Kling 4.0 on MITO: What It Is Best At, and When Director Picks It
One unbroken ski run, four different outfits. That is the shape of job Kling 4.0 is built for.
Kling 4.0 went live for everyone on 15 October, and it is now on MITO.
The short version is that it got longer, sharper and much better at holding on to a reference. The useful version is knowing which jobs that combination actually wins, which is what this post is about.
What changed
Four things matter more than the rest.
It holds a thirty-second take. One generation, up to 30 seconds, with up to 10 keyframes to pin the beats along the way. You are no longer assembling a long piece out of four-second fragments and hoping they match. On MITO that puts it in a club of three: Seedance 2.5, Wan 3.0 and now Kling 4.0.
It takes a lot of reference, and the reference sticks. Up to 15 items in a single generation: as many as 10 images, 5 videos totalling 30 seconds, and 7 elements, 3 of which can themselves be video. Those can be three-view character sheets, storyboards, wireframes, grids, depth maps or white-model passes, which is a far more precise way to say what you mean than a paragraph.
It does what the prompt said. This sounds like faint praise and is not. Video generation is probabilistic, and getting everything you asked for in one go is genuinely hard, so a model that follows a brief closely is worth more than one that occasionally produces something prettier. Kling 4.0 is strong here, which is what makes the 8,000-token prompt limit useful rather than decorative.
It outputs at production spec. 720p, 1080p and 4K, a 21:9 ultrawide option, 9:16 and 1:1 for social, and two-channel stereo with lip sync that tracks. Prompts run to 8,000 tokens, so a real shot brief goes in whole instead of being compressed into a sentence.
When Director reaches for it
Director picks the model against the requirement, and Kling 4.0 is what it points at when:
- The shot is one continuous move and the look has to survive all of it
- You need the piece to land on exact frames, which is the start and end frame problem
- There are story beats to hit in order, which is what the 10 keyframes are for
- The deliverable is 4K, 21:9, or HDR
- You are recreating a shot you already have with a different product in it
If none of those apply, a cheaper model is not a compromise. That is the whole logic of letting Director choose.
What it is good for
Wardrobe, colourways and SKUs. One shoot, many variants. The expensive part of a fashion or product spot is doing the same move again for every version, and this is precisely the thing Kling 4.0 is built to avoid.
Campaigns that ship in four ratios. 9:16 for social, 1:1 for feed, 16:9 for the site, 21:9 if it runs before a film. Native rather than cropped.
Recreating a proven ad. Reference an existing video’s pacing, framing and camera work, then put your product in it. For an agency sitting on a format that already performed, this is the cheapest possible second campaign.
Anything with a spoken line. Stereo and lip sync that holds, in a long list of languages and accents, which matters the moment a campaign localises.
Enterprise, specifically. The combination that counts is the 30-second take plus references that hold: it is what lets a brand run the same character, the same product and the same look across a series without the drift that makes AI work obvious. Pair it with Elements for the subject, Location Scout for the place and a Document for the rules.
How we made the ski film
Watch what it does not do. The camera never cuts.
It is one unbroken run down the mountain, and the skier changes outfit four times while the run continues: the rust skirt and straw hat, the tan suit, the black lace, the fur. The snow, the light, the trees and the motion are identical throughout. Only the wardrobe moves.
That is the Kling 4.0 case in a single shot. Generate the run once, then use the run itself as the reference and swap the garment on each beat, keeping everything else fixed. The alternative, in the real world, is skiing the same line four times in four outfits and hoping the light holds.
Fifteen seconds, and the kind of thing an agency would normally budget a location day for.
Kling 4.0 or Seedance 2.5
They overlap more than they differ, and both will do a 30-second take.
| Kling 4.0 | Seedance 2.5 | |
|---|---|---|
| Single take | Up to 30s | Up to 30s |
| Resolution | Up to 4K | Up to 1080p |
| Keyframes | Up to 10 | Start and end |
| Ratios | 21:9, 16:9, 1:1, 9:16 | 16:9 and the usual set |
The part a table cannot settle is which one you should actually reach for, and here our own filmmakers are better evidence than a benchmark.
They have been working mainly in Seedance 2.5, and in Kling 3.0 Pro before that. Their view is that those two families are the ones that give you the cinematic texture, the grain and the surface, and that the alternatives are not close on it.
So Kling 4.0 arrives in the family that was already winning that argument, and it is roughly three times cheaper than Seedance 2.5 at matched resolution and length.
That is the headline. Two models our own team already rates above everything else for texture, and the newer one costs a third as much to run.
Published head-to-heads will tell you other things, and some of them contradict each other. Most come from platforms that sell one or both models, so weigh them accordingly. What is not in dispute is that both will hold a 30-second take, and that the requirement decides which one fits.
Which is the argument for not choosing by hand at all. Describe the job, let Director read the requirement, and look at what it picked.
The short version. Kling 4.0 is the one to reach for when a look has to survive a long continuous move, when the beats have to land in order, or when the deliverable is 4K or ultrawide. It follows a brief closely, it holds a 30-second take, and it costs about a third of what the model our filmmakers were already using costs. For a campaign that ships in several ratios and several variants, that is a straightforward win.
More on making video well in part one and the advanced guide.
Frequently asked
- What is new in Kling 4.0?
Generations up to 30 seconds in one take, up to 10 keyframes to pin the story beats, output to 4K, a 21:9 ultrawide option, two-channel stereo with lip sync, and references that hold together better. Prompts can run to 8,000 tokens, so a full shot brief goes in as written.
- When will Director choose Kling 4.0?
When the requirements point at it. A piece that has to hold a specific look across a continuous move, land on exact frames, deliver 4K or run 21:9 is Kling’s shape of job. Director reads the requirement before the price and tells you what it picked.
- How many references can Kling 4.0 take?
Up to 15 items in one generation: as many as 10 images, 5 videos totalling 30 seconds, and 7 elements, 3 of which can be video-based. A voice reference is also supported.
- Can it keep a product consistent across a whole ad?
That is the case it is built for. You can reference an existing video’s pacing, framing and camera work and swap the product or the character into it, rather than rebuilding the shot each time.
- Kling 4.0 or Seedance 2.5?
Both generate up to 30 seconds in one take, and both come from the two model families our filmmakers rate highest for cinematic texture. The practical difference is cost: Kling 4.0 runs at roughly a third of Seedance 2.5 at matched resolution and length. Kling also goes to 4K, takes up to 10 keyframes and offers 21:9.
- Does it do vertical?
Yes. 9:16, 1:1, 16:9, 21:9 and Auto, generated at that shape rather than cropped down from a landscape master. The ski film in this post was made vertical first, for social.
- What does a Kling 4.0 generation cost?
Video is priced per second and the models are far apart on it. Kling 4.0 lands at roughly a third of Seedance 2.5 for the same resolution and length, which is a large part of why it is worth reaching for. Director confirms the resolution and the exact cost with a quote before it runs the video step.
- Is 4K available now?
4K and 1080p are listed in the specification. Kling’s own release note describes the 10-bit HDR output at those resolutions as coming soon, so check what the panel offers before promising a client a deliverable.
The AI tool for cinematic video.
Generate, direct, and publish professional videos — powered by the best AI models.
More posts
