Kling O3 API
Kling O3 视频生成与编辑模型,支持 3~15 秒文生、图生、参考生成和视频编辑;文生与图生支持 720p、1080p、4K 及可选音效,参考与编辑按当前计费规范支持 720p、1080p 且关闭音效。
What is Kling O3
Kling O3 is a video generation model from Kuaishou's Kling 3.0 family; upstream sources call it Kling Video 3.0 Omni, released with the Kling 3.0 series on February 4–5, 2026. Upstream claims include native 4K (60fps, announced April 29, 2026), a multi-shot storyboard of up to 6 cuts/15 seconds per run, character appearance and voice consistency driven by up to 7 reference images or a reference video, natural-language video editing, a visual chain of thought (vCoT), and native audio in five languages (English, Chinese, Japanese, Korean, Spanish). All of these are upstream claims; actual results are defined by task output.
Roko exposes kling-o3 through the /v1/videos/tasks async task API with four routes: text-to-video, image-to-video, reference-to-video, and video-edit, all billed per output second. Text- and image-to-video support an audio toggle (off by default); reference-to-video and video-edit are restricted to 720p/1080p with audio forced off. Image-to-video accepts 1–2 input images; reference and edit routes accept 0–7 images plus 1 video.
Prices come from the Roko public catalog: text/image-to-video without audio costs $0.088/s at 720p, $0.1166/s at 1080p, and $0.4378/s at 4K; with generated audio $0.1166/s at 720p and $0.1463/s at 1080p (4K stays $0.4378); reference-to-video and video-edit cost $0.132/s at 720p and $0.1761/s at 1080p. Duration is 3–15s (default 5), resolution defaults to 720p, and aspect ratios are 16:9, 9:16, 1:1 (default 16:9). Costs are estimated at submit time and settled on actual usage at completion.
Model pricing
Pricing follows the current production configuration and the billing dimensions supported by each route.
| Route | Price | Billing rule |
|---|---|---|
| 文生 / 图生(无音频) | 720p $0.088/sec1080p $0.1166/sec4k $0.4378/sec | 按输出视频秒数 |
| 文生 / 图生(生成音频) | 720p $0.1166/sec1080p $0.1463/sec | 按输出视频秒数 |
| 参考 / 编辑 | 720p $0.132/sec1080p $0.1761/sec | 按输出视频秒数 |
Cost examples
Examples below are computed as output seconds × the unit price of the selected tier; costs are estimated at submit and settled on actual usage. Amounts are catalog unit-price products and exclude any platform promotions.
$0.088/s × 5s = $0.44
The default combination — 5s duration, 720p resolution, audio off — is the cheapest way to validate a prompt.
$0.1463/s × 5s = $0.7315
Enable audio when you need ambience or dialogue; the same spec without audio is $0.583 ($0.1166/s × 5s), so the toggle clearly moves the bill.
$0.1761/s × 10s = $1.761
The reference route is always audio-off; feed 1 reference video plus up to 7 images, billed at $0.132/s (720p) or $0.1761/s (1080p).
$0.4378/s × 5s = $2.189
4K costs $0.4378/s with or without audio — roughly 5× the 720p audio-off rate. Lock the composition at 720p before rendering the 4K master.
Model information
Only provider-documented routes, capabilities, and limits implemented by this gateway are listed.
kling-o3-text-to-video
- Limits
- —
- Duration
- 3–15s
kling-o3-image-to-video
- Limits
- Images 1–2
- Duration
- 3–15s
kling-o3-reference-to-video
- Limits
- Images 0–7 · Videos 1
- Duration
- 3–15s
kling-o3-video-edit
- Limits
- Images 0–7 · Videos 1
- Duration
- 3–15s
Key capabilities
4K tier available
Upstream claims Kling O3 natively outputs 4K at 60fps (announced April 29, 2026). On Roko, text- and image-to-video routes can deliver 4K at $0.4378 per second — the same price with or without audio. Best for footage that needs reframing or large-screen delivery; validate prompts at 720p first.
Audio toggle cuts the unit price
Text- and image-to-video routes support an audio toggle, off by default. At 720p, audio-off costs $0.088/s versus $0.1166/s with audio — a $0.143 difference on a 5-second clip; 1080p with audio is $0.1463/s. Keep it off when you don't need dialogue or ambience to get the lowest rate.
Reference-to-video and video-edit routes
The reference-to-video and video-edit routes accept 1 reference/edit video plus 0–7 images, billed at $0.132/s (720p) or $0.1761/s (1080p). Both routes are capped at 720p/1080p with audio forced off, ideal for extending, rewriting, or restyling existing footage.
Upstream multi-shot and consistency claims
Upstream sources describe a storyboard mode of up to 6 cuts/15 seconds per run, character appearance and voice consistency extracted from up to 7 reference images or a reference video, and improved on-screen text rendering. These are upstream claims, not Roko-measured results — verify against actual task output.
Main parameters
Parameters come from structured model configuration; unsupported or unconfigured fields are omitted.
modelstringSubmodel ID
Default:kling-o3-text-to-video / kling-o3-image-to-video / kling-o3-reference-to-video / kling-o3-video-editpromptstringKling O3 视频生成与编辑模型,支持 3~15 秒文生、图生、参考生成和视频编辑;文生与图生支持 720p、1080p、4K 及可选音效,参考与编辑按当前计费规范支持 720p、1080p 且关闭音效。
Default:RequireddurationintegerOutput video seconds
Default:3–15squalityenumCurrent billing
Default:720p / 1080p / 4kaspect_ratioenumCapability
Default:16:9 / 9:16 / 1:1generate_audiobooleanCapability
Default:falseimage_urlsarrayImage to video: 1–2 · Reference to video: 0–7 · Video editing: 0–7
Default:Varies by routevideo_urlsarrayReference to video: 1 · Video editing: 1
Default:Varies by routeCapabilities and usage guidance
Key features
- Text to video · kling-o3-text-to-video
- Image to video · kling-o3-image-to-video
- Reference to video · kling-o3-reference-to-video
- Video editing · kling-o3-video-edit
- 720p / 1080p / 4k output
- 3–15s Duration
- Generated audio output
- Audio-capable workflow
Limits
- Image to video: Images 1–2
- Reference to video: Images 0–7 · Videos 1
- Video editing: Images 0–7 · Videos 1
- Capability source: https://evolink.ai/zh/kling-o3
Limits and caveats
- Reference-to-video and video-edit are restricted to 720p/1080p with audio forced off; 4K or native audio is only available through text/image-to-video.
- The 4K tier applies to text/image-to-video only, at $0.4378/s with or without audio — about 5× the 720p audio-off rate of $0.088/s.
- A successful submission only means the task was accepted, not that generation will succeed; cost is estimated at submit and settled on actual usage, with the terminal task state as authoritative.
- Roko output links expire; once a task succeeds, transfer the file to your own storage promptly.
- Only three aspect ratios are supported — 16:9, 9:16, and 1:1 (default 16:9); other framing requires cropping or padding.
- Upstream claims — multi-shot storyboarding (up to 6 cuts/15s), character/voice consistency, and five-language native audio — come from upstream material and are not Roko-measured; verify against actual task output.
How to call it
- Prepare the prompt and parameters: set model to kling-o3, choose inputs for your route (text prompt; 1–2 images; or 1 video plus 0–7 images for reference/edit), and set duration 3–15s, 720p/1080p/4K resolution, and a 16:9, 9:16, or 1:1 aspect ratio; enable audio on text/image-to-video if needed.
- POST /v1/videos/tasks to submit the job and receive a task_id; success only means the task was accepted, and the cost in the response is an estimate.
- Poll GET /v1/tasks/{task_id} to track status until the job leaves queued/processing and reaches a terminal state.
- On succeeded, read the video URL from the task result; Roko output links expire, so transfer files to your own storage promptly.
- On failed, read the error, adjust the prompt or parameters, and resubmit with the same flow; failed tasks are settled per the platform refund rules, and the final query result is authoritative.
Typical use cases
Multi-shot shorts with a consistent character
Input: character reference images (up to 7 per upstream) plus storyboard-style prompts; method: reference-to-video route or storyboard prompting for multi-shot sequences; expected: the same character keeps a consistent look across cuts (upstream claim, needs hands-on validation); limits: reference routes are 720p/1080p and audio-off only.
Reference-video-driven character/voice carry-over
Input: 1 reference video that upstream claims can yield character appearance and voice signatures; method: generate new clips via reference-to-video; expected: the character likeness carries over (upstream claim); limits: audio is forced off and output capped at 720p/1080p, and the voice-carry-over capability is not Roko-verified.
Natural-language video editing
Input: 1 video to edit plus 0–7 images, with changes described in plain language; method: submit through the video-edit route; expected: restyling or targeted edits without frame-by-frame tooling; limits: output is 720p/1080p and audio-off, so add sound in post if needed.
Choosing against other video models on Roko
A rough split by billing style and workflow fit; final decisions should rely on each model's catalog and your own testing.
Kling O3 vs Kling 3.0 Turbo
Two siblings with different jobs: O3 goes broad (reference-to-video, video editing, a 4K tier, and upstream-claimed multi-shot/consistency features), while Turbo targets speed and low cost. Pick O3 for spec headroom, Turbo for high-volume short clips on a budget; prices for both live in the Roko catalog.
Kling O3 vs Hailuo 3
O3 offers an audio toggle (from $0.088/s at 720p audio-off) and a 4K tier ($0.4378/s), with reference and editing routes; Hailuo 3 leans on omni-modal input with up to 12 reference files. Choose Hailuo for multi-asset fusion, O3 for 4K masters or per-second, toggle-controlled audio costs.
Kling O3 vs Veo 3.1
The 4K pricing conventions differ: O3 4K is $0.4378/s flat (audio on or off), while Veo 3.1 4K is $0.308/s without audio or $0.3608/s with audio. Billing details differ too (Veo bills per result with a 4-second minimum per result), so don't compare unit prices alone — each catalog is authoritative for itself.
Frequently asked questions
What is the Kling O3 model ID on Roko?
Set model to kling-o3 when submitting; the gateway routes to kling-o3-text-to-video, kling-o3-image-to-video, kling-o3-reference-to-video, or kling-o3-video-edit depending on the route. Actual routing is reflected in the task response.
How much does one Kling O3 video cost?
Billing is per output second. Text/image-to-video: $0.088/s at 720p without audio or $0.1166/s with audio; $0.1166/s at 1080p without audio or $0.1463/s with audio; 4K is always $0.4378/s. Reference-to-video and video-edit: $0.132/s at 720p and $0.1761/s at 1080p (audio off). Example: 5s at 720p without audio is $0.44.
Is 4K output supported?
Yes. Text- and image-to-video routes support 720p, 1080p, and 4K (default 720p); 4K is $0.4378/s with or without audio. Upstream claims native 4K 60fps — an upstream claim. Reference-to-video and video-edit do not support 4K.
What are the limits of the reference and edit routes?
Both routes accept 1 video plus 0–7 images (image-to-video instead takes 1–2 images and no video), but output is capped at 720p/1080p with audio forced off, priced at $0.132/s (720p) or $0.1761/s (1080p).
Which durations and aspect ratios are supported?
Duration is 3–15s with a default of 5s; aspect ratios are 16:9, 9:16, and 1:1 with 16:9 as default. The Roko catalog is authoritative.
Why can't I get the video immediately after submitting?
Video generation is an async task: POST /v1/videos/tasks returns a task_id, and you poll GET /v1/tasks/{task_id} until it reaches a terminal state before reading the result. A successful submit only means the task was accepted.
Can I try Kling O3 for free?
Roko API calls are billed at the rates above with no free-tier promise. Top up after signing up to call the model; any limited-time promotion is subject to the current site announcements.
Can I use the output commercially?
Commercial compliance depends on the rights to your input material, the generated content itself, Kuaishou's and Roko's terms of service, and applicable law. Clear copyright, likeness, and content review before publishing; reference material involving another person's face or voice especially requires authorization.
Sources and verification scope
Upstream capability claims come from Kuaishou Kling's public materials; Roko-callable parameters, routes, and prices come from the current production catalog. Verified on 2026-10-09.
- Kuaishou: Kling AI 3.0 official launch Official launch: Video 3.0 Omni, multimodal generation and editing, up to 15 seconds, native multilingual audio, multi-shot storyboards, and reference-video consistency
- Kuaishou: Q2 2026 results and native 4K rollout Official company disclosure for native 4K output in the Kling AI 3.0 model series
- HowAIWorks: Kling 3.0 family Upstream: kling-o3 model name, 3-15s default 5s, up to 7 reference images, natural-language video editing, 5 audio languages
- Roko API video task docs Async submit, task query, and billing lifecycle on Roko
- Roko API model catalog Current routes, prices, and generation config
Models in this family
Explore other available variants from the same model family.
Other video models
Explore more models available through the unified asynchronous video task API.