Kling 4.0 for Short Drama, Tested: API Timings, Real Costs and 30-Second Shots

On 15 October 2026, Kuaishou’s Kling AI announced that Kling 4.0 is officially available: on the new creation page at kling.ai, and through MCP and CLI. On 23 October 2026 fal opened the Kling 4.0 API to all customers, and on 24 October we switched it on in DramaSo after running real renders through it.

Most coverage will retype the spec sheet. This post asks the short-drama question instead: which of these changes actually touch the beats that make or break a vertical episode, and what you can do about them tonight in DramaSo, where Kling 4.0 now sits at the top of the model picker. We also publish the numbers from our own API test runs below.

The frames below are the kind of storyboard this post keeps coming back to: a few stills that a Kling render has to honour.

What is live, what is announced

Kling’s launch post from 30 September describes the model; the 15 October announcement moves it from early access to general availability. Split the two lists before you plan an episode around them:

	Kling 4.0 (per Kling)	Status on 24 Oct 2026
Single generation	Up to ~30 seconds	Live on kling.ai
Keyframes	Up to 10 keyframe images	Live on kling.ai
Omni reference	Up to 15 assets: ≤10 images, ≤5 videos, ≤7 subjects (a subject can carry a voice)	Live on kling.ai
Audio	Two-channel stereo, improved lip-sync	Live on kling.ai
Access	kling.ai creation page, MCP, CLI	Live
On fal	fal web UI and fal Agent	Live since 21 Oct
fal API	Text, image and reference to video	Open to all customers since 23 Oct
In DramaSo	Model picker, via fal’s API	Live since 24 Oct
10-bit HDR	4K and 1080p	Coming soon
Kling’s own developer / enterprise API	kling.ai API	Coming soon
Extension to 2 minutes	From an uploaded video	Coming soon

Practical rule: Plan around the “live” column. HDR and the 2-minute extension are still a promise, not a deadline.

On 21 October fal announced Kling 4.0 in its web UI and fal Agent only. Two days later the API followed for every customer. That second step is what DramaSo needs, because the studio calls models through their APIs. The picker now lists Kling 4.0, Kling O3 Pro, Kling O3 and Kling v2.5 Turbo; see the Kling short drama generator for the current rows.

We ran Kling 4.0 through the API: the numbers

Before turning it on, we sent real jobs to fal’s kling-video/v4 endpoints on 24 October 2026 and measured the files that came back. Early-access calls a week earlier had returned an unrelated 12-second placeholder clip, so we checked every output with ffprobe instead of trusting “completed”.

What we checked	Result
Request	Image to video, first + last frame, 3 s, 720p
Output length	3.04 s (the placeholder era returned 12 s for the same request)
Output format	1280×720, 24 fps, H.264
Audio	Stereo AAC, 44.1 kHz, always included: the v4 API has no audio-off switch
Time to render	About 162 s for two parallel 3 s jobs; about 298 s for a single direct call that sat in the queue
List price on fal	$0.157 per second for text, image and reference to video, same at 720p and 1080p
Start frame	The start image is used as the exact first frame

For comparison, our earlier runs put Kling O3 Pro at roughly 318 s for a 5-second shot and Kling O3 at roughly 94 s. Kling 4.0 is not the fast option; it is the long-shot option.

What the price means per shot at fal’s list rate:

Shot length	Kling 4.0 ($0.157/s)	Kling O3 Pro ($0.112/s, audio off)	Kling O3 ($0.084/s, audio off)
5 s	$0.79	$0.56	$0.42
15 s	$2.36	$1.68	$1.26
30 s	$4.71	not possible in one shot	not possible in one shot

One behaviour worth knowing: in one test our prompt described a different scene from the start image. Kling 4.0 kept the start image as the opening frame and then cut to the scene in the prompt. For short drama that is a free jump cut you did not ask for.

Practical rule: Write the prompt about the keyframe you hand it. If the prompt and the image disagree, Kling 4.0 will honour both and cut between them.

A few API facts that shape how you board for it: duration is any whole number from 3 to 30 seconds; resolution is 720p or 1080p; image to video takes its aspect ratio from the start frame; text to video offers 16:9, 9:16, 1:1 and 21:9; reference to video takes up to 10 images, 5 videos and 7 elements; there is no negative prompt field.

30 seconds: one beat per shot, not one shot per beat

A vertical episode runs one to three minutes, and it lives or dies on a few beats: the hook, the reveal, the cliffhanger. With 5–15 second shots, every one of those beats is two or three clips glued together, and each cut is a chance for the lead to change face or jacket.

A typical episode still: the hook frame that the cliffhanger has to match.

A ~30-second generation lets a whole beat — confrontation, reaction, the line that lands — stay inside one continuous render. That is fewer seams in exactly the places viewers rewind.

What it does not fix: a 30-second shot of a weak beat is still a weak beat. Write the episode first; the seven-step workflow still applies.

Practical rule: Spend long shots on the beats with dialogue and reaction in them. Keep establishing shots and B-roll short.

10 keyframes: storyboard stills become the contract

Kling O3 already gives you first and last frame. Kling 4.0 raises that to up to ten keyframe images for poses, expressions, transitions and story beats. For short drama that is the storyboard turned into a control surface: the slap lands on frame four, the smirk on frame seven, the door opens on frame nine.

That only helps if the stills exist and agree with each other. Lock them on the AI Storyboard Generator before you spend video credits, and keep each character’s reference sheet constant — the same habit the Kling Omni storyboarding guide teaches for 3.0.

Plan the time budget on paper before you plan keyframes: a 30-second shot only helps if the beat inside it is already decided.

Voice in the reference: casting carries across shots

Omni reference in 4.0 takes up to 15 assets per generation, and a subject can include a specified voice. Combined with stereo audio and better lip-sync, that addresses the second most common giveaway in AI drama after the changing face: the lead sounding like a different person in every clip.

The practical move is the same one that works today: fix the cast first — face, wardrobe and voice — and reuse it on every shot. DramaSo already voices the whole script so you can hear the episode before rendering video; that audio pass is where you decide what each character should sound like.

How to use Kling 4.0 in a short drama today

Kling 4.0 is in the picker, but it is the most expensive Kling row per second, so spend it where it changes the episode:

1. Open DramaSo and paste a script, or start from the AI Drama Generator.
2. Listen to the episode as dialogue audio and fix pacing before any video credits are spent.
3. Mark the two or three beats that deserve one continuous shot — those are your future 30-second Kling 4.0 shots.
4. Lock the storyboard stills and character references for those beats.
5. Render those beats with Kling 4.0, and the rest with Kling O3 (up to 15 seconds, first/last frame) or Kling O3 Pro (1080p). Mix in Seedance or MiniMax H3 per shot where they fit.
6. Export captioned 9:16 for TikTok, Reels and Shorts.

If you boarded an episode on Kling O3 before 4.0 arrived, those marked beats are the only ones to re-render; the script, stills and cast carry over unchanged.

Practical rule: Treat Kling 4.0 as an upgrade to a few shots, not a reason to rewrite the episode.

When to reach for 4.0 — and when not

Worth it for long dialogue beats, choreographed reveals that need several pinned poses, and scenes where voice consistency matters more than raw resolution.

Not needed for short inserts, establishing shots and drafts. At list price 30 seconds of Kling 4.0 costs $4.71 against $2.52 for the same 30 seconds on Kling O3. A cheaper row does those shots just as well, and DramaSo shows the credit price of each shot before it runs.

The comparison with other tools lives in Runway vs Kling for short drama and AI short drama generators compared. This post has one job: separate what Kling 4.0 shipped from what it promised, put real numbers on it, and point it at the beats that matter.