AI videos are moving from short clips (using simple text and /or images) to highly controllable, long-form video production.
And Kling 4.0, which is launching this month, is a big step forward on that development path.
The Main Tech Challenges of Video Generation Today
Making one good frame or short clip is now a solved problem.
But making thousands of frames that hold together (and do not cost a fortune) is difficult.
Here are some of the main tech challenges.
- Consistency over time. This is the core problem everyone talks about. A video has to keep the same characters, objects and scenes from the first frame to the last. And small errors in each frame can add up, so longer clips drift. Faces change. Objects appear and vanish. This is why clip length has been measured in seconds.
- Compute cost. Longer and sharper video takes far more compute, both to train the model and to run it. Cost also sets serious limits on length, resolution and speed. It is also why you still submit a request and wait for quite a while.
- A text prompt is still a blunt instrument. Professionals want to set the shot, the camera move, the timing and the performance. Video generation is moving away from one long prompt and towards reference images, keyframes and specific edits.
- These models learn 2D visual patterns that look like physics. Objects usually fall and liquids usually flow. But the models do not know or actually enforce physical laws, so unusual motion can break. They just infer physics from the prediction of 2D images.
- Sound and performance. Professional videos need dialogue, lip-sync and audio that are generated in step with the picture. This has been problematic.
Solve these and you can move from clips to Hollywood-type stories.
And further out, you might get to world models, which are interactive environments generated in real time. But that really is a different generation process entirely.
Given that, here’s what new in Kling 4.0.
Kling 4.0: Five Big Upgrades
Kling says the full Kling 4.0 will officially launch this month. And Kling 4.0 Flash (a lighter, faster version) went to a limited group for early access on September 28.
Keep in mind, Kling’s core customers are (according to Kling) focused on filmmaking, short-form drama, animation, advertising and music videos. That means lots of product ads, user-style product promotion, game trailers and short dramas.
So, these are the professional (and paying) customers, which I mentioned in the first article.
Here is what Kling has announced for 4.0.
Upgrade 1: 30-Second Scenes in One Pass
Kling 4.0 will generate up to 30 seconds in a single pass. That’s double the 15-second limit of Kling 3.0. And video extensions up to 2 minutes is coming soon. That’s the upgrade that people will notice the most.
Kling also says 4.0 will include fast motion and complex camera moves. Sot that’s push-ins, pans, tracking and orbit shots, staying smooth and stable, and so on.
On-screen text is also new. The model can generate text in multiple languages, plus emojis, and Kling says text and logos stay clear through movement and scene changes. Advertisers will really like this.
Screenshot from Kling 4.0
Screenshot from Kling 4.0
Upgrade 2: From Prompting to Directing
Today, content creators are making longer videos by stringing together lots of pretty short clips (10-15 seconds). But the big limitation is that goes hand-in-hand with longer videos is the ability to edit and craft those videos. To get away from one shot prompts where you hope you get what you want. And then try again if you don’t.
The key is this is advanced editing capabilities. For this, Kling 4.0 has:
Up to 10 keyframe images.
These temporal markers set character poses, scene changes and story beats. First and last frames are also supported. This lets you craft a sequence of events across the video.
Up to 15 reference items go in one generation.
Reference items let you set characters, scenes, motion sequences and objects. This is big for stopping character drift. For 4.0, it can handle up to 10 images, 5 videos (with a combined length of 30 seconds), and 7 subjects (of which 3 can be video based).
A voice is the only audio reference. Note: ByteDance announced up to 50 for Seedance 2.5, so Kling still trails on raw count. But this is a big step forward.
- Image references can be character sheets, grids, storyboards and wireframes.
- Video references can guide performance, camera movement and pacing. White-model and depth-map videos are supported, which lets a studio block out a scene in 3D first and have the model follow it.
Kling says references now work together more reliably, keeping characters, actions and shots closer to what was intended.
Editing up to five clips, 30 seconds combined.
You can replace a character or a scene, remove an element, or change the style while the rest stays stable.
A feature Kling calls creative recreation.
This lets you copy the pacing and shots of an existing video while swapping in your own products and characters. That one is aimed squarely at advertisers.
Screenshot from Kling 4.0
Screenshot from Kling 4.0
Upgrade 3: Stereo Sound and Multiple Languages
- Audio for 4.0 is two-channel stereo, with more accurate lip-sync.
- Dialogue covers Chinese, English, Japanese, Korean, Spanish and Portuguese. German, French and Hindi are added in this release. It also handles a range of dialects and accents.
Upgrade 4: Cinema-Grade Output
- Resolution is 720p, 1080p or 4K. That’s important. 4K is now table stakes for the top tier.
- 10-bit HDR at 4K and 1080p is coming soon.
- A 21:9 ultrawide format is new, alongside 16:9, 1:1 and 9:16.
Screenshot from Kling 4.0
Upgrade 5: Agents Enter the Workflow (The Big One)
This is my area and what I have been waiting for.
All the improving quality, length and editing capabilities are great. But if they are tools an agent can use (instead of a human), this becomes way more powerful. Why not just have an agent that is a highly trained director?
For Kling 4.0, this includes:
- A creation page that has been redesigned around one input box for all references, grid and list views, and a single timeline for previewing and editing.
- A new canvas that lets you connect assets, steps and results as nodes in one view.
- An Agent can take your request, use tools and carry out tasks. Kling already opened itself to outside agents via MCP and CLI. Now it is adding its own built-in Agent.
I really like this.
Also, Kling 4.0 Flash: The Faster, Cheaper Option
As mentioned, Kling 4.0 Flash is the fast, lower-cost version, built for frequent creation. It generates 3 to 20 seconds at 720p in standard dynamic range, and it still accepts references.
My Take: The Leaders Are Pulling Away
You can see Kling 4.0 moving along the video generation development path. I expect the leaders (Kling, Seedance, MiniMax’s Hailuo, Wan/HappyHorse, Google’s Veo) to continue to put distance between themselves and the second tier.
The harder problems, such as physics, real-time generation and world models, are questionable. It’s not obvious that you need physics encoded. Or world models at all. This is the first thing I would ask them about.
Cheers, Jeff
———
Related articles:
- AutoGPT and Other Tech I Am Super Excited About (Tech Strategy – Podcast 162)
- AutoGPT: The Rise of Digital Agents and Non-Human Platforms & Business Models (Tech Strategy – Podcast 163)
From the Concept Library, concepts for this article are:
- Videos
- AI Generative Video
From the Company Library, companies for this article are:
- Kling AI
———-
I am a founder, consultant and keynote speaker specialized in digital and AI agentic strategy.
I am the founder of Agentic Org Now, a firm that helps leadership teams launch AI and agentic initiatives fast.
I write (a lot) about digital and digital AI strategy (3 best selling books, +2.9M followers on LinkedIn). You can get a free book at the email sign-up below.
My Moats and Marathons book series is a framework for building and measuring competitive advantages in digital and AI businesses.
Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.
