Domestic platforms give creators a 6–7% revenue share, with a monthly quality-rate review that suspends cooperation if standards aren’t met. ByteDance’s漫剧 collaboration document marks two overseas tracks—“overseas production” and “overseas scripts”—as “paused.” Meanwhile, the overseas short-drama market is projected to hit $60 billion in 2026. On new platforms like YourChannel, a Chinese creator pulled in $586,700 with a 5-episode short drama in 72 hours, with the platform’s revenue share going up to 90%. This article does the math: where the money is, and what the real barrier in front of that money actually is.

Let’s Start With That 5-Episode Drama

The Persian Revenge is 5 episodes in total—first 2 free, latter 3 paid, with a total runtime of 10 to 20 minutes. The story follows a Middle Eastern assassin who time-travels to the present day to assassinate a tyrant.

One solo creator. Within 72 hours of launch, the backend showed total revenue of $586,700, nearly 50,000 paying users, and a daily average close to $200,000. It wasn’t on TikTok, YouTube, ReelShort, or DramaBox—it was on a platform called YourChannel.

This news was reported by the Associated Press and other foreign media, and relayed domestically by “短剧自习室.”

Let’s pour some cold water on the numbers first: this is a platform-backend figure, a single data point, and highly non-replicable. But its value isn’t in the size of the number—it’s in how it lays bare one question. The same drama, on different platforms, can mean the creator takes home more than ten times as much.

The Domestic Path: The Terms Are Face-Up

ByteDance’s Short Drama Copyright Platform—Finished 漫剧 Business Introduction and Collaboration Guidelines lays out the domestic cooperation model in two tables. Here’s the key difference in one sentence:

Model 1, “Finished Product Standard Revenue Sharing”: 5-year contract, 5-year+ content license, standard authorization. Revenue rule: free segments use standard per-unit pricing; paid segments take 6% of the paid mini-program recharge amount.

Model 2, “Finished Product Framework Cooperation”: 1-year contract, perpetual content license, special authorization. Paid segments take 7% of the recharge amount.

The entry thresholds differ. Model 1’s requirements are straightforward: ability to produce complete finished content, clear ownership with no infringement, horizontal or vertical format, generally open to all 漫剧. Model 2’s threshold is “meets our data evaluation criteria”—after platform confirmation, a business team proactively reaches out. Put bluntly: you need to be noticed first.

Model 2 also lists several hard metrics in the document: content types are limited to simulated live-action, 2D animation, 3D animation, or dynamic narration 漫剧; minimum 60 minutes per AI drama; series and serialized works supported.

The most critical part is the evaluation mechanism. The document states: after signing, continuous assessment is conducted; if the quality rate falls below standard during the cooperation period, the partnership will be suspended; if the quality rate meets the standard again, cooperation will resume. The evaluation cycle is once per month.

60 minutes, monthly reviews, quality rate达标—put these three together, and you have the complete picture of what domestic platforms expect from an individual creator right now.

The Overseas Doorway: The Document Says “Paused”

Here’s where it gets interesting. Chapter 2 of the document is titled “漫剧 Overseas Cooperation Guidelines (Paused).” Two items hang beneath it:

1. Overseas production cooperation: Overseas AI Simulated Live-Action Drama—Production Cooperation Plan (Paused)
2. Overseas script cooperation: 番茄 Overseas—Latin America AI Simulated Live-Action Dialogue Script Solicitation (Paused)

The document’s last modified date is September 24—yesterday.

So the easiest path—“work for a platform and go overseas through them”—is currently closed. That’s not my inference; it’s what the document literally says.

But the Overseas Market Is Growing

At the same time, the overseas short-drama market data points the other direction. The 2026 overseas short-drama market is projected at $60 billion; AI short-drama export revenue reached $1.27 billion in the first half of the year. Reports suggest AI dramas already account for 40% of TikTok’s short-drama consumption.

The platform side is also diverging. ReelShort and DramaBox are two reference points: the former had roughly $404 million in revenue in 2024, pursuing a premium high-margin strategy but facing retention pressure; the latter brought in about $323 million, runs on a subscription model, has lower user churn, and has already turned a profit. The two operate completely separate account and payment systems—two independent businesses.

And beyond them, a third category of platform has emerged. YourChannel positions itself as a “one-person content monetization platform,” with a core model of paid viewing plus ultra-high revenue share plus creator-controlled distribution. Its claimed revenue share goes up to 90%, with daily withdrawals, and it promises no monopoly on traffic or compute—creators retain copyright, pricing power, and ownership of their private-domain traffic.

It also offers a low-cost testing mechanism: creators only need to upload 3 episodes to get market feedback data, which they can use to adjust their subsequent content direction.

If you place the domestic 6% and the overseas 90% side by side, a brutal conclusion emerges: going overseas isn’t “working harder for the same money”—it’s the same labor priced more than ten times differently under different rules.

How Far Can You Free-Ride on Google’s Models

Now let’s talk tools. The easiest thing to overestimate on this path is “free.”

The text side genuinely has free tiers. Gemini’s free tier is still available—the Flash series allows roughly 1,500 requests/day, and Flash-Lite about 15 requests/minute. Writing scripts, breaking down storyboards, generating character profiles, translation and polishing—all of this runs fine, at zero cost.

The video side does not. Veo is not on the free tier. To use it for generating finished content, you must enter a paid tier. So the phrase “free-ride Google’s models to make English short dramas” should be stated precisely: you can free-ride the scripts and storyboards, but you cannot free-ride the finished films.

One more reminder: free-tier quotas are for humans interacting with the model, not for running scripts to batch-process. Industrial-scale usage like 2,000 uploads per day won’t be covered by free quotas.

Finished content is where you have to switch lanes. Three options:

If you go the “pure free-ride” route, you get a complete script and storyboard package—plus a finished film you can’t generate. If you go the minimum-paid route, a personal monthly budget runs three to four figures, depending on your output volume and quality requirements.

Tools are the one part of this chain where there’s no shortage right now.

Tools Solve Throughput, Not “Quality Rate”

This is the part I most want to make clear.

Did you notice a contradiction buried in the data above? On one side, the market is growing and platforms are competing for content. On the other, domestic platforms are taking down content. Before the end of April 2026, 红果 had cumulatively taken down or blocked over 10,000 low-quality AI short dramas, with 漫剧 accounting for more than 60%. Meanwhile, on just 红果 alone, the daily new upload volume of 漫剧 in March reached roughly 2,000—20 times that of live-action dramas.

Massive uploads and batch takedowns happening simultaneously means platforms have moved past the “content-starved” phase into the “content-selecting” phase.

Platform policy trajectories confirm this: from guaranteed minimums to no minimums, from fixed splits to pure revenue sharing, from “buying content” to “filtering content.” When supply surges to 2,000 uploads per day, the platform no longer pays for “your ability to produce content”—it pays only for “your content passes review and someone finishes watching it.”

And as for how much AI can actually help with “passing review and getting watched”—that’s worth breaking down item by item.

Novel—the story itself. AI can give you a structurally complete template, but it can’t give you “why this story must be told.” Templated stories show up in the data as low opening retention.

Writer’s sensibility—the rhythm and restraint of prose. AI-generated dialogue tends to be overfull, lacking the judgment of “leave one line unsaid.”

Screenwriter’s instinct—beats, hooks, and cliffhanger endings per episode. AI can apply structure, but the hooks it produces tend to be “information suspense” rather than “emotional suspense”—and the latter is what actually retains viewers.

Director’s thinking—why this shot, this angle, this pacing for this particular scene. AI gives you “average correctness,” not “the best choice for this scene.”

Cinematography—the emotional expression of lighting, composition, and color. AI’s optimal solution is “not bad-looking”—but a hit needs “you remember this one shot.”

Art direction / aesthetic judgment—determining the unified visual tone of an entire drama. This is something AI currently cannot provide at all, because it requires looking at a pile of all-correct options and insisting on one: “This is the one I want.”

These six roles are six separate positions in a traditional production crew. On an individual creator, they’re six loads to carry. What domestic platforms evaluate monthly as “quality rate” and what overseas platforms’ testing mechanisms collect as “market feedback”—they’re screening for the exact same thing: these six abilities.

In The Persian Revenge’s case, the creator was one person, but the story’s skeleton, the assassin time-travel hook, and the pacing design of 5 episodes with the first 2 free—all required human judgment. AI may not have been involved in the script at all at that point.

Risks and Boundaries

Three things must be stated clearly, or this article is misleading you.

First, third-party commercial terms are cited only as reference. ByteDance’s document is an internal platform guideline for partners. I’ve cited publicly stated portions as supporting evidence, without reproducing the complete tables. Terms change. The document was last modified on September 24; tomorrow it could be a different version.

Second, the data itself has discrepancies. 红果’s policies on AI short-drama guarantees and revenue sharing have very inconsistent media reporting: some say guaranteed minimums were fully eliminated and converted to pure revenue sharing at roughly 20%; some say the revenue-share coefficient dropped from 60 to 40; and the platform itself responded that “adjustments were made only for certain high-risk scripts.” I’m inclined to view this as口径 discrepancies arising from rapid policy iteration. Do not treat any single number as definitive.

Third, “paused” does not mean permanently closed. Those two overseas cooperation tracks in the document are marked as paused, but the supply-demand relationship between the overseas market and Chinese content production capacity hasn’t changed. It’s more likely a temporary state during a policy adjustment, not a strategic abandonment.

Let’s Do the Final Math

Per the document’s requirements, a single AI drama must be at least 60 minutes. At the standard short-drama length of 1 to 2 minutes per episode, that’s roughly 30 to 60 episodes. Assuming each episode needs 15 to 25 effective shots, a single drama requires between 450 and 1,500 shots.

Using free tiers for scripts and storyboards brings this segment’s cost close to zero. But 450 to 1,500 shots—each one requires a human to judge: “Is this angle right? Should this line stay? Will this image be remembered?”

Tools have torn down the wall on the throughput side. The wall behind throughput is still built from those six abilities.