Kling 4.0 launched on October 15, 2026, and its official API update log now records full availability on October 28. These are separate milestones. For motion-comic production, the useful question has moved from “is it out?” to which endpoint fits a shot, what references it accepts, and how the job is billed.
What does the official API support now?
| Feature | API documentation checked October 30, 2026 |
|---|---|
| Single generation | 3–30 seconds |
| Omni Reference | Up to 15 assets in total |
| Reference sublimits | Up to 10 images, 5 videos, 7 Elements and 3 voices; shared limits also apply |
| Keyframe control | Up to 10 input keyframes |
| Omni prompt length | 26,000 English characters or 11,000 non-English characters; mixed text has a weighted limit |
| Output | 720p or 1080p; 21:9 and adaptive aspect ratio supported; 10-bit 4K still coming soon |
| Longer extension | Multiple extensions up to two minutes are described as coming soon |
| Agent access | Integration with Kling MCP and CLI announced |
| Developer API | October 28 update confirms availability; account funding requirements still apply |
The caps are not additive. Images, videos, Elements and voices share the 15-item allowance. Video character Elements also share a five-item cap with reference videos, with no more than three video character Elements. Input video duration, including those Elements, totals 3–30 seconds. For mixed-language Omni prompts, the documented condition is E / 26000 + N / 11000 <= 1, where E and N count English and non-English characters. The old 8,000-token launch description is not the current API character limit.
Is the Kling 4.0 API available?
Yes. The official API update log replaces the launch announcement's “coming soon” status. The platform says access requires a Flagship Resource Pack or an account-manager balance top-up. Confirm eligibility, available region and quota in the account you will use; a consumer subscription or an older resource pack is not proof of API entitlement.
| Workflow | Official create-task path | Typical use |
|---|---|---|
| Text to video | /text-to-video/kling-4.0 | A shot described in text |
| Image to video | /image-to-video/kling-4.0 | Animating an approved frame |
| Omni video | /omni-video/kling-4.0 | Mixed references, keyframes and video editing |
The Omni endpoint reference uses a contents array and settings object. Give reference assets unique IDs and refer to them as @id in the prompt. refer_image supplies a reference; first_frame, middle_frame and last_frame define keyframes; feature_video supplies a motion or other feature reference; base_video supplies the clip to edit. These roles are not interchangeable.
Omni defaults to 1080p when resolution is omitted, with duration chosen from the prompt, then a base clip, then 10 seconds. Set resolution and duration explicitly when comparing costs. Follow the current authentication documentation for the regional host and API key. A third-party model ID, URL or credit rate belongs to that provider's contract; it cannot be pasted into the official request unchanged.
How input video changes the bill
The official unit price table distinguishes these cases:
| Input | Billable duration | 720p | 1080p |
|---|---|---|---|
| No video | Output seconds | 1 unit/s | 1.5 units/s |
| With video | Output + max(input, output), in seconds | 0.6 units/s | 0.9 units/s |
Here “input” includes reference videos and video character Elements. For a hypothetical 10-second output, no video costs 10 / 15 units at 720p / 1080p. An eight-second input costs 12 / 18 units; a 15-second input costs 15 / 22.5 units. The lower video-input rate does not necessarily mean a cheaper finished clip.
These are arithmetic examples, not executed jobs or dollar quotes. Unit purchase costs and balance billing depend on the account. Save the task ID, returned usage.input_duration and billing details for reconciliation; do not assume one unit always equals a fixed dollar amount or a consumer credit.
How to use the new controls for a story
Start with an approved character sheet and a scene reference. Map each keyframe to a specific story beat: entrance, reaction, handover, departure. Use only enough references to define identity, space, performance, and voice; a larger reference budget is not a reason to upload conflicting examples.
For a first evaluation, compare a short close-up, a two-character exchange, an action with hand-to-object contact, and a matching shot later in the scene. Keep the input assets and retry allowance fixed. Record rejected takes, time to approval, and total spend per approved second. This is an evaluation plan, not an AniKuku benchmark.
For a dialogue shot, use the character sheet as a reference and approved storyboard frames as keyframes. Add a video reference only when its movement is needed, describe exactly what it controls, and budget for the extra billable duration. AniKuku's project workflow helps organize the assets; this article does not announce a new Kling integration in AniKuku.
What happened to the Preview and Flash rumors?
Our September article documented unconfirmed interface screenshots. The October release is stronger evidence and replaces that status. The official announcement should not be read as confirmation of every rumored Flash mode, price, or image-model specification.
FAQ
Is Kling 4.0 out?
Yes. The model launched October 15. The official API log records full availability October 28, subject to the platform's account requirements. 4K and longer extensions remain separate availability questions.
Does 4K mean Kling 4.0?
No. 4K is a resolution. Earlier Kling 3.0 models also offered 4K configurations. Verify the model ID as well as the output size.
Should I switch an entire project immediately?
Test the shots that currently fail first. Keep approved assets and shot history so a version change does not force a whole-episode restart.
Continue with our AI video model comparison and the script-to-storyboard workflow.
Sources: Kuaishou's model launch announcement, official API updates, Omni API reference and unit pricing. API documentation and pricing checked October 30, 2026; the model launch date remains October 15.