KlingParameters & settingsPrompting basics
Kling 4.0 Guide: Omni Reference, Keyframes and 30s Clips
Kling 4.0 generates 3 to 30 second videos at 720p, 1080p or 4K from two entry points: Omni Reference, where you combine up to 15 assets (images, videos, voice profiles and elements) and refer to them in the prompt as @Image1 or @Video1, and Keyframes, where up to 10 images fix what appears at set moments. Kling 4.0 Flash is the faster, cheaper mode, up to 20 seconds at 720p, without keyframes.
Kling released its 4.0 model user guide on 14 October 2026. It is the biggest change to how you work in Kling since VIDEO 3.0 added native audio and multi-shot generation earlier in the year: clips can now run to 30 seconds, references from many sources work together in one task, and you can pin the picture at up to ten moments along the timeline. Kling's own manual notes that features and parameters are subject to the actual product interface, so treat the numbers below as launch specifications and check them in the app.
What is new compared with VIDEO 3.0
| Setting | VIDEO 3.0 | Kling 4.0 |
|---|---|---|
| Duration | 3 to 15 seconds | 3 to 30 seconds, in 1-second steps |
| Resolution | Not listed in the comparison | 720p, 1080p or 4K; 10-bit HDR announced as coming soon |
| Aspect ratios | 16:9, 1:1, 9:16 | Adds 21:9 (plus Auto) |
| Keyframes | Start and end frames | Up to 10 keyframes |
| References | Elements and images | Up to 15 assets: images, videos, voice profiles, elements |
| Audio | Native audio | Two-channel stereo, more languages and accents |
| Prompt length | A lower limit | Up to 8,000 tokens (marked new) |
The 3.0 column summarises Kling's VIDEO 3.0 user guide. Kling also lists German, French and Hindi as new languages in 4.0, alongside Chinese, English, Japanese, Korean, Spanish and Portuguese, and says multilingual text, emojis and logos render more reliably in motion.
Omni Reference or Keyframes
Kling 4.0 has two entry points, and they cannot be combined in one generation.
- Omni Reference is the default and the most capable. Use it when you have assets to reference, such as a character, a product, a location or a video whose pacing you want, or when you start from an idea alone. Text to video also runs here.
- Keyframes keeps the familiar image-to-video workflow (a first frame, or first and last frames) and extends it to as many as 10 keyframes. Use it when particular moments must look exactly like your images.
Both offer 720p, 1080p or 4K, and durations from 3 to 30 seconds with a default of 10. In Keyframes, the aspect ratio follows the first keyframe once you upload images.
Working with Omni Reference
Kling's limits per task are up to 10 images (30 MB each), up to 5 videos (3 to 30 seconds each, 30 seconds in total, 300 MB each), up to 3 voice profiles made from 5 to 30 seconds of audio, and up to 7 elements, of which up to 3 can be video elements. The total across all types is 15. Reference videos, videos for editing and video elements share the same quota of 5 clips and 30 seconds.
Assets are numbered in upload order: @Image1, @Image2, @Video1. The single most useful habit, according to Kling, is to state each asset's role, because without a role the model guesses:
@Image1 defines the scene. Preserve its layout, furniture and lighting.
@Image2 defines the woman. Preserve her face, hairstyle and clothing.
The woman walks to the window, stops and looks back.
Kling's suggested prompt structure has three parts: asset roles, overall settings (scene and texture, visual style, camera tone), and a sequence of events, which can include timings such as "3.0–8.0s". Omni Reference can also edit an existing video: change expressions, movement, camera position, style or background while keeping the rest, using up to 5 clips with one primary video.
Writing prompts for 4.0
Kling's manual is unusually direct about prompting:
- Concrete beats abstract. Replace "premium look" with what you would see: side backlight on the object's edges, a blurred background, fine matte grain. Replace "vintage feel" with yellowed walls, worn furniture edges and warm incandescent light.
- Describe what you want, not what you do not. Kling says negative instructions can increase errors. The two it does recommend are "no subtitles" and "no background music", because the model may add them by default.
- Start short, then add what matters. One sentence covering subject, scene, action and emotion first; then camera work, timing, audio and what must be preserved.
- For a style change, add exclusions. When asking for animation, explicitly exclude photorealism so styles do not mix.
Audio and smart storyboarding are on by default with no separate switches. You cannot upload music or sound effects; audio input is limited to voice profiles, and background sound is generated from the picture and your prompt's audio description.
Kling 4.0 or 4.0 Flash
Flash mode is built for speed and cost: Kling describes it as ideal for rapid iteration and short-form content, with videos up to 20 seconds at a maximum of 720p. It does not currently support start and last frames or multiple keyframes. A sensible workflow is to test ideas in Flash and generate finals in 4.0 at 1080p or 4K. Kling also announces video extension for 4.0, up to 2 minutes, as coming soon; until then, earlier models' Extend with Prompts remains the route to longer clips, covered in how to extend a Kling video.
Settings at a glance
| Setting | What it does | Start with |
|---|---|---|
| Entry point | Omni Reference or Keyframes | Omni Reference unless frames must be exact |
| Model | 4.0 or 4.0 Flash | Flash to explore, 4.0 for finals |
| Resolution | 720p, 1080p, 4K | 720p tests, 1080p delivery |
| Duration | 3 to 30 s | 10 s (the default) |
| Aspect ratio | 21:9, 16:9, 1:1, 9:16, Auto | The delivery shape |
| Asset roles | Tells the model what each @asset is for | One line per asset |
For consistent characters across clips, our guide to Kling Elements still applies; elements are one of the asset types Omni Reference accepts.
Try it
@Image1 defines the café interior. Preserve its layout and warm window light.
@Image2 defines the barista. Preserve her face, hair and apron.
The barista slides a cup across the counter towards the camera and smiles. Slow push in, shallow depth of field. Soft café ambience, no background music, no subtitles.
Run it first in 4.0 Flash at 10 seconds to check the action, then in 4.0 at 1080p. If the face drifts, add an element of the barista built from a front and a side view and reference it instead of @Image2.