’s own 4.0 API isn’t open yet.
- Veo 3.1 vs Runway Gen-4.5
Runway brings 10-second clips and video-to-video editing. Runway’s API has no audio setting; Veo generates sound natively and adds 1080p, 4K and first-and-last-frame control.
- Veo 3.1 vs Luma Ray3.2
Luma suits silent, grade-ready HDR footage in more aspect ratios; Veo suits clips that need sound, or 4K.
Looking for a replacement rather than a head-to-head, for example after Sora 2 closed or Google Flow dropped the model you use? Start with our guide to alternatives by need, or read about the Veo 3.1 changes at Google.
How we build these comparisons
- Specs and prices come from the makers’ own pages where they exist; press reports and reseller prices fill gaps only where a maker hasn’t published a figure, and we label them.
- Where a maker’s pages contradict each other, we say so instead of picking the flattering number.
- We don’t publish side-by-side “tests” of models we haven’t run. Our clips are Veo 3.1 only, from our public gallery.
- Every comparison has a section on when to choose the other model.
- We name a better option only against a stated criterion, never overall.
How we checked: every row was read from the linked pages (official documentation first) on November 8, 2026. Models change quickly; check the source before relying on a number.
Sources
Prices, model specs and dates on this page were checked on November 8, 2026. Third-party prices change; check the source before you rely on a number.
- Google Cloud — Veo 3.1 model documentation
- Google Cloud — generative AI pricing
- Google Cloud — extend Veo videos
- Google Cloud — generate Veo videos from reference images
- Google blog — Veo 3.1 Lite announcement
- Google AI for Developers — Gemini Omni
- Google AI for Developers — Gemini API pricing (Omni Flash)
- Google Flow Help — credits and models
- Kling AI — Kling 4.0 feature page
- Kling AI — developer API pricing
- Bernama — Kling 4.0 launch (press)
- The Decoder — Kling 4.0 and 4.0 Flash (press)
- ByteDance Seed — introducing Seedance 2.5
- BytePlus ModelArk — Seedance 2.5
- BytePlus ModelArk — model pricing
- fal.ai — Seedance 2.5 (reseller price)
- Alibaba Cloud Model Studio — Wan3.0 API reference
- Alibaba Cloud Model Studio — model pricing
- xAI — Grok Imagine Video 1.5
- xAI docs — video generation
- xAI docs — pricing
- MiniMax — MiniMax H3
- MiniMax API — video generation reference
- MiniMax API — pricing
- Runway — introducing Gen-4.5
- Runway API — models
- Runway — official Node SDK (parameters)
- Runway API — pricing
- Luma AI — introducing Ray3.2
- Luma API — video generation guide
- Luma API — pricing
- Luma AI — Ray product page
- Pika — Pika H3 Flex
- OpenAI — API deprecations (Sora 2 / Videos API)
- OpenAI — Sora 2 Pro model page (historical prices)
- OpenAI — Videos API reference (create)
- OpenAI — video generation guide
- Google AI for Developers — Gemini API deprecations
Try Veo 3.1 Lite, Fast and Quality in one place
Text-to-video, image-to-video and first-and-last-frame, 4 to 8 seconds, up to 4K. Packs start at $9.99 with no subscription.
Sign in with Google to create an account; buy credits only when you need them.