Author: Changan | Biteye Content Team
Shortly after the 2026 Spring Festival, actors in Hengdian began posting videos on Douyin (TikTok) complaining about not getting any work.
"The crew's group chat is silent. In previous years, announcements would come in before the Lantern Festival (the 15th day of the first lunar month), but this year, we've waited until the end of February and haven't received a single one."
During the Spring Festival of that year, ByteDance's video generation model Seedance 2.0 was quietly launched, sweeping across the entire short drama industry.
This article aims to clarify three things:
What happened in the industry after Seedance 2.0?
How exactly are AI short dramas made?
And what opportunities does this mean for ordinary people?
I. One model changed the entire industry.
During the Spring Festival, ByteDance's SeeDance 2.0 video generation model was officially launched. Tim from Filmstorm said "terrifying" six times in a test video.
It has reshaped the entire video industry from the production side: you don't need a camera crew, actors, or a location; a text description, a reference image, and you can generate a publishable video in minutes.
With the threshold lowered, two types of needs that were difficult to meet in the past have been released.
Turn impossible scenes into videos: Create derivative works based on existing films and television shows, such as "Have you ever saved a fox at the foot of the Tianshan Mountains?"
Scenes we want to see but can't see: These are basic emotional needs, and some scenes may never have a chance to be captured again, but AI gives these scenes a chance to exist.
These two examples, taken together, illustrate the same point: the emergence of AI video generation tools has changed the way video is used as a medium. It is no longer the exclusive product of professional teams and equipment, but has become something that every ordinary person can use to express themselves, convey emotions, or even simply for entertainment.
This capability led to an explosion of two types of video content.
1️⃣ Short video content centered on entertainment and traffic
This type of content is not as complex as short dramas. It does not require considering the consistency of characters across multiple videos, nor does it require maintaining a continuous storyline. Essentially, it strips out some tedious and repetitive tasks and lets AI do them.
The most typical example is AI-powered digital voiceovers. The process is very simple: upload a photo of yourself to generate a digital avatar, write the text, and the AI will automatically complete the lip-syncing, voice-over, and video output.
Another type is the visualization of jokes. Many text-based jokes circulating online are funny but lack visuals, limiting their reach. Now, some people specialize in turning these jokes into videos, adding subtitles and voiceovers, transforming a text-based joke into a short video.
2️⃣ AI short dramas with storyline at their core
Short dramas are far more complex than short videos because the plot is continuous. The same character must appear from the first episode to the sixtieth, with their face, clothing, and scene style remaining consistent. This requirement for consistency increases the workflow's difficulty by an order of magnitude.
Due to ByteDance's restrictions on real-person images, many creators have turned to areas that do not require real faces—comics and animation.
By using AI-generated anime characters instead of real people, animated series circumvent compliance issues and simultaneously open another door: adapting online novel IPs. Fantasy, underdog stories, and system-based narratives—these genres, with hundreds of millions of readers on platforms like Tomato and Qidian—are naturally suited for making animated short dramas.
II. From Script to Finished Product: The Complete Workflow of an AI Short Drama
Many people see a video and think that the model is automatically created by inputting a plot description.
That's not actually the case. A high-quality AI short drama is backed by a well-defined workflow, with corresponding tools for each step, and the quality of each step directly affects the final product.
Step 1: Write the storyboard
The script requires you to clearly write out every shot. The format is something like this: Shot 3, kitchen, close-up, the male protagonist takes ingredients out of the refrigerator, the camera moves from his hand to his face, his expression is tired, duration 5 seconds, voiceover "It's that time of year again".
The more detailed this step is, the more stable the subsequent generation will be. AI models understand explicit visual instructions, not vague narratives. A well-written storyboard reduces the randomness of each subsequent step.
Step 2: Establish a reference image library for characters and scenes.
This is the most easily overlooked yet most indispensable step in the entire process.
The biggest problem with AI-generated videos isn't image quality, but consistency. The same character might have the same face in one episode but a different face in the next. Background colors will shift, and clothing details will disappear. Without a fixed reference image to constrain it, it's almost impossible to create a series of three or more episodes.
The solution is to use a dedicated image editing tool to "finalize" the character before generating the actual video—generating one image each from the front, side, three-quarter angle, and eye, and fixing the hair color, skin tone, clothing, and style. The same process is applied to the main scenes. This image library will be used in all subsequent shots and forms the foundation of the entire workflow.
⚠️A little trick: If you want to use JiDream to generate real-life videos, you can blur the eyes in a person's face photo, add "This person was generated by AI" to the image, and display the eyes separately in the image. This can bypass the platform's face detection restrictions.
Step 3: Control the card draw rate using the first frame.
Anyone who has worked on AI-generated videos knows the term "gacha." What's the probability that a generated video is directly usable? High-quality prompts and reference images can significantly reduce the number of gacha pulls.
The professional team's approach is to first use a raw image tool to generate the first frame of each shot, and then use this image as a reference input so that Seedance can generate subsequent motion from this frame.
In this step, the quality of the raw image tool directly determines the ceiling of the final video. The better the generated image and the more consistent the details, the better the video will be after being fed into Seedance.
This is why GPT Image 2 has had such a profound impact on the entire industry. GPT Image 2's ability to understand image descriptions has leaped to a new level; given a scene description, it can generate high-quality reference images with more stable faces and more controllable styles. Improved reference image quality leads to improved image quality across all downstream processes, creating a chain reaction.
Step 4: Editing and Compositing
Once the clips are confirmed, they can be spliced together using CapCut or other editing tools, and subtitles, voiceovers, and background music can be added. Seedance 2.0 supports generating sound effects and background music while generating video, and the lip-sync and audio synchronization are quite stable. This feature can save a significant amount of post-production work.
III. Traditional Short Dramas vs. AI Short Dramas: An Asymmetrical War
Having discussed the process at length, what about the cost? How much does a 60-episode AI short drama actually cost? How does it compare to a traditional live-action short drama?
Standard Membership: 199 RMB per month for continuous subscription, including 2210 points per month, which can generate approximately 200 seconds of video, translating to a cost of approximately 1 RMB per second.
Premium Membership: 499 RMB per month for continuous subscription, including 6160 points, which can generate approximately 560 seconds of video, with a cost of around 0.89 RMB per second.
But this price wasn't always the case.
This year, JiMeng has raised its prices several times. The original annual membership was priced at 2,599 yuan, which translates to 216 yuan per month, and members would receive 15,000 points per month.
The annual membership fee has now increased to 5199 yuan, and in April of this year, the monthly points were cut from 15000 to 6160, a reduction of more than 60%. This is equivalent to a reduction of more than half in the video length that can be generated with the same budget, resulting in an actual increase of 60%.
In the game "Jimeng", generating a 1-second video costs 11 points. Assuming a short drama episode is 1 minute long, without any card draws, the actual cost of a short video episode is about 46 yuan.
The success rate of drawing cards varies greatly depending on the quality of the prompts and the complexity of the scene. Assuming an average of four generation attempts are needed per video to obtain a usable segment, the actual computational cost for one episode of a short drama is approximately 184 yuan. This is under the premise of stable prompt quality and relatively simple scenes. If the plot is complex and the characters' movements are large, the number of card draws will only increase.
Besides computing power, there are also operating costs. A small AI short drama team typically consists of 3 to 5 people, including a screenwriter, a gacha artist, and an editor. The fixed expenses for personnel, office rent, and utilities are around 35,000 to 70,000 yuan per month. If the cost is spread across 10 dramas produced per month, the total cost per episode, including operating costs, is approximately less than 500 yuan.
Traditional live-action short dramas are divided into male-oriented and female-oriented categories based on their themes, with significant differences in cost.
Male-oriented dramas: They feature more action scenes and special effects, and the production cost of a 60-episode series is usually over 500,000 yuan, which translates to about 8,300 yuan per episode.
Female-oriented dramas: These focus on emotional storylines and have relatively controllable costs. A 60-episode series costs approximately 350,000 to 400,000 yuan, with each episode costing around 5,800 to 6,700 yuan.
In contrast, even with the added team operating costs, the total cost per episode of an AI-generated short drama is no more than 500 yuan. The cost difference between traditional live-action production and AI production for a single episode is between 15 and 40 times.
This gap means that a traditional short drama requires a bet of hundreds of thousands of yuan, and a wrong topic selection can be a serious blow, potentially requiring the entire team to recover for several months. AI short dramas, on the other hand, cost only a few hundred yuan per episode, allowing the same budget to process ten topics simultaneously, trading quantity for probability and speed for a window of opportunity.
Fourth, what does this mean for ordinary people, and are there any opportunities for them?
In 2025, the Chinese micro-drama market reached 67.79 billion yuan, with a user base of 696 million, meaning more than half of Chinese netizens watch micro-dramas. This is the fertile ground for AI micro-dramas; there's no need to cultivate a new market, as AI micro-dramas have already established a stable paid subscription habit.
Based on this, Douyin has also begun to proactively allocate traffic and funds to AI-generated original videos.
Douyin, in partnership with Jimeng, launched the "AI Creation Wave Plan S2": every two weeks, 10 high-quality pieces of content will be comprehensively evaluated, with regular content receiving a cash reward of 1,500 yuan. Authors who make the list will also have priority access to industry cooperation opportunities, commercial orders, and short film project application support.
Driven by platform incentives, a batch of content of significantly higher quality has emerged from Douyin's creative boom this month. AI-generated public service short dramas such as "Paper Airplane," "Centenarian Kindergarten," and "Farewell" have generally garnered high numbers of likes.
The monetization path is also straightforward. Domestic creators can pursue three parallel avenues: platform traffic revenue sharing, Tomato Novel CPS commission, and brand orders.
With 10,000 to 50,000 followers, the price for a single commercial order is 500 to 2,000 yuan;
With 50,000 to 100,000 followers, a single post can cost 2,000 to 5,000 yuan.
With 100,000 to 500,000 followers, a single post can earn between 5,000 and 20,000 yuan.
The platforms also have traffic revenue sharing programs. Douyin's mid-video program pays about 60 yuan per 10,000 views, and Kuaishou's Magnet Star program pays about 40 yuan per 10,000 views. The thresholds are not high.
Advertisers already heavily invest in short drama content for game promotion, app user acquisition, and brand integration; AI short dramas simply provide them with a lower-cost option.
In conclusion
Given such a large market, what kind of people are suitable to enter it?
People without any background should not jump into making short dramas. Short dramas have high requirements for character consistency and scene coherence, the workflow is complex, and the cost of trial and error is not low. A more pragmatic approach is to start by practicing with short videos.
Douyin (TikTok) features numerous videos that transform popular online phrases and sentences into content. These videos don't require a continuous storyline or fixed characters; each video is independent. These accounts gain followers quickly and achieve high view counts, making them ideal for building a personal brand and audience base. More importantly, these videos almost eliminate the need to consider character or scene consistency, allowing users to focus entirely on content selection and pacing.
Once the account is up and running and you get a feel for the tools and platform rules, then focus on improving the quality of the content and gradually try more complex short drama workflows.
The AI short drama market hasn't yet seen a true monopolist. Tools are iterating, workflows are evolving, and a team that gets a process working today might be overtaken by a better model tomorrow. This means the first-mover advantage isn't as significant as it seems, and newcomers still have a chance.
