Micro-drama: The Platform Is Not the Product

💬 Micro Drama Platforms & Tech Evolution Quick Answers

What is a card puller in the AI video production industry?

A card puller (抽卡师) is a specialized technician who manages generative AI models for video production pipelines. Instead of coordinate planning with massive film sets, cameras, and physical actors, a card puller feeds scripts and visual reference assets into models like Seedance 2.0, generates vertical clips in massive algorithmic batches, and manually curates the structural fragments that meet quality benchmarks.

Why are micro drama platforms shifting from live-action to AI content?

The transition is driven entirely by extreme production cost-efficiencies and hyper-accelerated timelines. Traditional vertical short-form crews spend roughly three weeks shooting live-action sets, whereas a small team utilizing advanced generative AI applications can engineer a complete 60-episode digital drama in less than seven days at a fraction of the capital expense.

What is the primary difference between Hongguo and ReelShort’s revenue models?

ReelShort relies on external user acquisition ads, while Hongguo functions inside an integrated ecosystem. ReelShort monetization depends on direct viewer paywalls while renting its target traffic from external platforms at immense cost. Hongguo operates on a free ad-supported tier, retaining its entire subscriber circulation cycle within ByteDance’s native ad networks, shopping apps, and video distribution systems.

💬 The Deep Dive Article

For the past year I’ve been following China’s entertainment industry closely, and one of its fastest-growing sectors is micro-drama. It is purely fascinating to observe. You may know those under the names short or vertical dramas. What interests me is not how fast it’s growing, but how it works and I already have covered the segment when it emerged in previous articles.

The thing is, it is so fast-paced that my previous articles may only give you the background to understand where I am today.

Honestly, I held this one article back on purpose. I wanted to see how AI would change production, advertising and growth once it made real scale possible at the start of 2026. The changes turned out to be more intriguing than I expected, and they rearranged the industry in less than a year. The shifts exposed one interesting point though. Speed is a side-effect of technology. The real thing is where the money went when production stopped being the bottleneck.

I’ll start with the numbers, because they set up everything else. DataEye’s figures on overseas short-drama apps: in the first half of 2025, these apps earned about $1.08 billion from 665 million downloads. A year later, downloads had more than doubled, but revenue reached only around $1.27 billion. Make the calculation and you’ll see that a download was worth $1.62 in 2025 and just $0.88 in early 2026. Sensor Tower’s figures, measured differently, point the same way: by my calculation, first-quarter downloads rose more than 130% year on year, revenue only about 7%.

Vivek Couto of Media Partners Asia summed it up: micro-drama has become “an earnings story rather than a growth story.” An earnings story is about who brings in the viewer, who keeps what they spend, and who absorbs the loss when a title fails. This piece is organized around those three questions, in the order the industry answered them. What I am also sure of is that we haven’t seen it all, so there will definitely be more to come in terms of changes.

AI Automation: What Got Cheap First in the Content Supply Chain

What I love about China’s business environment is what I call its “cowboy approach” to innovation: you don’t wait for perfection, you deploy, let it play out and fix things as you go. It’s why structural change and its flaws show up there first. The trigger here was ByteDance’s Seedance 2.0 video model, released in February 2026. Within weeks, live-action work in Hengdian collapsed (Caixin).

The production line changed its shape and where a crew of 40 or more once spent three weeks on preparation, shooting and post-production, a team of five with a few computers can now make a 60-episode drama in about a week. Cameras, sets and actors gave way to scripts, prompts and generation. A new job appeared: the “card puller” (抽卡师), who feeds the model a script and visual references, generates clips in batches, and keeps the ones that work.

Cheaper, faster production flooded the market. By the first quarter of 2026, around 95% of new micro-dramas in China were AI-generated, and a premium live-action budget of about 1.5 million yuan now had an AI alternative at under 200,000.

It wasn’t surprising that the flood consumed itself. Only around one in a hundred new AI dramas on Douyin earned back its costs, and earnings per view kept falling because when everyone can make a drama, a drama’s value changes dramatically. That answers the first question almost by default: when production got cheap, nobody made money from production anymore.

Live action, meanwhile, found its level. It became a small share of releases but drew a far larger share of views, and by the second quarter output was rising again, backed by an NRTA programme launched in April, under which six major platforms, Douyin and its Hongguo app among them, committed in May no less than 6 billion yuan to live-action production. Early reports suggest the money has not yet reversed the downturn, but the direction is clear. Human production didn’t get eliminated but rather got an overhaul of its value and pricing.

The Distribution Pivot: Why the Production Bottleneck Moved to the Feed

And with the bottleneck moved, the real question moves too. If a drama costs almost nothing to make, what does it cost to succeed? Industry estimates put paid promotion, the traffic that brings viewers to a title, at around 70% of total cost. Even experienced teams earn only a few percent above what they spend.

Note the asymmetry. Production changed in months. The cost of acquiring a viewer’s attention is falling only gradually, over years. Analysts tracking ReelShort expect its acquisition spending to fall on exactly that slow curve, from 55% of revenue to 44% by 2028. Considering the speed at which production changed, I wonder if 2028 is not way too far to predict at this stage.

Now the bottleneck has moved, from the set to the feed, from cameras to distribution, and honestly this is how progress operates. Production speed solved, exposes other weak points of the whole supply chain. Once this is solved, I am sure it’ll move further. Whoever controls the bottleneck at a particular moment controls the industry’s economics. Everything that follows further down in this piece is a variation on that sentence.

Ecosystem Convergence: Tracking Who Owns the User Spend

Hongguo and ReelShort are the clearest examples of the two models, and they run opposite answers. How a drama is made (live action or AI) and how it earns (free or paid) are separate questions; Hongguo and ReelShort answer both questions, the earning and the ownership, completely differently.

Hongguo is free and it earns through advertising and links to Douyin’s e-commerce. It has well over 300 million monthly users spending about two hours a day in the app, and doesn’t publish its finances. ReelShort is paywalled, earns mostly from viewer payments, passed three-quarters of a billion dollars in revenue in 2025, with most of it coming from North America.

On a rough per-user basis, my own estimate from reported figures, the two land in a surprisingly similar range (roughly $8 per Hongguo user in 2025, based on reported revenue, against around $12 per ReelShort user; not a like-for-like average, but enough to show the gap is smaller than the two models suggest). The real difference comes from how each user arrives. ReelShort rents its audience from platforms it doesn’t own: most of its viewers come through paid advertising on someone else’s platform, and in 2025 that advertising cost ReelShort 55% of its revenue. Hongguo sits inside ByteDance, which also owns the video platform where its clips circulate, the ad system and the AI model (Caixin). For Hongguo, that spend largely stays within the group.

This is why the two models are converging, but the convergence is easily misread. ReelShort is adding ads and its own payment channels; Hongguo is adding shopping to advertising. Free and paid are becoming hybrids, yet hybridization matters less than control. It is not the charging method that determines a platform’s economics, but how much of the chain, from drama to money, it owns. ReelShort charges its viewers but rents their attention. Hongguo gives the drama away and keeps most of what surrounds it.

One caveat, and it’s the opening left for everyone else: overseas, Chinese apps take the great majority of short-drama app revenue, but locally made dramas are quickly replacing translated Chinese titles in foreign markets. The advantage still within reach of non-Chinese players is exactly that localization. Their own audience in their own language and the stories made to match the audience’s attention.

Risk Mitigation: Dissecting Who Carries the Project Loss

The third question is who absorbs the loss when a title fails, and this is where the industry’s answer shifted hardest.

Hongguo once guaranteed producers a fixed payment per title. After some suppliers exploited the system with ultra-cheap dramas, those guarantees were withdrawn from most producers, then redirected in 2026 to selected live-action projects. Producers outside that tier now carry the full risk that a title finds no audience and most projects reportedly lose money.

Actors were the ones that felt the impact of the production shifts the fastest. State media reported pay for lead actors halved and mid-tier pay cut by more than three-quarters. You’ve probably already seen stories circulating about Hengdian actors now working at theme parks. A new market appeared for faces: an actress was offered 500 yuan to license her likeness for a year of AI dramas. Her words strike me as the sharp reality which is not felt by everyone yet: “We used to get paid by the hour,” she said. “Now we get paid by the pixel.”

In the US, by contrast, the vertical-drama boom came with a union contract setting minimum pay. Two different labor markets that have two different approaches to the same disruption.

The thing is, human work isn’t disappearing from micro-drama. It’s being sorted into a small premium tier at the top where the live-action projects platforms now underwrite selectively, and a per-pixel tier at the bottom.

The route is the product

In the end I decided to put the three answers together. Who brings in the viewer? The traffic machine, whoever owns it. Who keeps what viewers spend? Mostly the platform and increasingly, either platform. Who absorbs the loss? The producer, the actor, the person paid by the pixel.

When production got fast and cheap the value moved to the distribution layer. The earning models, despite being different, converged on ownership. The human layer got sorted into either premium or pixel. Growth is a side measure, not the real criteria for success.

If an entire industry can be restructured this fast: cameras replaced by prompts in weeks, guarantees to full producer risk within a year, then for me micro-drama may be one of the clearest previews we have of how AI disruption will unfold elsewhere. Technology doesn’t just make something cheaper. It decides who gets paid, and the answer is: whoever owns the route to the audience.

The platform is not the product. The route is. And the route is still moving.