Kling 3.0 launched Feb 5: what Sume's kling-3 row accepts
Kling 3.0 launched 2026-02-05 with 15 s clips and storyboards. Sume's kling-3 takes 4-15 s at 720p or 1080p, start/end frames, optional audio, no references.

Kling AI's 3.0 launch release is dated February 5, 2026 and lists clips up to 15 seconds, native speech in several languages and storyboard-style multi-shot control in Video 3.0 Omni. Sume's kling-3 row (named Kling Video v3 Pro in the catalog) takes text-to-video and start/end-frame image-to-video for 4 to 15 seconds at 720p or 1080p, and rejects reference URLs.
The release is read from Kling's press release as syndicated on Yahoo Finance on 2026-10-07; the Sume side is from the Video Router docs.
Release claims next to the Sume row
The release says the models were initially open to Ultra subscribers, with broader access to follow.
| Topic | Kling 3.0 release | Sume kling-3 |
|---|---|---|
| Length | Up to 15 seconds | 4 to 15 seconds |
| Audio | Native speech in English, Chinese, Japanese, Korean, Spanish | Optional through generate_audio; one price with audio off, one with on |
| Multi-shot storyboard | Video 3.0 Omni feature: per-shot duration, shot size, perspective, camera | No storyboard field documented |
| Inputs | Not itemized in the release | Text, start frame, end frame; no reference_*_urls |
| Resolution | Not stated | 720p or 1080p |
| Aspect ratios | Not stated | 16:9, 9:16, 1:1 |
What the "no references" rule means
Sume's kling-3 is the row for cinematic motion from a prompt or between two stills. If you send reference_image_urls, reference_video_urls or reference_audio_urls to it, the request is rejected, so a character-consistency job with several photos belongs on a row that lists references, such as minimax-h3 (up to 9 images, 3 videos and 3 audios, 12 in total).
Price at a glance
The 720p and 1080p rates are the same in the catalog, so choose 1080p when the delivery needs it.
| Setting | Per second | 10 seconds | 15 seconds |
|---|---|---|---|
| Audio off | $0.14 | $1.40 | $2.10 |
| Audio on | $0.21 | $2.10 | $3.15 |
Two stills and a move
The start and end frame mode is kling-3's most distinctive use on Sume. You give a first frame and a last frame as frame_images, write a prompt for the motion between them, and pick 4 to 15 seconds. It is a good way to animate a before-and-after, a product turning to a new angle or a slow reveal, because the endpoints are fixed by your pictures.
Keep the two frames similar in framing and lighting. If they differ a lot, the model has to invent the bridge, and the result is less predictable. A 10-second clip at 720p with audio off is $1.40.
- Same aspect ratio for both frames and for the request.
- Subject in a similar place in both frames.
- A prompt that names the move, for example a slow push-in.
Where the release and the row part ways
The press release is about the Kling 3.0 family as a whole, and the family is bigger than one row. Features it names for the Omni variant, such as per-shot storyboards, are not in the Sume row's documented fields. When a client cites the launch, ask which feature they need, then check it against the row before you agree to it.
Sources
Related posts
More in Models
- Lip-sync and motion control cost per clip: H3 Max, Fabric, Kling
Sume prices H3 Max lip-sync at $0.0625 to $0.20 per audio second, Fabric at $0.1875 and Kling 3.0 Motion Control at $0.1575. A 10-second clip is $0.63 to $2.00.
- Music Router prompt limit: 5,000 characters, and what to put in them
Sume Music Router prompts run 1 to 5,000 characters, with no duration field or negative prompt; length goes in the text. A brief budget and a checker.
- MAI-Transcribe-2-Streaming 0.13 s to final: when the clock starts
The 0.13 second figure is measured from end of speech found by a VAD, and partials arrive in about 100 ms. Why neither number is the wait for a file transcript.
- MAI-Transcribe-2-Streaming 2.5% WER: what the test audio mix is
The 2.5% word error rate comes from a chunked-streaming index: 50% AA-AgentTalk, 25% VoxPopuli, 25% Earnings22. Here is how that maps to a recorded call.
Written by Sume