Kling Video O1
Kling Video O1 (Omni One) is Kuaishou's industry-first unified multimodal video model that merges video generation and editing into a single engine. Integrating text-to-video, image-to-video, element referencing, localized inpainting, and video restyling, it supports referencing up to 7 subjects simultaneously to lock character and prop consistency. Creators can perform conversational video editing via natural language prompts—ideal for advertising, VFX, and end-to-end film production.
Related Models
Kling v3.0 Image-to-Video
Kling v3.0 Image-to-Video is Kuaishou's next-generation multimodal AI video model. Built on the unified Omni architecture, it takes static images or subject references to generate up to 15-second cinematic videos in up to 4K resolution. It features native audio-visual synchronization, multilingual lip-sync, and enhanced subject consistency to prevent visual drift, along with intelligent multi-shot control—ideal for commercial ads, film VFX, and narrative short videos.
Kling v3.0 Text-to-Video
Kling v3.0 Text-to-Video is Kuaishou's next-generation AI video model. Powered by the native Omni architecture, it accurately parses complex prompt text to generate up to 4K cinematic-grade videos up to 15 seconds long. It natively supports integrated audio-video generation (ambient audio, music, and multilingual lip-sync) alongside exceptional visual realism, multi-shot coherence, and complex physical simulation—ideal for commercial advertising, film VFX, and content creation.
Kling 3.0 Omni
Kling 3.0 Omni delivers high-quality text-to-video generation with smooth motion, cinematic visuals, accurate prompt adherence, and native audio for ready-to-share clips. Ready-to-use REST inference API, best performance, no cold starts, affordable pricing.
Read Me
Kling Video O1 Video API
Kling Video O1 is Kuaishou's general-purpose video generation model in the Kling (KeLing) series, unifying long-video generation, multimodal references (image/video/character), automatic multi-shot assembly, and optional audio generation into a single model interface.
On iCreat, it is called via the kwaivgi/kling-video-o1 capability code using an asynchronous task queue. It supports two resolution tiers — Standard (720p) and Pro (1080p) — along with advanced capabilities such as multi-shot assembly, reference-video style transfer, and first-frame/end-frame specification.
Model Positioning
Kling Video O1 is positioned as the general-purpose video generation interface of the Kling series. Unlike Kling V3 T2V, which only accepts pure text input, and Kling V3 I2V, which only accepts image plus text, Video O1 unifies text, image, video, and character assets into a single request structure, supporting advanced capabilities such as automatic multi-shot assembly, reference-video style transfer, first/end-frame specification, audio generation, negative prompts, and watermark control — suited for production scenarios that require complete creative control.
For teams that need to generate video from mixed asset types or require multi-shot storytelling, Video O1 offers a single interface covering the full capability set. If you only need to quickly generate short videos from text or a single image, V3 T2V or V3 I2V is more lightweight.
Core Capabilities
Multimodal input with image placeholders
Supports three types of multimodal input: image_list (image assets), element_list (subject asset library), and video_list (video assets). Placeholders such as <<<image_1>>> in the prompt can reference images in image_list for precise image-to-video control.
Multi-shot assembly
Set multi_shot: true with shot_type: intelligence to let the model automatically break a long narrative into 3-6 shots; or use shot_type: customize with 1-6 manual multi_prompt entries for full control over each shot.
First-frame / end-frame specification
Use type: first_frame and type: end_frame in image_list to specify the opening and closing frames for precise start/end visual control.
Reference video and style transfer
Use refer_type: feature in video_list to learn the color grading, lighting, and cinematography style of a reference video, or refer_type: base to rewrite the scene while preserving the original subject's motion.
Optional audio generation
Set sound: on to generate sound synchronized with the visuals. Note: when the request contains video_list, sound can only be set to off.
Two resolution tiers
Switch tiers via the mode field: std (720p) is the lowest cost, pro (1080p) adds detail.
Negative prompts
negative_prompt describes content that should not appear in the video, helping the model avoid unwanted visual elements.
Watermark control
watermark_info.enabled controls whether a watermark is added, useful for branded content that must be delivered watermark-free.
Pricing
| Reference video | Mode | Unit price (USD/second) | 5-second cost |
|---|---|---|---|
| No | Standard | $0.084 | $0.4200 |
| No | Pro | $0.112 | $0.5600 |
| Yes | Standard | $0.126 | $0.6300 |
| Yes | Pro | $0.168 | $0.8400 |
Total = unit price × output video durationA reference video means video assets provided via video_list. Requests with a reference video cost 1.5x (a 50% premium) the price of those without.
Use Cases
- E-commerce and product showcase: Upload product images via
image_list, reference them precisely in the prompt with<<<image_1>>>placeholders, and generate product usage scenes, feature demos, and selling-point videos. - Short video and social media: Automatic multi-shot assembly (
intelligence) generates story-driven content for Douyin, Kuaishou, Xiaohongshu, and Bilibili;customizemode provides precise creative control. - Advertising and marketing videos: Brand stories, product ads, and campaign videos, with optional sound generation and Pro quality.
- Film and VFX concept design: Multi-shot assembly and reference-video style transfer for concept validation and previz in pre-production.
- Games and virtual worlds: Use the
element_listcharacter library to generate game cutscenes and character demos with consistent visual style. - First/end-frame control: Specify start and end frames via
first_frameandend_framefor precise shot transitions and scene changes.
Model Comparison
Kling Video O1 vs. Kling V3 T2V and Kling V3 I2V
| Dimension | Kling Video O1 | Kling V3 T2V | Kling V3 I2V |
|---|---|---|---|
| Endpoint | kwaivgi/kling-video-o1 |
kwaivgi/kling-v3/text-to-video |
kwaivgi/kling-v3/image-to-video |
| Positioning | Full-featured video generation | Text-to-video only | Image-to-video |
| Input | Text + image + video + character | Text only | Image + text |
| Multi-shot assembly | Supported | Not supported | Not supported |
| Reference video | Supported | Not supported | Not supported |
| Audio generation | Supported | Not supported | Not supported |
| Negative prompts | Supported | Not supported | Not supported |
| Watermark control | Supported | Not supported | Not supported |
| Resolution | 720P / 1080P | 720P / 1080P | 720P / 1080P |
| Max duration | Longer | 15 seconds | 15 seconds |
| 720P unit price | $0.084/sec | $0.084/sec | $0.084/sec |
| Best for | Full-featured video generation | Quick text-to-video | Quick video from an image |
Why Choose Kling Video O1?
When video production requires mixed multimodal input (image + video + character), multi-shot narrative assembly, reference-video style transfer, or precise first/end-frame control, Kling Video O1 covers the most complete set of video generation capabilities in the Kling series. A single interface handles the full pipeline from mixed asset input to final output, with no need to switch between interfaces.
Through iCreat, teams can call Video O1 via the asynchronous task queue, billed by output video duration, with no charge for failed tasks. Use Standard (720p) during testing to control costs, then switch to Pro for high-quality delivery once the approach is confirmed.
API
Kling Video O1 is called through an asynchronous task queue: submit a request to https://api.icreat.ai/v1/task/submit/kwaivgi/kling-video-o1 to get a task_id, poll https://api.icreat.ai/v1/task/query-status until the status becomes SUCCEEDED, then call https://api.icreat.ai/v1/task/get-result with the same task_id to retrieve the generated video. Authentication uses the iCreat API Key, and the request body Content-Type is application/json.
Placeholders such as <<<image_1>>> in the prompt reference assets in image_list. For single-shot generation, set multi_shot to false and provide prompt; for intelligent multi-shot, set multi_shot: true with shot_type: intelligence and provide prompt; for custom multi-shot, set multi_shot: true with shot_type: customize and provide 1-6 multi_prompt entries.
Specifications
Top-level fields
| Field | Type | Required | Description |
|---|---|---|---|
prompt |
string | Conditional | Single-shot text prompt. Required when multi_shot=false, or when shot_type=intelligence. Chinese ≤ 2000 characters, English ≤ 2000 words |
negative_prompt |
string | No | Negative prompt describing content that should not appear in the video |
multi_shot |
boolean | No | Whether multi-shot generation is enabled |
shot_type |
string | Conditional | Required when multi_shot=true. Supports customize, intelligence |
multi_prompt |
object[] | Conditional | Required when multi_shot=true and shot_type=customize; 1-6 entries |
image_list |
object[] | No | Image asset list |
element_list |
object[] | No | Subject asset list |
video_list |
object[] | No | Video asset list |
sound |
string | No | Whether to generate sound. Supports on, off; can only be off when video_list is present |
mode |
string | No | Generation mode. Supports std (720p), pro (1080p) |
aspect_ratio |
string | No | Output aspect ratio. Supports 16:9, 9:16, 1:1 |
duration |
string | No | Video duration in seconds |
watermark_info |
object | No | Watermark config; currently only supports the enabled field |
image_list[] items
| Field | Type | Required | Description |
|---|---|---|---|
image_url |
string | Yes | Image URL |
type |
string | No | Image type. Supports first_frame, end_frame |
element_list[] items
| Field | Type | Required | Description |
|---|---|---|---|
element_id |
string | Yes | Subject ID in the subject library |
video_list[] items
| Field | Type | Required | Description |
|---|---|---|---|
video_url |
string | Yes | Video URL |
refer_type |
string | No | Video reference type. Supports feature (style transfer), base (preserve motion, rewrite scene) |
keep_original_sound |
string | No | Whether to keep the original video sound. Supports yes, no |
watermark_info
| Field | Type | Required | Description |
|---|---|---|---|
enabled |
boolean | No | Whether the watermark is enabled |
Model specifications
| Category | Description |
|---|---|
| Model name | Kling Video O1 |
| Developer | Kuaishou |
| Endpoint | kwaivgi/kling-video-o1 |
| Model type | General-purpose video generation model |
| Invocation | Asynchronous task queue (submit → query-status → get-result) |
| Input modalities | Text, image, video, character |
| Output modalities | Video (optional audio) |
| Resolution | 720P / 1080P |
| Aspect ratios | 16:9 / 9:16 / 1:1 |
| Multi-shot | Supported (intelligence auto 3-6 shots / customize manual 1-6 shots) |
| Reference video | Supported (feature style transfer / base motion preservation) |
| Audio generation | Supported (sound: on) |
| Negative prompts | Supported |
| Watermark control | Supported |
| Billing | By output video duration |
Architecture
Kling Video O1 uses Kuaishou's Kling series video generation architecture: text prompts, image assets, video assets, and character assets are unified into a single encoding before entering the generation network. The model processes multimodal input tokens in a single context, uses a multi-shot planning module to break long narratives into shot sequences, then generates each shot and assembles them into a complete video.
Unlike the single-shot generation of Kling V3 T2V/I2V, Video O1 has built-in multi-shot assembly, so shot planning and automatic stitching happen at the model level without developers manually calling T2V/I2V multiple times and editing the results together. Reference videos pass through a style encoder that extracts color grading, lighting, and cinematography features, injecting them into the generation process for style transfer.
Notes
Small elements in complex scenes may drift in detail between shots; constrain the model with more detailed prompts or a reference video.
In multi-shot customize mode, each multi_prompt needs enough cinematography and subject detail to avoid jarring shot transitions.
When a request contains video_list, sound can only be set to off; new audio cannot be generated at the same time.
Non-Chinese/non-English prompts are untested and may reduce generation quality.
The same prompt produces different variations each run (by design); for deterministic output, use more detailed prompts or a reference video to constrain the model.
Evaluate the full task outcome, including visual quality, shot coherence, audio sync, generation latency, and cost. We recommend testing prompts and composition with Standard first, then switching to Pro for delivery once the approach is confirmed.
FAQ
When should I use a reference video?
Use refer_type: feature to keep visual consistency across a series (color grading, lighting, cinematography style); use refer_type: base to rewrite a scene or replace the subject while preserving the original motion. Requests with a reference video cost 1.5x (a 50% premium) the price of those without.
How many shots does multi-shot mode support?
shot_type: intelligence automatically plans 3-6 shots; shot_type: customize supports 1-6 manual multi_prompt entries.
Should I choose Standard or Pro?
Use the lowest-cost Standard (std) for testing motion, composition, and prompts; switch to Pro (pro) once the approach is confirmed for high-quality delivery.
Am I charged if generation fails?
No. Failed tasks (FAILED) incur no charge and can be safely resubmitted. Regenerating after SUCCEEDED is a new request and will be billed again.