Kuaishou’s Strategy: Riding the Rise of Short Mini Dramas + Industry Expertise Over AGI Chase
Kling: New Subcategory of AI-Native Entertainment
Hi everyone,
Hong Kong is going through a heartbreaking tragedy right now. My heart goes out to the families affected by the deadly fire.
It’s been deeply sickening to watch parts of X turn it into fuel for political narratives, when the only thing that should matter at this moment is getting survivors to safety and supporting the families who’ve lost so much.
My family and I are doing what we can, gathering winter clothing, basic necessities, and direct donations to local charities and NGOs. I’m acutely aware of how small that is compared to the scale of what’s needed, but if you’re in a position to help, I’ll share a list of organizations at the end of this piece.
And a day after Thanksgiving, I just want to say thank you for supporting AI Proem, and remember - there is always room for more kindness in this world.
Ok, now, let’s talk AI.
I met with a representative of Kling, Kuaishou’s AI team, this week. The meeting was meant to be a fairly “normal” catch-up and check-in on models, product roadmap, and use cases. But it ended up becoming something else: a window into how AI is diffusing into creative fields, and how differently investors fixated on value creation, builders focused on technological advancement, and creatives who hold the utmost respect for craftsmanship experience and evaluate AI.
Are we creating AI for the current generation or the next?
What stayed with me from that conversation wasn’t the tech demo. Kling’s models are objectively impressive: cinematic motion, controllable camera paths, and increasingly photoreal characters. Apparently, the “cat skating on ice” video was a huge deal in the industry because of how sophisticated the movement and physics were.
But the thing that really lingered was something the Kling representative said almost in passing: their team had been genuinely surprised by the kind of feedback they were getting from people in the creative fields.
When directors, actors, and editors start using Kling, their biggest concerns are not about resolution or latency. They ask: Am I accidentally infringing on someone’s IP? Does this cheapen what I’ve spent my life learning to do? Who actually owns the shot if AI generated half the frames? What about the martial art artists, the chefs, and the musicians we hire for those expert shots? Copyright, authorship, and respect for craft remain top of mind for them.
In tech, people are often so fixated on building the best system that they can become oblivious to the potential harm or disruption that system creates. It’s not that the developers are evil; it’s just that they’re optimising for a different objective. Investors focus on monetary value creation. Policy analysts focus on safety and guardrails. Developers focus on “进步” — advancement — and, as she put it, that can become the only thing they see. They don’t necessarily know what the real pain points of end users are. That’s why Kling has tried to flip the sequence: let users lead the product, and let the product guide the technology.
She shared another anecdote that I found fascinating. At a film school event, a well-known director and professor talked about how he always has to think about the physical space he can work with when staging a scene and planning actors’ movements.
A current student then challenged that assumption: why should we constrain ourselves to physical limits at all if we’re working with AI? Why not design the world exactly as we want? He went further and said that using AI means a director has to give up some control. The model will develop its own interpretation of the prompt, and it won’t always give you the precise image in your head. If you prompt “a sofa in a room,” you can’t expect it to perfectly match your imagination. In a way, making someone literally make that sofa in real life is easier, but that is also much more costly. Working with AI, in other words, means treating it as a creative partner rather than a tool completely in your control - like Photoshop?
That same student gave a simple example: when you want a cat to move in the frame, why must it always walk from left to right? What if the AI decides the cat should walk from right to left? He seemed far more relaxed about that kind of flexibility than many older generations in the industry. At the same time, anyone who has seriously used text-to-image or text-to-video tools knows that it’s incredibly hard to surgically tweak small details after the image or video is generated. So the way you prompt, including the language, the know-how of the industry, the context, and all, becomes highly technical. So it seems like prompting itself is turning into a craft.
All of this made me wonder if we are thinking about AI integration in the wrong way or in a narrow-minded way. We spend so much time discussing how AI will increase productivity within the workflows we already recognise. But what if AI-native generations reshape the workflows themselves? Thirty years ago, cartoonists drew with pencils on paper and spent hours sometimes to mix the perfect color. Today, they have digital tablets, animation suites, audio tools, and colour grading, so you can see that the entire work process is probably unrecognisable to their predecessors. The craft of creating cartoons has essentially changed completely.
So now the question is, do we evolve the craft or evolve the result? How do we not lose that craftsmanship, or at least continue to respect the aesthetics and skills that took decades to hone? The friction that is creating “slop” is actually because the creators of the tools do not understand the needs of the users.
Very focused, very clear target market
So that brings me to this point. Amid so many AI startups and big tech I’ve met, many have been pivoting their strategy from A to B within just months (or weeks trying to capture the gold rush). But what makes Kling interesting to me is that they’ve actually been quite committed to their vertical since the beginning. And they’re quietly inverting the usual AI hierarchy.
Expanding on what I mentioned above, take a step back and think it through.
The standard playbook is: build a powerful model, ship a flashy demo, and only then scramble to figure out which industries want it. Most big tech companies have been building general-purpose tools and startups and finding ways to “wrap” around them for vertical use cases.
With Kling, they’ve only worked with videos essentially, given that its parent company, Kuaishou, is China’s second popular short-video platform.
The Ghibli Hype: Next Frontier is in Video & Introducing China’s Leading AI Video App, Kling
Last week, design software maker Figma submitted paperwork to the U.S. Securities and Exchange Commission for an initial public offering. This comes just over a year after it uncoupled from Adobe, following a public courtship of $20 billion, largely due to concerns about monopoly and regulatory controls.
They have also been very specific with their user demographic: film directors, marketers, post-production teams, and AI-native drama studios. And you can see that in the business mix: roughly 70% of revenue comes from prosumers, think professional creators and marketers who are close enough to the end audience to care deeply about craft, but pragmatic enough to pay for speed and scale. The remaining 30% comes from enterprise API customers across advertising, animation, gaming, and film.
So what they’re doing to ensure satisfaction is letting the users lead the product, which then reverts to the tech. The team first adjusts the product surface to make it easier to mark up shots, control motion, and line up start and end frames. Then the model itself is nudged in response: better multi-view consistency, more controllable motion, finer editing handles. You can see that evolution in the progression from the first release to things like the Master Edition 2.0 and now 2.5 Turbo. The model is no longer the hero on stage; it’s the part backstage that quietly updates to keep up with what the outside world is asking for.
They’re also hiring industry people, who act like translators between the film industry and the tech people, and partnering up with film festivals and associations across the world. What we’re seeing is industry → workflow → product → model, this is actually quite a radical rewiring.
Looking at the business numbers. In under a year from launch in June 2024, Kling had already crossed an annualised revenue run-rate of over US$100m by March 2025, according to the material Kling shared with me. Quarterly revenue stepped up from around RMB 150m in Q1 to over RMB 250m in Q2.
Its deliberate positioning as an AI creative platform for the professionals actually is a pretty overlooked vertical.
When Hollywood is protesting against the use of AI, there are people who will embrace it, notably in markets where there is less money for creatives. And it made me think of when TikTok took off, a whole sub-industry started from zero- MCN agencies that managed influencers. In some ways, I think we’ll see a subcategory of AI native content take off. As for deepfake concerns, she noted that in China, there is a strict requirement that all AI content be watermarked, and technically, if you remove that watermark, you’re breaking the law and can face legal repercussions. But the thing is, this requirement varies across markets. [I’m a bit scared by the lack of an international standard on this tbh]
And while I thought ByteDance would be its biggest competitor, she actually said it’s only ByteDance’s Seedance that is doing something similar. As we’ve been writing about at AI Proem, ByteDance seems to be doing everything and is probably the most aggressive of the Chinese big tech in talent poaching right now. It’s doing everything from LLMs, to consumer apps, to video editing, and well- this.
To note, one of the four tigers, Minimax, has also been shipping out multi-modal models and is said to be very competitive. But as it has been on an investor courtship, two investors have shared with me that they feel the company is a bit inconsistent in its strategy right now. It seems distracted with high-profile apps like Hailuo and Talkie, but it is also still trying to compete on frontier models. However, given its grassroots beginning, it is slowly running out of money. It doesn’t seem to have strategically chosen a lane to focus on yet.
And in this case, Kling, by comparison, looks like it’s the least ambitious among the three, yet the most focused.
New subcategory: AI-native entertainment
One of the most unexpected and interesting wedges for Kling has been vertical mini-dramas. One use case of AI-generated content that is taking off quietly is the potential of creating 1080p, smartphone–OK-quality videos. Think of the telenovela-like short mini-dramas that are taking off. Many of the producers and studios behind them are actually Chinese.
China’s 短剧 has actually taken a very unexpected life on its own. Zeyi Yang actually wrote about it for Wired in July. These short videos are pretty much like what he wrote: acting quite meh, plot somewhat the same, actors tbh usually some kind of a D-lister, not to be mean (it is what it is). You know what drives their popularity? – our universal appetite for romance, melodrama, storytelling.
I mean, who doesn’t love a good romance? I, for one, miss the 90s–00s romcoms and am too braindead at the end of the day to watch high-brow, deep-thinking, soul-searching sci-fis that seem to dominate Netflix / Apple these days (I do appreciate Severance, but it requires too many brain cells to follow tbh).
And everything seems to have a satirical take these days, but sometimes I just want The Summer I Turned Pretty, which I shamelessly binged religiously.
My husband made fun of me for coming home from work to find me glued to the screen, watching teenagers half my age fall in love and flirt with their eyes. (Team Conrad, thank you. Don’t judge me, talk to my hand.)
Anyway, you know how AI creation can be plugged into this?
Wellllll, it’s not like anyone’s zooming in on their phones to see the details of actors’ expressions or if the fingers match the melody played on the piano. So Kling is plugging into this space and co-creating AI-assisted short films with directors like Jia Zhangke and Timmy Yip, and these are explicitly framed as human–machine co-creation rather than full replacement.
So, to end my rant today. I think the conversation with Kling serves as a useful lens on where AI might actually be heading - beyond the industries as we know them. It’s not the first time traditional media/ content/ entertainment has been disrupted.
[Ollie Forsyth actually put together a guide on “The New Titans of Media” here, and humble me is included too. SanQ.]
What we’re seeing is that on one hand you’ll always have the ‘AGI’ chasers, blinders on, optimising for benchmarks and rankings but in another case of “good enough” combined with deep tacit knowledge on the other hand, AI native videos will have a potential place in this world. I’m not sure how much I would like them, but I see it as a potential new subcategory and new business model.
For those who want are interested. Here’s a list of organizations you can reference to help the families affected by the tragic fire in Tai Po, Hong Kong. 🙏
This is one of those article that got my immediate like on the title before going into the details 👍 obviously well research