Short answer: Omni Reference in Kling 4.0 is the multimodal entry for locking how assets look, move, and sound—up to 15 combined references—while Keyframes is a separate entry for placing exact visual moments on a timeline. The two entry points cannot be combined in one generation task: pick Omni when you need “like this asset,” pick Keyframes when you need “look like this at second N.”
- Omni total cap: ≤15 references across images, videos, voice profiles, and elements.
- Per-type caps: images ≤10; videos ≤5 (3–30s each, total ≤30s); voice profiles ≤3; elements ≤7 (video elements ≤3).
- Labeling: assets auto-number as
@Image1,@Video1, and so on—state each role in the prompt. - vs Keyframes: Keyframes uploads images only and targets timeline moments; Omni + Keyframes capabilities stay separate per task.
- Launch context: Kling 4.0 officially launched October 15, 2026 (Kuaishou IR).
What is Omni Reference in Kling 4.0?
Omni Reference is Kling 4.0’s default creation entry for multimodal control. You upload images, videos, voice profiles, and elements, then describe how each asset should contribute. The official User Guide positions Omni for asset- and concept-led generation—appearance, style, actions, and camera language—while Keyframes focuses on distributing key visuals along the timeline.
Native generation length is 3–30 seconds. Aspect ratios include 21:9, 16:9, 1:1, 9:16, and Auto. Prompt text can run up to 8,000 tokens. Last checked October 19, 2026 against the Kling 4.0 User Guide and feature page.
How many references can Omni Reference take?
Stay inside both the total and per-type ceilings. Reference videos, videos for editing, and video elements share the same video quota.
As you organize your reference assets, consult the Kling 4.0 overview and decide which appearance, motion, and sound details your brief needs to preserve.
| Input type | Quantity / size limits (User Guide) |
|---|---|
| Images | Up to 10; ≤30 MB each; aspect ratio 2:5–5:2 |
| Videos | Up to 5; 3–30 seconds each; total duration ≤30 seconds; ≤300 MB each; aspect ratio 2:5–5:2 |
| Voice profiles | Up to 3 per task; 5–30 seconds of audio to create a profile; ≤50 MB per file |
| Elements | Up to 7 total, including up to 3 video elements; source footage for a video element must be 3–8 seconds |
| Combined | Up to 15 reference assets across images, videos, voice profiles, and elements |
| Text | ≤8,000 tokens |
Credit costs and wall-clock generation times are not listed as fixed Omni prices on the User Guide tables above—use the on-screen quote in the product UI, and treat missing numbers as not listed on the official page.
How does Omni Reference differ from Keyframes?
Kling 4.0 exposes two primary entry points. Capabilities cannot be mixed inside a single generation task.
| Comparison | Omni Reference | Keyframes |
|---|---|---|
| Best for | Assets or a creative concept to reference | Predefined visual moments that must land precisely |
| What it controls | Appearance, style, actions, camera movements | Distribution of key moments along the timeline |
| Supported uploads | Images, videos, and voice profiles (plus elements within Omni limits) | Images only |
| Keyframe count | Keyframe-to-video is also available through Omni workflows described in the guide | Up to 10 keyframe inputs on the Keyframes entry |
| Combine in one task? | No—Omni and Keyframes capabilities cannot be combined in one generation task | |
For a deeper Keyframes walkthrough, see Kling 4 keyframes. For motion-led character work that is separate from Omni, see Kling Motion Control. New-user orientation lives in how to use Kling, and launch timing is summarized in Kling 4.0 release date. Automation via MCP/CLI is covered in the MCP and CLI guide.
How should I structure an Omni Reference prompt?
Assets number automatically in upload order (@Image1, @Image2, @Video1). State each role explicitly; if you only list IDs, the model infers intent and results get less predictable. Official structure:
- Asset roles — what each
@reference must preserve or contribute. - Overall settings — scene texture, visual style, camera tone.
- Sequence of events — timed beats with visuals, action, and emotion.
- Audio — dialogue/voiceover, ambience, music, effects when needed.
Prefer concrete visual language over abstract adjectives. The guide notes that only a few negative instructions are usually necessary (for example, no subtitles / no background music when the model might add them by default).
What prompt skeletons work for Omni Reference?
1) Character lock (identity consistency)
@Image1 defines the female lead; preserve facial features, hairstyle, and wardrobe.
@Image2 defines the cafe interior; preserve layout, wood tones, and window light.
Overall: realistic handheld, soft daylight, 12 seconds, 9:16.
0–4s: she enters and scans the room.
4–8s: she sits, places a notebook on the table, and looks toward the window.
8–12s: slow push-in on her face as she smiles.
Audio: quiet cafe ambience; no background music; no subtitles.
2) Product ad (creative recreation)
@Video1 supplies narrative pacing, shot scale, and camera path.
@Image1 is the product hero; keep logo, silhouette, and material exact.
Replace the original product in @Video1 with @Image1.
Keep camera moves, lighting direction, and duration aligned with @Video1.
15 seconds, 16:9, clean commercial look.
Audio: short VO only if needed; otherwise product foley and room tone.
3) Expression / local edit
Edit @Video1.
Change the character’s expression from neutral to a clear surprised smile at the midpoint.
Keep wardrobe, background, camera path, and duration unchanged.
Keep everything else unchanged.
Do not add new characters, text overlays, or music.
When editing, the User Guide requires explicit “Edit @Video1” language. Writing only “Reference @Video1” may be treated as reference generation instead of an edit.
Which Kling 4.0 features are not available yet?
Write status carefully. Official Kling pages mark 4K/1080p 10-bit HDR output and Video Extension up to 2 minutes as not yet released (the User Guide and feature page use “coming soon” only as a status label—not as a ship date). Enterprise/developer API access remains not yet released. Use the live product UI for what your account can actually select.
Want the same character in every shot? See Kling 4.0 character consistency: reference images vs Omni.
Frequently asked questions
What is Omni Reference?
Omni Reference is Kling 4.0’s multimodal entry for combining images, videos, voice profiles, and elements—up to 15 total—to guide appearance, motion, camera, and audio within one generation.
How many references does Kling 4.0 Omni allow?
Up to 15 combined assets, with per-type caps: ≤10 images, ≤5 videos (≤30s total), ≤3 voice profiles, and ≤7 elements (≤3 video elements).
Can I use Omni Reference and Keyframes together?
No. The User Guide states the capabilities of the two entry points cannot be combined within a single generation task. Run separate tasks if you need both styles of control.
Do Omni Reference videos share a quota with edit clips?
Yes. Reference videos, videos for editing, and video elements share the same video quota: up to 5 clips and ≤30 seconds combined duration.
How do @Image1 and @Video1 work?
Uploads are numbered automatically. In the prompt, name each ID’s role (preserve face, borrow camera path, edit this clip). Unlabeled IDs force the model to guess.
What is the maximum Omni generation length?
Native generation is 3–30 seconds per pass. Multi-pass Video Extension toward 2 minutes is not available yet on the shipping product described today.
Is 4K HDR available in Omni Reference today?
4K/1080p 10-bit HDR is not available yet. Do not plan delivery workflows as if HDR output is already shipping.
Does Omni Reference cost a fixed number of credits?
Fixed Omni credit tables are not listed on the User Guide specification tables cited here. Use the on-screen quote in the Kling UI; treat unpublished numbers as not listed on the official page.
When did Kling 4.0 launch?
Kuaishou’s IR materials state Kling 4.0 officially launched on October 15, 2026, highlighting Omni-style multi-reference control and longer native generation among other upgrades.
Should beginners start with Omni or Keyframes?
Start with Omni Reference for most asset-led and text-led ideas. Switch to Keyframes when you already know exact frames that must appear at specific timestamps.
Where do these facts come from?
- Omni vs Keyframes, input limits, prompt structure, edit wording: Kling 4.0 Model User Guide.
- Feature overview, 15 references, 10 keyframes, 30s generation, HDR/extension status language: Kling 4.0 feature page.
- Official launch date October 15, 2026: Kuaishou IR Kling 4.0 launch release.
Try Kling AI Today
Open Kling 4.0, choose Omni Reference, upload only the assets you can label, and generate a short 3–10 second test before a full 30-second pass—with free credits on sign-up, enough for one video.