Kling: Video O1

Kling VIDEO O1 is Kuaishou’s unified reference-led video model. It combines a text description with images, prepared subject elements or a source video to guide new scenes and focused changes to existing footage. Its core interface is a visual brief: identify the referenced material, describe its role and state the appearance, action or transformation you want in the resulting video.

A new shot can start from text, an opening image or a set of visual references. Use an image to establish a character or object, and describe the setting, action and viewpoint in words. When several references contribute different features, assign their roles explicitly. An original product brief might take the object from one image and the environment from another, then describe the object’s movement within that setting.

O1’s prompt can name images, elements and videos with specific reference markers, such as <<<image_1>>> or <<<video_1>>>. Keep those markers associated with the actual supplied asset and explain what should be taken from it. The maker’s prompt allowance is 2,500 characters. A concise instruction that identifies the subject and change is more useful than a collection of unrelated style terms.

For image-based references without a video input, the documented combined budget is up to seven reference images and multi-image elements. With a reference video, that image/element budget is four. These limits are scoped to image-based elements; other element types have separate compatibility rules. Choose the material that supplies the essential appearance or scene information rather than filling the budget with redundant views.

Image inputs use JPG, JPEG or PNG, supplied as an accessible URL or image data, with files up to 10 MB, dimensions of at least 300 pixels and an aspect ratio between 1:2.5 and 2.5:1. Prepare clear subjects and retain important details in the crop. If the task needs a character or product to remain recognisable, state which features should remain and inspect them throughout the completed clip.

Start and end frames condition a transition between visual boundaries. The documented O1 boundary-frame workflow requires a starting frame when an ending frame is supplied, and does not support start/end-frame control with more than two images in the array. When both boundary frames are used, additional subject elements are not part of that O1 configuration. Keep a frame-controlled task separate from editing a supplied base video.

A source video can have two distinct roles. As the base clip, it is the footage being edited: direct a specific addition, removal, style treatment or compositional change while stating what should remain. The edit follows the input clip’s duration and does not combine that base-video task with boundary-frame control or multi-shot generation. As a feature reference, the video supplies content, style or camera information for a new shot.

Reference-video prompting can also request a preceding or following shot. Identify the source clip and describe the new visible event, such as a continuation of the subject’s action or a reveal of the surrounding environment. This uses the reference to communicate visual information; review the join and subject continuity before assembling the new segment with the original footage.

The producer’s video-input profile accepts one MP4 or MOV source, up to 200 MB, with a minimum duration of three seconds. Its detailed input profile specifies dimensions from 720 to 2160 pixels and frame rates from 24 to 60 fps, with an input aspect ratio between 1:2.5 and 2.5:1. That documented source-video workflow returns 24 fps video. The exact allowed clip duration also depends on the selected model profile. O1’s new-generation range is three to ten seconds, distinct from the later 3.0 Omni duration range.

The maker’s O1 modes provide 720p Standard or 1080p Pro output. New-shot framing supports documented landscape, portrait and square choices where the selected workflow uses that control; a base-video edit follows its source framing. Source-video audio can be retained or discarded. O1’s reference and editing workflow should remain separate from the later 3.0 Omni native-audio and custom-storyboard profile.

Generation produces a task whose completed result contains the video artifact and actual duration. The maker’s result documentation says generated videos are deleted after 30 days, so save the completed clip promptly. Preserve the source assets and the brief with that result, then review the requested change, reference use and continuity. For an edit, compare the affected area and the parts that should remain. For a reference-led new shot, check that each asset supplied the intended visual information. Keep the completed clip available for the subsequent edit or production sequence.