When did AI video clips reach 30 seconds?
Per generation, the notes read go from 10 seconds with Hailuo 02 in June 2025 to 15 with Wan 2.6 in December, 20 with Sora in March 2026, and 30 with Wan 3.0 on 6 August 2026 and Seedance 2.5, announced on Higgsfield and Luma the same day. Kling's Motion Control reached 30 seconds of action in December 2025. As of 2026-09-25.
| Date | Line | Version | What changed |
|---|---|---|---|
| 7 April 2025 | Amazon Nova Reel | Nova Reel 1.1 | Multi-shot videos up to two minutes, in an automated mode from one prompt or a manual mode with a prompt per shot |
| 18 June 2025 | MiniMax Hailuo | Hailuo 02 | Hailuo 02 generates 1080p and clips of up to 10 seconds |
| 23 September 2025 | Wan | Wan 2.5 | Wan 2.5 preview generates synchronised audio and 10 second clips |
| 3 December 2025 | Wan | Wan 2.6 | Wan 2.6 text-to-video plans several shots inside one clip of up to 15 seconds |
| 17 December 2025 | Kling AI | Kling 2.6 | Kling 2.6 Motion Control copies an action and expression, up to 30 seconds in one take |
| 12 March 2026 | Sora | Sora 2 | Character references, clips up to 20 seconds, 1080p on Sora 2 Pro, extensions and batch jobs |
| 6 August 2026 | Seedance | Seedance 2.5 | Seedance 2.5 arrives on Higgsfield with clips up to 30 seconds |
| 6 August 2026 | Seedance | Seedance 2.5 | Seedance 2.5 arrives on Luma with 30-second single-pass generation |
| 6 August 2026 | Wan | Wan 3.0 | Wan 3.0 folds reference, editing, replication and driving into one model, with clips up to 30 seconds |
| 11 August 2026 | Seedance | Seedance 2.5 | Seedance 2.5 reaches Runway through its MCP, up to 30 seconds |
| 20 August 2026 | Wan | Wan 3.0 | Wan 3.0 Prime, a faster version with the same capabilities |
| 24 August 2026 | Wan | Wan 3.0 | Wan 3.0 and Wan 3.0 Prime arrive on Higgsfield, up to 30 seconds |
| Undated version row | Seedance | Seedance 2.5 | Seedance 2.5: 4 to 30 seconds, 1080p in 10-bit |
Inclusion rule. Entries whose note bears directly on the question. Order. By date, oldest first; undated rows last.
Two different numbers hide behind clip length in release notes. One is how long a single generation can be. The other is how long a clip can grow when it is extended, one generation after another. Runway's Gen-3 Alpha reached 40 seconds in total through extensions as early as September 2024, but each piece was still a short generation. This page follows the first number.
The steps are not evenly spaced. Ten seconds held through most of 2025; fifteen arrived with Wan 2.6's multi-shot planning at the end of the year; Sora's API went to twenty in March 2026. Thirty arrived twice in one day in August 2026, once in Alibaba Cloud's list for Wan 3.0 and once in platform posts for Seedance 2.5, whose own model list gives 4 to 30 seconds.
Kling's December 2025 Motion Control note is a different kind of thirty seconds: an action copied from a reference video in one take, not a clip generated from a prompt.
Amazon's Nova Reel 1.1 is a third kind. From April 2025 one request can return up to two minutes, but as 6-second shots joined together, so each piece is still a 6-second generation.
1In the makers' words
Amazon Nova Reel, 7 April 2025
This update lets you generate multi-shot videos up to 2-minutes in length with consistent style across shots.
AWS News Blog, Amazon Nova Reel 1.1
- April 7Multi-shot videos up to two minutes, in an automated mode from one prompt or a manual mode with a prompt per shot2025
MiniMax Hailuo, 18 June 2025
Released MiniMax Hailuo 02, a next-generation video generation model with support for 1080p resolution and video generation of up to 10 seconds.
MiniMax, model release notes
- MiniMax-Hailuo-02Hailuo 02 generates 1080p and clips of up to 10 seconds
Wan, 23 September 2025
newly upgraded model architecture supports synchronized audio generation with visuals, enables 10-second long video generation
Alibaba Cloud Model Studio, newly released models
- wan2.5-t2v-previewWan 2.5 preview generates synchronised audio and 10 second clips
Wan, 3 December 2025
Intelligent shot scheduling supports multi-shot storytelling, generating multi-shot narrative videos with consistent subjects, scenes, and atmosphere, with a maximum duration of 15 seconds.
Alibaba Cloud Model Studio, newly released models
- wan2.6-t2vWan 2.6 text-to-video plans several shots inside one clip of up to 15 seconds
Kling AI, 17 December 2025
30 seconds of high-difficulty action in one seamless take, with total command over movement, expressions, lip-sync, and gestures.
Kling AI, release notes
- 2025-12-17Kling 2.6 Motion Control copies an action and expression, up to 30 seconds in one take
Sora, 12 March 2026
Expanded the Sora API with reusable character references, longer generations up to 20 seconds, 1080p output for sora-2-pro , video extensions, and Batch API support for POST /v1/videos .
OpenAI, API changelog
- March 12Character references, clips up to 20 seconds, 1080p on Sora 2 Pro, extensions and batch jobs2026
Seedance, 6 August 2026
Seedance 2.5 from ByteDance is available on Higgsfield, the latest version in the Seedance video family.
Higgsfield, changelog
- Seedance 2.5Seedance 2.5 arrives on Higgsfield with clips up to 30 seconds
Seedance, 6 August 2026
With Seedance 2.5 on Luma you can get native 30-second generation in a single pass
Luma, Seedance 2.5 on Luma post
- Seedance 2.5: Now Live in LumaSeedance 2.5 arrives on Luma with 30-second single-pass generation
Wan, 6 August 2026
It supports four-modal all-reference input, up to 30 seconds of video generation, and can parse files, web pages, and complex images.
Alibaba Cloud Model Studio, newly released models
- wan3.0-videoWan 3.0 folds reference, editing, replication and driving into one model, with clips up to 30 seconds
Seedance, 11 August 2026
Generate up to 30 seconds of video with native audio
Runway, changelog
- Seedance 2.5Seedance 2.5 reaches Runway through its MCP, up to 30 seconds
Wan, 20 August 2026
Wan3.0-Video-Prime is the high-speed version of the Wan3.0 video generation model, with capabilities aligned with the Wan3.0-Video standard version.
Alibaba Cloud Model Studio, newly released models
- wan3.0-video-primeWan 3.0 Prime, a faster version with the same capabilities
Wan, 24 August 2026
Up to 30 seconds per generation, at 480p, 720p, or 1080p.
Higgsfield, changelog
- Wan 3.0 and Wan 3.0 PrimeWan 3.0 and Wan 3.0 Prime arrive on Higgsfield, up to 30 seconds
Seedance, Undated version row
Duration: 4–30 s
BytePlus ModelArk, model list
- dreamina-seedance-2-5-260628Seedance 2.5: 4 to 30 seconds, 1080p in 10-bit
Clip length and extension · Which AI video models let you set how much moves? · Which AI video models generate sound with the picture?
2Notes read
- AWS News Blog, Amazon Nova Reel 1.1, read 2026-09-25
- MiniMax, model release notes, read 2026-09-25
- Alibaba Cloud Model Studio, newly released models, read 2026-09-25
- Kling AI, release notes, read 2026-09-25
- OpenAI, API changelog, read 2026-09-25
- Higgsfield, changelog, read 2026-09-25
- Luma, Seedance 2.5 on Luma post, read 2026-09-25
- Runway, changelog, read 2026-09-25
- BytePlus ModelArk, model list, read 2026-09-25