Quick Take
A short AI video can look polished and still miss the shot you needed. The character reaches the wrong mark, the turn comes too late, or the ending lands on a different image. Kling 4.0 raises the stakes by offering up to ten keyframe inputs for one generated scene. Can a creator plan the beginning, turning point, and end—and get a shot that follows them?
That is a test, not a result. AI Bulgogi has not generated a Kling 4.0 video.
What Happened
In an October 16 company-issued announcement, Kling described native generation up to 30 seconds, up to ten keyframes, and up to fifteen combined reference assets. Those numbers describe available controls, not proof that a particular scene will keep its timing or identity.
The Kling 4.0 user guide makes the creative choice more specific. Its Keyframes entry point lets a maker place image milestones at chosen moments on the timeline. Omni Reference is for a broader mix of reference assets and direction. The guide says those entry points cannot be combined in one generation task. It also says Kling 4.0 Flash does not currently support multiple keyframes.
Kling says the model is available on its website, with developer API access to follow. The guide warns that launch specifications are subject to the actual product interface. Its prose marks HDR and two-minute extension as coming soon, despite some present-tense specifications and announcement language. For this story, the news is the documented scene-planning control. The result of using it is still the open part.
Why It Matters
The headline sounds like a bigger prompt: ten chances to tell the model what to show. The more interesting change is a timed visual brief. A creator can put the intended opening, change, and closing image on the timeline before asking for motion between them. That turns “looks good” into a sharper question: did the shot arrive at the required states when the story needed them?
The distinction between the two entry points matters here. If precise visual beats are the requirement, the guide points to Keyframes. If a task needs a broader asset mix, it points to Omni Reference. Treating those as one giant control panel would misstate the workflow—and make a test impossible to interpret.
The Creator Angle
Picture a 15-second brand scene: a product is hidden, a character reveals it, and the final frame holds a clean pack shot. A beautiful output with an early reveal or a changed product is still a failed brief. The team's real unit of value is the accepted shot, with the right beat order, character, object, framing, and export.
Kling's announcement argues for more stable motion and narrative coherence. Those are vendor claims. A creator deciding whether to use the tool needs evidence from their own approved material, account, settings and output. The number of generated seconds tells less than the number of revisions needed to get one usable scene.
Workflow Drop
Build the test on paper before spending anything:
- Draw three original frames. Specify the opening, turning point, and closing image. Use rights-cleared characters and props. Write the required action and approximate time beside each frame.
- Choose the correct entry point. The guide's Keyframes workflow is the documented path for timed image milestones; do not assume it can share a task with Omni Reference assets or that Flash has the same controls.
- Set acceptance rules. Make a short checklist for beat order, moment of the turn, character and object identity, unintended changes, sound, and output metadata.
- Check access and cost first. Confirm the signed-in account's model, plan, terms, quote and export choices before any authorized generation. Log every attempt, retry and actual charge.
- Compare the finished file with the plan. If testing is later authorized, inspect it frame by frame and calculate the cost of the accepted shot, not just the first attractive preview.
That ledger reveals whether the control improves a deliverable or merely makes the setup feel more direct.
Hot Take
Ten keyframes are useful if they make a director's intent easier to test. They are not ten guarantees.
The strongest promise in Kling's launch is that creators can plan more of a scene before generation. But the production win only exists when the resulting file keeps the beats that matter, survives export, and costs less to accept than the alternatives. A gorgeous miss is still a miss.
Bottom Line
Kling 4.0 documents a way to place up to ten visual milestones across a short generated scene. It does not, by itself, prove that the model will follow your storyboard.
Make the plan visible. Then judge the shot against it.