Best Video Generation APIs in 2026: Models, Pricing, and Limits Compared

Key Takeaways

• Cherry 2 Pro on the Mage API is our top pick, with clips up to 30 seconds at up to 1080p, characters and voices by @handle, and no monthly fee, with 12 more video models behind the same key.

• Pricing units differ, so price the same clip on each API. Mage bills Gems per clip (618 Gems, about $0.62, for Cherry 2 Pro at its 4-second, 480p default), Runway and Kling bill per second, and BytePlus bills Seedance per token.

• Clip length now reaches 30 seconds on Cherry 2 Pro, Kling 4.0, and Seedance 2.5. Runway's Gen-4.5 stops at 10 seconds and Veo 3.1 at 8.

• OpenAI removed the Sora API on September 24, 2026, and Google shut down the Veo 3.1 preview models on the Gemini API on October 22, 2026. Veo 3.1 remains on Google Cloud's Gemini Enterprise Agent Platform.

What Is the Best Video Generation API?

Choosing a video generation application programming interface (API) comes down to clip length at the resolution you ship, a character that stays consistent across calls, and billing you can predict before a job runs. On Mage, that's Cherry 2 Pro, our flagship video model and a Mage exclusive, with 12 other video models a single parameter away.

We compared 6 models on billing, job handling, length and resolution, audio, character consistency, and content rules. The demos all look good, so the differences show up in the docs.

What Sets a Good Video Generation API Apart

Here's what we checked in each API's docs.

Waiting for a render

Every video API here is asynchronous. You submit a job, get an ID, and then poll or wait for a webhook. Idempotency keys stop a retried request from paying twice.

Length and resolution

The longest clip you can get in one call ranges from 8 to 30 seconds here, and top resolutions run from 720p to 4K. A short cap means stitching, and stitching is where continuity breaks.

Sound on the clip

Native audio saves a second pipeline for dialogue and effects. Some APIs price audio separately, so check whether the rate you're quoting includes it.

Keeping a character across clips

If you're making a series, you need the same person in every clip, so look for saved characters or a generous reference-image budget per request.

Content rules and failed runs

Every API here moderates, and they treat blocked or failed jobs differently. Some refund failures, some charge for a moderated request, and some only bill successful videos.

The Top 6 Video Generation APIs in 2026

1. Cherry 2 Pro - by Mage

Cherry 2 Pro is the default model in the Cherry line on the Mage video generation API, a REST API at api.mage.space that runs 13 video models with one key, one request lifecycle, and the same field names. There's no monthly fee. Each clip is paid in Gems, and 1,000 Gems cost $1 on the standard pack.

What it does well:

• Clips of 4 to 30 seconds at 480p, 720p, or 1080p

• Up to 29 images, characters, and references per request, plus up to 3 audio sources

• Saved characters and their voices by @handle, exactly as in the Mage app

• Video editing on a source clip of 4 to 30 seconds, keeping its length and aspect ratio

• Idempotency keys, so a retried submit never charges twice

• Refunds on failed generations

Every response reports the exact charge in billing.gems_charged, accounts can run 20 generations at once, and results stay available for 30 days. Official TypeScript and Python SDKs wrap the submit-and-poll loop, and the same models are reachable from Claude and ChatGPT through the Mage Model Context Protocol (MCP) server. The API docs list every model's fields and price.

The catch: there are no webhooks, so you poll status_url until the job completes.

Best for: apps that need long, character-consistent clips and access to a range of models with different strengths, all on one key.

2. Gen-4.5 - by Runway

Runway's developer platform documents its models, pricing, and moderation in detail. Gen-4.5 is its newest text-to-video and image-to-video model, and the same API also sells Aleph 2.0 for video editing and Act-Two for character performance.

What it does well:

• Credits at $0.01 each, with Gen-4.5 at 12 credits a second

• Third-party models on the same key, including Kling, Veo 3.1, and Seedance

• Act-Two for driving a character with a performance video

• Clear task statuses, from PENDING to SUCCEEDED or FAILED

Gen-4.5 runs 2 to 10 seconds, and its text-to-video output comes in 1280 by 720 or 720 by 1280. Runway says a task's status won't update more than once every 5 seconds, and output URLs expire within 24 to 48 hours.

The catch: moderated generations cost the same as successful ones, and the API reference documents no webhooks.

Best for: teams that want several labs' models behind one well-documented account.

3. Kling 4.0 - by Kuaishou

Kling 4.0 is the newest model on Kling's official API, alongside 3.0, 3.0 Turbo, and older versions. A single call takes up to 15 inputs across images, videos, characters, and voices.

What it does well:

• Clips of 3 to 30 seconds at 720p or 1080p

• Up to 10 images, 5 reference videos, 7 Elements (characters), plus voices, with 15 inputs in total

• Webhooks through a callback_url

• Native audio from dialogue, music, and sound described in the prompt

Kling 4.0 costs 1 unit a second at 720p and 1.5 at 1080p, which works out to about $0.15 and $0.23 a second on the $500 package. Trial packages start at $9 for 64 units.

The catch: units expire with no rollover, after 30 days on trial packages (which allow 5 concurrent jobs) and after 180 days on the larger ones (which allow 20).

Best for: production apps that need long clips with several characters and voices in one request.

4. Veo 3.1 - by Google

Veo 3.1 is generally available on Google Cloud's Gemini Enterprise Agent Platform, formerly Vertex AI. Google shut down its preview models on the Gemini API on October 22, 2026, and pointed Gemini API developers to Gemini Omni 1.1 Flash instead.

What it does well:

• 720p, 1080p, and 4K output

• Native sound generation

• Up to 4 videos per prompt

• A Fast variant for cheaper drafts, plus Lite in preview

You'll pay $0.40 a second for Veo 3.1 with audio at 720p or 1080p, and $0.20 without. Veo 3.1 Fast starts at $0.10 a second with audio, and Lite at $0.05.

The catch: you can't go past 8 seconds in one call, and reference-image video is fixed at 8.

Best for: short, high-resolution clips with sound for teams already on Google Cloud.

5. Seedance 2.5 - by ByteDance

Seedance 2.5 runs on BytePlus ModelArk and generates video with audio from up to 50 reference assets (30 images, 10 videos, and 10 audio clips).

What it does well:

• Clips of 4 to 30 seconds

• A 480p draft mode for cheap previews

• Webhooks through a callback_url

• Charges only for successfully generated videos

Pricing is per token, worked out from the clip's duration, size, and frame rate. BytePlus's own example puts a 5-second, 16:9 clip at about $0.51 at 480p, $1.16 at 720p, and $2.84 at 1080p.

The catch: Seedance 2.5 doesn't accept uploaded references with real human faces, so any people in the clip have to come from BytePlus's preset digital characters, your own earlier outputs, or a real person who has verified and consented through BytePlus. BytePlus also recommends a balance above $30 before you enable the model.

Best for: apps that build scenes from many references and want draft previews before the final render.

6. MiniMax-H3 - by MiniMax

MiniMax-H3 is MiniMax's current video line, and MiniMax now labels its older Hailuo models legacy.

What it does well:

• $0.08 a second at 768p and $0.13 at 2K

• H3-Lite from $0.025 a second at 480p

• Up to 9 reference images, 3 reference videos, and 3 audio clips, with 12 files in total

• A simple create, poll, and download flow

MiniMax recommends polling a task every 10 seconds, or you can pass a callback_url and get a POST on every status change. Mage also runs MiniMax-H3 through its own API.

The catch: clips stop at 15 seconds, and the cheaper H3-Lite tops out at 768p.

Best for: high-volume apps where the price per second decides the model.

Calling Cherry 2 Pro Through the Mage API

The whole flow, from key to finished MP4, takes 5 steps.

Step 1: Sign in to Mage, open API, then API Keys, and create a key. It starts with mage_sk_ and is shown once, and an account can hold 10 active keys. Keys start working after one purchase, and a single Gem pack counts. You'll find your keys on the Mage API page.

Step 2: POST a JSON body to https://api.mage.space/v1/cherry/generate with an Authorization: Bearer header, model_id set to cherry-2-pro, and an Idempotency-Key header.

Step 3: Write the prompt the way you'd brief a camera operator, naming the move, the subject, and the light.

Example prompt:

A slow push-in on a lighthouse at dusk, waves breaking below

Step 4: Poll the status_url from the response until the status is completed, failed, or cancelled.

Step 5: Download result.url within 30 days, and check billing.gems_charged to log the cost.

Video Generation API Questions, Answered

What is the best video generation API?

Cherry 2 Pro on the Mage API, for 30-second clips at up to 1080p, saved characters and voices, and pay-per-clip pricing with no monthly fee. Veo 3.1 renders 4K, and Kling 4.0 takes up to 15 inputs in one call.

Can I still use Veo through the Gemini API?

Not the preview models. Google shut them down on the Gemini API on October 22, 2026, and replaced them there with Gemini Omni 1.1 Flash. Veo 3.1 is still generally available on Google Cloud's Gemini Enterprise Agent Platform, formerly Vertex AI.

Which video API makes the longest clips?

Cherry 2 Pro, Kling 4.0, and Seedance 2.5 all reach 30 seconds in a single call. MiniMax-H3 stops at 15, Gen-4.5 at 10, and Veo 3.1 at 8.

Can a video generation API keep the same character across clips?

Yes, if it takes saved characters or enough references. Cherry 2 Pro on the Mage API calls saved characters and their voices by @handle, with up to 29 images, characters, and references per request. Kling 4.0 takes up to 7 Elements, and Seedance 2.5 takes up to 50 reference assets but won't accept real human faces.

Which video generation APIs make sound?

Cherry 2 Pro, Veo 3.1, Kling 4.0, Seedance 2.5, and MiniMax-H3 all generate audio with the clip. Cherry 2 Pro also takes up to 3 audio references and your characters' saved voices. Runway's Gen-4.5 doesn't document audio output.

Can I edit an existing video through an API?

Yes. Cherry 2 Pro edits a source clip of 4 to 30 seconds and keeps its length and aspect ratio, and Magejutsu on the Mage API recasts footage while keeping its motion, camera, and sound. Runway sells Aleph 2.0 for video editing on its own API.

Do I need a subscription to use the Mage API?

No. Any Mage account can create an API key and pay per generation in Gems, with no monthly fee. The key starts working after one purchase, and a single Gem pack counts. 1,000 Gems cost $1 on the standard pack.

Generate Your First Clip

Create a key, send the lighthouse prompt to Cherry 2 Pro at its 4-second, 480p default, and check billing.gems_charged on the response. Then switch the duration to 30 and see what a full scene costs before you build around it.

Building an agent instead of an app? See Mage's guide to Claude connectors for video. Have fun creating!