Kling
Kling is a video model family by Kuaishou. Use Kling when a shot needs character references, an existing first frame, or a separate movement donor. These are different modes: an image anchors appearance, while a motion donor describes performance. The motion-donor operation is called Avatar (Аватар) in the model picker; for cartoon or animal heroes and clips cut off at the knee, see also DreamActor 2. Avatar has two drivers, chosen in the model window under “What animates” (Чем оживить): clip (ролик) takes the movement from a source clip; voice (голос) takes it from a sound, and Kling’s Avatar-by-voice address is described below. For a richer body and gestures under a sound, see also OmniHuman 2.
Resolution, quality and sound
Section titled “Resolution, quality and sound”Choose Standard for 720p or Pro for 1080p. The quality choice selects the corresponding provider operation; there is no separate resolution switch. Standard is the initial choice. Both qualities are connected for generation, reference generation, editing, motion transfer and Avatar by voice.
Generation supports 3–15 seconds with model-generated sound on by default; sound can be turned off. Editing and motion transfer preserve the source sound by default and can turn it off. Their length follows the source rather than an independently chosen generation length.
The separate Kling Turbo variants are not connected by this selector. Settings were checked against the current Pro API and motion-control API on 20 October 2026; this is a settings check, not a new visual-quality test.
Avatar by voice
Section titled “Avatar by voice”With the driver voice, Kling animates one hero picture (first frame) to one sound (sound 1): the mouth, head and shoulders follow the sound, and the sound of the input arrives in the output. This is a separate provider address from the motion donor, but for you it is the same operation, Avatar. The quality choice here changes the address, the output size and the price: Standard gives 720×1280 and Pro gives 1072×1920 at 30 frames per second; the inputs are identical. Text is optional.
The output is as long as the sound, rounded up to a whole number of 8-frame steps and never shorter than about 2.1 seconds, and the price is counted by this output length, so a very short sound still costs the minimum. The cost is shown before the run. The vendor publishes no limits for the sound; the studio accepts one to 300 seconds. In the owner’s trial of 22 October 2026 (one cartoon picture, 8.5 seconds of a Russian vocal) Kling moved the mouth and head in a restrained way, with less body than OmniHuman 2. This is one observation of one scene type, not a result for real faces or plain speech.
| Setting | Value |
|---|---|
| Model ID | kling_v3_standard |
| Name | Kling |
| Availability | available |
| Default for new nodes | no |
| Earlier IDs resolving to this record | kling_o3_standard, kling_v3_motion_control |
Common declared limits
Section titled “Common declared limits”| Setting | Value |
|---|---|
| Capabilities | text to video, image to video, native audio, multi reference, negative prompt, dialogue |
| Prompt required | yes |
| Maximum prompt length | 2500 |
| Duration | 3–15 seconds; fallback 5 seconds |
| Audio | model-generated audio, off; fallback model-generated audio |
| Aspect ratio | 9:16, 16:9, 1:1; fallback 9:16 |
| Declared language support | prompt: English; speech: English, Chinese; tier: bronze |
| Controls | quality: standard, pro; new node standard; fallback standard |
| Element input form | sheet |
| Content constraints | people: restriction level: ok |
| Negative prompt strategy | separate negative-prompt field |
Mode: from_first_frame
Section titled “Mode: from_first_frame”| Setting | Value |
|---|---|
| Aspect ratio | not declared |
| Combined reference limits | video frame: min side: 720 |
| Video reference | refused |
| First frame | required |
| Last frame | accepted; optional |
| References | maximum Elements: 3; views per Element: 4; video: maximum Elements: 1 |
Mode: text_to_video
Section titled “Mode: text_to_video”Uses the common declared limits above.
Mode: from_references
Section titled “Mode: from_references”| Setting | Value |
|---|---|
| Combined reference limits | video frame: min side: 720 |
| Element input form | frames |
| Video reference | refused |
| Negative prompt strategy | instruction in prompt text |
| References | maximum Elements: 3; views per Element: 4; video: maximum Elements: 1 |
Mode: edit_video
Section titled “Mode: edit_video”| Setting | Value |
|---|---|
| Duration | same as source video; fallback same as source video |
| Audio | origin, off; fallback origin |
| Aspect ratio | not declared |
| Combined reference limits | video frame: min side: 720; videos: 1; audio inputs: 0; audio without visual input: no |
| Negative prompt strategy | instruction in prompt text |
| Source video | required |
| References | maximum Elements: 3; views per Element: 4 |
Mode: motion_transfer
Section titled “Mode: motion_transfer”| Setting | Value |
|---|---|
| Prompt required | no |
| Duration | same as source video; fallback same as source video |
| Audio | origin, off; fallback origin |
| Aspect ratio | not declared |
| Combined reference limits | video frame: min side: 340; images: 5; videos: 1; audio inputs: 0; combined maximum: 6; audio without visual input: no; video reference seconds: minimum: 3; maximum: 30.05 |
| Negative prompt strategy | instruction in prompt text |
| First frame | required |
| Source video | required |
| References | maximum Elements: 1; views per Element: 4 |
Mode: avatar_audio
Section titled “Mode: avatar_audio”| Setting | Value |
|---|---|
| Prompt required | no |
| Duration | same as source video; fallback same as source video |
| Audio | origin; fallback origin |
| Aspect ratio | not declared |
| Combined reference limits | images: 1; videos: 0; audio inputs: 1; combined maximum: 2; audio without visual input: no; audio reference seconds: minimum: 1; maximum: 300 |
| Negative prompt strategy | instruction in prompt text |
| First frame | required |
Unavailable modes: recast.
Limits apply together. initial is the studio choice for a new node; default is the fallback for an existing node without that value. Declared language or input support does not guarantee a successful result.
Prompting playbook
Section titled “Prompting playbook”- Name each character and the action that belongs to them. For a reaction shot, describe the visible reaction instead of several abstract emotions.
- Keep camera movement separate from actor movement. Request one dominant camera move per shot.
- For dialogue, identify the speaker and exact words. Keep an off-screen narrator out of the on-screen character dialogue.
The built-in Kling authoring skill adapts a scene request for this family through Prompt for the model. It prepares instructions; it does not render video or certify the result.
Studio example
Section titled “Studio example”Prepare an Element for a courier and a scene image of a station. Select Kling and first-frame mode on Video. Ask Prompt for the model: “The courier lifts a small parcel from the bench, checks its label and smiles. The camera slowly approaches. Only station ambience.” Review the resulting prompt, then generate a short clip.
A run may consume credits. Review the prompt and inputs, then inspect the finished clip for action, identity, framing and sound. For automation, use the supported MCP workflow.
Known failure modes
Section titled “Known failure modes”A character sheet produced a second copy of the character in a reference-mode test on 30 September. Clean character frames passed the corresponding test on 28 September. Sheets worked as additional character inputs with a scene first frame and with motion transfer on 30 September. Input choice is mode-specific.
Evidence and limits
Section titled “Evidence and limits”The older editing exam used individual frames. Editing now uses a sheet input with an outstanding replacement exam; the older editing success does not certify the current input.
On 28–30 September 2026, the tested reference, first-frame, edit and motion-transfer cases retained the fictional character. The reference and edit requests used O3; the first-frame and motion-transfer requests used v3. The edit retained the source street and movement. This is a small set of storyboard-reviewed cases, not a general identity guarantee. A video sample with multiple people was refused on 4 October; that does not establish that all possible single-character clips are refused.
The studio registry establishes available inputs and modes. A recorded render establishes only its observed outcome. The applicability guide explains the distinction. See also character identity, static camera and video style samples.
See the model–technique applicability matrix for connected operations, dated outcomes and configurations that remain untested.
Reviewed 14 October 2026 against the current studio contract and dated studio research. Official model-author documentation checked on that date; author capabilities may exceed Vini’s connected controls. Recheck this playbook when the model version, quality or mode changes.