Kling Video O3 Pro Reference-to-Video API by Kuaishou
kwaivgi/kling-video-o3-pro/reference-to-video
Kling Omni Video O3 Reference-to-Video generates creative videos using character, prop, or scene references. Professional quality with up to 7 reference images and optional video input.
Kling Video O3 Pro Reference-to-Video은(는) Kuaishou에서 개발한 모델입니다. Atlas Cloud(운영: Atlas Cloud AI LLC)는 해당 모델에 대한 액세스를 제공할 뿐이며 이를 소유하지 않습니다. 모든 상표는 각 소유자에게 귀속됩니다.
Kling Video O3 Pro Reference-to-Video
Kling Video O3 Pro Reference-to-Video uses Kuaishou's Kling 3.0 Omni model in Professional mode (1080P output) to generate new videos guided by reference images, subjects (elements) and an optional reference video. Describe a scene that involves the people, objects or styles in your references, and the model combines them into one coherent video with consistent identity.
Why Choose This?
- Character-consistent generation Keep the identity of people, products and objects from your reference images.
- Multi-reference support Up to 7 reference images, or up to 4 together with a reference video.
- Video reference Borrow camera movement, motion or style from a reference video, or generate the next or previous shot of it.
- Audio options Keep the reference video's original audio, or generate native audio when no video is used.
- Flexible output 16:9, 9:16 or 1:1, any length from 3 to 15 seconds, with single-shot or multi-shot generation.
Parameters
| Parameter | Required | Description |
|---|---|---|
| prompt | Conditional | Description of the scene and action. Max 2,500 characters. Refer to inputs with <<<image_1>>>, <<<video_1>>> and <<<element_1>>>, numbered in upload order. Required unless multi_shot is on with shot_type = customize. |
| video | No | One reference video (URL), used as a feature reference: camera movement, motion, style, or the next/previous shot. MP4 / MOV, at least 3 s long, width and height 700–4553 px, 24–60 fps, up to 200 MB, aspect ratio between 1:2.5 and 2.5:1. |
| images | No | Reference images of characters, objects, scenes or styles. JPG / JPEG / PNG, up to 10 MB each, at least 300 px on each side, aspect ratio between 1:2.5 and 2.5:1. Images and image-type elements together: up to 7 without a reference video, or up to 4 with one. |
| multi_shot | No | Generate a multi-shot video (default: false). With a reference video, only shot_type = intelligence is supported. |
| shot_type | Conditional | Required when multi_shot is true. customize: you define every shot in multi_prompt. intelligence: the model splits prompt into shots. |
| multi_prompt | Conditional | Required when shot_type is customize. 1–6 shots, each {index, prompt, duration}. Each shot prompt can be up to 512 characters, each shot lasts at least 1 second, and the shot durations must add up to duration. |
| elements | No | Up to 3 subject references (mapped to Kling element_list). Each item either points to an existing subject by element_id, or creates one inline with element_name, reference_type (image_refer / video_refer) and frontal_image + refer_images (up to 3) or refer_videos (1). Video-type subjects cannot be combined with a reference video. |
| aspect_ratio | No | 16:9 (default), 9:16 or 1:1. |
| duration | No | Video length in seconds: any integer from 3 to 15 (default: 5). |
| keep_original_sound | No | Keep the reference video's original audio (default: true). Only applies when video is provided. |
| sound | No | Generate native audio (default: false). Only available without a reference video. When video is provided, sound generation is turned off. |
Output resolution is fixed at 1080P for Pro mode.
How to Use
- Upload reference images — characters, products, scenes or style references.
- Add a reference video (optional) — for camera movement, motion or style, or to continue the shot.
- Add elements (optional) — attach saved or inline subjects.
- Write your prompt — refer to each input by its token, e.g.
<<<image_1>>> walks through the park with <<<image_2>>>, following the camera movement of <<<video_1>>>. - Choose aspect ratio and duration — 16:9 / 9:16 / 1:1, 3 to 15 seconds.
- Set audio — with a reference video, choose whether to keep its original sound. Without one, you can turn on
sound. - Run — submit and download your video.
Best Use Cases
- Character-Driven Storytelling — Scenes starring specific people from your reference photos.
- Video Continuation — Generate the next or previous shot of an existing clip.
- Camera & Style Transfer — Reuse the camera movement or look of a reference video with new subjects.
- Marketing & Ads — Spokesperson or product videos from still photos.
- Premium Production — 1080P character-consistent scenes for final delivery.
Pro Tips
- Number the tokens in upload order: the first image is
<<<image_1>>>, the second is<<<image_2>>>, and so on. - Reference images with clear faces and distinct features give the best identity consistency.
- To extend a clip, provide it as
videoand prompt "Based on<<<video_1>>>, generate the next shot." - Test character consistency with O3 Std or short durations (3–5 s) before rendering long Pro clips.
- Remember the lower image limit (4) when you use a reference video.
Notes
- With a reference video: up to 4 images plus image-type elements, no video-type elements, no generated
sound, and multi-shot only withintelligence. - Without a reference video: up to 7 images plus image-type elements.
- Only one reference video per request.
- Duration accepts any integer from 3 to 15 seconds.
- Make sure image and video URLs are publicly accessible.
Related Models
- Kling Video O3 Std Reference-to-Video — Same model in Standard mode (720P) at lower cost.
- Kling Video O3 Pro Video Edit — Edit existing videos with natural-language instructions.
- Kling Video O3 Pro Text-to-Video — Generate video from text with Kling 3.0 Omni.
- Kling Video O3 Pro Image-to-Video — Animate a start frame with Kling 3.0 Omni.
