kling-v3-omni is Kuaishou Kling's video model for reference-based creation, also known as Kling 3.0 Omni. It can generate short videos from text or a first-frame image, and can also create and edit using multiple images, reference videos, or existing footage. Its value lies in bringing character, scene, and style assets into a single task, making it suitable for short-form video production with existing visual assets and a need for repeated adjustments to shot content.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
api.acedata.cloudkling-v3-omniInput parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
The values and operating ranges above apply to this platform's access point; standard generation and Omni reference editing use different parameter combinations.
Multi-image references let characters, scenes, and styles avoid being fully restated in text. You can place subject photos, environment images, and style images in the same task, then explicitly reference them in the prompt and explain their respective uses. This is especially suitable for creating with existing brand assets, allowing shots to develop around clear visual references instead of starting from scratch each time.
Video references offer two different uses: base uses a video as the foundation to be edited, allowing changes to elements, composition, color, weather, or overall style; feature extracts characteristics such as style and camera movement to guide a new video. Before choosing, clarify whether you want to alter the original footage or borrow its characteristics, which helps organize assets and prompts.
Image-to-video uses the first frame as a starting point and can also add a last-frame constraint for the ending image, making it suitable for short videos with existing storyboards or key visuals. Standard generation can select aspect ratio, duration, and whether to generate accompanying audio; when combining reference assets, switch to the Omni workflow and use assets and text together to describe subject actions and scene changes.
Enter product images, brand environment images, and shot descriptions to generate short videos for social content or advertising concepts. Prompts should describe the product's position, display actions, and lighting atmosphere, while clearly specifying the role of each reference image. After delivery, focus on checking the product's appearance and image details, then adjust different creative versions around the same set of assets.
Use an existing video as base input and describe the style, background elements, or weather you want to change, such as transforming live-action footage into anime visuals. This is suitable for creating creative variations based on the original footage rather than rebuilding an entire storyboard. Audio can be retained or removed, with music, subtitles, and final editing added after the visual editing is complete.
Set a reference video as feature and pair it with character or scene images to explore new videos with specified visual styles or shot characteristics. This is suitable for director previews, storyboards, and series content concepts. For each task, first focus on a single shot objective; after delivering the short video, compare composition, action, and style before deciding the creative direction for subsequent shots.
If tasks mainly rely on text, first and last frames, and explicit camera parameter controls, kling-v3 is better suited to this workflow; if you need multi-image combinations, reference videos, or direct editing of existing shots, prioritize kling-v3-omni. Both can generate short videos with sound, but Omni does not support camera_control, so do not confuse reference camera movement with camera parameter controls.
kling-o1 also supports all-in-one references and is not limited to processing a single image. Specific reasons to choose kling-v3-omni are the need for more flexible generation durations, sound in standard generation, or processing longer reference videos. Existing O1 workflows can retain similar asset organization approaches, but when switching models, reset the duration and audio combination rather than merely replacing the name.
Use text2video and image_list/video_list for reference creation, and specify the subject, style, and actions by number in the prompt. base is for editing existing videos, while feature is for video feature reference.
Specify model=kling-v3-omni for /kling/videos, starting with std/pro and whole-number 5-second increments. When reference video is included, set generate_audio=false; original audio is controlled by keep_original_sound. Do not mix reference tasks with standard generation parameters.
Save the task_id and video ID, and retrieve the final video through queries or callbacks; compare it with the assets to check the subject, actions, and preserved areas. If you choose to preserve the original audio, verify that track; when generating from scratch with sound enabled, then check the new audio.
Reference images define the character's clothing and appearance, while a reference video defines a slow panning shot; generate a clip of the character standing in a bookstore while maintaining the subject's characteristics.
Use image_list/video_list and reference the assets in the prompt; when video is included, do not generate new sound at the same time, and use keep_original_sound to preserve the original audio.
Kling O3 and Kling 3.0 Omni are names for this model; kling-v3-omni is used when calling it on this platform. It corresponds to a different model from kling-v3 and kling-o1 respectively. When creating a task, explicitly specify the target model to avoid relying on the default selection.
If you already have a definite starting image, use image2video and provide a first frame, adding a last frame if necessary; if you need to combine character, environment, and style materials, use text2video with image_list. Reference images must not only be uploaded, but also explicitly referenced in the prompt, explaining their purpose in the shot.
Yes, but the materials have different roles. refer_type=base means editing this base video; feature means drawing on characteristics such as its style and camera movement to guide creation. The former is suitable for modifying existing content, while the latter is suitable for reference-driven new shots; the two task types should use different prompt objectives.
Not all tasks can be covered by the same combination. Regular generation supports 4k and audio options, while Omni reference requests do not use 4k; after adding a reference video, new audio generation must be disabled. If you want to retain the material's sound, set keep_original_sound instead of enabling generate_audio.
Submit the model, operation, and creative input to POST /kling/videos. Set async=true to first obtain a task_id, then query the task; you can also use callback_url to receive the result. After completion, obtain the video through video_url, and save the task ID, status, and material configuration for convenient tracking and iteration.
Choose a development approach or application suited to this model and follow its setup guide.
Python SDK Integration Guide
SDK installation, authentication and examplesTypeScript SDK Integration Guide
SDK installation, authentication and examples