AI micro-dramas are now the mainstay and costs have fallen sharply. A beginner's look at two hits, skills, rules, tools, podcasts and room for small teams.
2026-10-21 · Tony Cai, SolanaLink Co., Ltd.
Prologue: one evening, 3 hours of micro-drama
The production quality of micro-dramas has clearly improved, and that has made me start thinking seriously about content creation.
Last night I watched a micro-drama for 3 hours straight. The plot moved smoothly, the visual style stayed consistent from start to finish, and it was easy to watch. The movements of the lead and supporting characters, including several dance sequences, were well done too.
This is quite different from how many people picture micro-dramas. The rising quality shows that the production capability of the whole industry is improving fast.
I have worked in software engineering for more than 20 years and have never made a short video. This article is my first pass at the subject as a beginner: where the industry stands now, where the barriers are, what knowledge and tools I need to pick up, and where the opportunities for small teams lie.
Where the industry stands: AI micro-dramas are now the mainstay
In 2026, AI micro-dramas have gone from novelty to the mainstay of the industry: in China, AI live-action-style and animated dramas are expected to exceed RMB 40 billion for the full year, and more than 95% of newly released micro-dramas are made with AI.
| Metric | Figure | Source |
|---|---|---|
| China's micro-drama market (including AI live-action-style and animated dramas) | Expected to reach RMB 121 billion in 2026, of which AI live-action-style and animated dramas exceed RMB 40 billion | DataEye (compiled by 西南证券 (Southwest Securities)) |
| Micro-drama users | 851 million (May 2026), 66.7% of internet users | DataEye (compiled by Southwest Securities) |
| New releases | About 128,000 titles in Q1 2026, of which about 122,000 were AI micro-dramas, more than 95% | 中国网络视听协会 (China Netcasting Services Association) (澎湃新闻 (The Paper)) |
| Annual output | From 60,000 titles in 2025 to 1 million in 2026 (forecast) | 中泰证券 (Zhongtai Securities) (南方财经 (Southern Finance)) |
| Overseas AI live-action-style and animated dramas | Expected to exceed US$4 billion in 2026, up 390% year on year, nearly 70% of overseas micro-dramas | DataEye (Southern Finance) |
What made this scale possible is that both cost and labor have fallen sharply:
- Cost: In the first half of 2026, the production cost of animated dramas and photorealistic AI micro-dramas fell 80%–90% from 2025, to only 2%–10% of a traditional micro-drama.
- Team and schedule: A traditional micro-drama takes about 15 people and more than 4 weeks; an AI micro-drama now takes 3–5 people and around 10 days on average.
- Models: Video models such as Seedance 2.0/2.5, Kling (可灵) 3.0 and MiniMax H3 keep improving. By the estimate of 财联社 (Cailian Press), generating a 15-second clip with Seedance 2.0 costs about RMB 15, or RMB 1 per second.
AI is also moving from micro-dramas into long-form series. 《后西游记》 (Later Journey to the West), produced by 芒果 TV (Mango TV), premiered in prime time on 湖南卫视 (Hunan TV) on August 31. Its characters, action and scenes are all AI-generated, with no human actors.
The barrier has moved: making it is no longer scarce; making it good and getting it sold is the hard part
Once AI brought production costs down, being able to make a drama was no longer the barrier, and competition moved to content and operations.
- Prices have bottomed out: Quotes for standardized AI micro-dramas have fallen from RMB 5,000 per minute to a few hundred, close to cost; custom projects can still be quoted at RMB 10,000–20,000 per minute (Southern Finance). The gap between the two is the premium for content and creativity.
- Tools are converging: A report by The Paper tested a batch of mainstream AI micro-drama tools and found that the differences at the model and workflow level keep getting smaller. The author concludes that in the end the industry counts only views, conversions and collected revenue, and that whoever owns a content methodology will matter more.
- Platforms are turning to live action: As AI dramas flooded the market, platforms stepped up their support for live-action content instead. The "微短剧精品创作传播计划" (Quality Micro-Drama Creation and Distribution Program) of 广电总局 (National Radio and Television Administration, NRTA) calls for six key platforms to invest at least RMB 6 billion in live-action micro-dramas; Hongguo (ByteDance's micro-drama app) has also committed RMB 0.5 billion specifically to live-action shooting and realistic subjects (Southwest Securities).
- Regulation is tightening: In July 2026, 《微短剧发展管理办法》 (Measures for the Administration of Micro-Drama Development) was formally issued, and micro-dramas entered a stage of standardized management (Southwest Securities).
Zhongtai Securities' view is that user attention will become a scarce resource, and the platforms that control distribution will become where the profit is (Southern Finance).
For beginners, this means tools will keep getting cheaper and more alike. What is truly scarce comes down to three things: the ability to tell a story well, distinctive subjects and perspectives, and the operational skill to get content in front of an audience.
Two real cases
These two works happen to represent two paths: one takes the payoff-driven drama formula to the extreme, while the other broke out on the strength of one person's story.
Case 1: 《六十倍觉醒》 (Sixty-Fold Awakening): the payoff-driven formula taken to the extreme
The drama I mentioned at the start, the one I watched for 3 hours straight, is 《六十岁山野老樵夫一朝觉醒六十倍修炼速度》 (A 60-Year-Old Mountain Woodcutter Awakens Overnight to Sixty Times the Cultivation Speed) (YouTube). Taken apart, each of its hit elements maps to a formula you can learn:
| Hit element | How the drama does it |
|---|---|
| High-concept contrast | The lead is a 60-year-old woodcutter who has chopped wood for 60 years; his cheat ability ("golden finger") is "the longer he lives, the faster he cultivates," so 1 session of cultivation is worth 60 of anyone else's |
| Hidden-master trope | The villains all take him for an old man with one foot in the grave; he first plays weak to lower their guard, defeats them with a single move, and goes straight back to being a stooped old man afterwards |
| Emotional anchor | He must protect the granddaughter he has lived with, each the other's only family; his younger brother's family once tried to force him to sell his ancestral home for 80 taels, and later come to his door with precious medicine, begging |
| Numeric feedback | His realm rises step by step from Qi-Blood Level One to the Inner Qi realm, and numbers such as "one level broken through in 6 days" give an instant payoff, like leveling up in a game |
| Map upgrades | From mountain village to market town, to county seat, then to prefectural city; each new map brings a new set of opponents |
| Chain of suspense | A piece of purple wood leads to a giant serpent deep in the mountains, which leads in turn to a cult, a divine blade and a temple's secrets, each link tied to the next |
Dramas like this rely not on visuals but on a structure that has been proven again and again. For a beginner learning to write scripts, the best textbook is to take apart hits like this one.
Case 2: 《我这一生最大的罪，是把人写成了妖……》 (The Greatest Sin of My Life Was Writing People as Demons…): one person, one essay, nearly 18 million views
This is the short film in which the crowd cries "苍天已死，黄天当立" ("The Azure Heaven is dead; the Yellow Heaven shall rise"), the slogan of the Yellow Turban Rebellion of the late Eastern Han dynasty (184 AD). The Bilibili uploader "青瓜蛋丶" released it on August 17, 2026 (original on Bilibili); it runs 26 minutes.
- Story: In the late Eastern Han dynasty, the court historian Pei Lingshi is ordered to investigate "demonic omens" among the common people, only to find that most of the so-called demons are commoners driven to the brink by famine and harsh rule. The supernatural episodes draw on Han-dynasty calamities recorded in 《搜神记》 (In Search of the Supernatural). After Zhang Jue delivers a speech in the film, the crowd shouts "苍天已死，黄天当立" in unison, and the Yellow Turban Rebellion begins (unwire.hk).
- Numbers: Views passed 12.8 million in the first week, and viewer tips reached RMB 1 million (unwire.hk); by September 1 the count was close to 18 million (界面新闻 (Jiemian News)).
- Production: The creator did the script, AI generation, retouching and editing alone. He used medium and long shots, silhouettes and smoke to hide AI's flaws in lip movement and motion (HK01). The workflow: first create the characters, costumes, scenes and key frames with GPT Image 2, then use Seedance 2.5 to set the characters in motion, adding camera moves, action and sound (unwire.hk).
- Against the formula: It unfolds slowly over 26 minutes instead of following the micro-drama pattern of a twist every few seconds (HK01).
- Where the script came from: The script started as an essay the creator wrote several years ago and was gradually reworked into its present form (Jiemian News).
Jiemian News noticed that the discussion in the comments has moved from "Was this made with AI?" back to the characters, the story and the emotions themselves. Everyone can use the same tools; what really sets one work apart from another is whether you have a story you want to tell.
Seen together, one case shows that formulaic payoff-driven dramas still have a large audience, and the other shows that a work that slows down and deals with real problems can break out too. Small teams can take either path, but neither gets around the same thing: a good script comes first.
Industry knowledge a beginner needs
The tools are quick to learn. What really takes time is knowledge of storytelling and of the business, plus a compliance baseline.
| Area | What to understand | Why it matters |
|---|---|---|
| Script and pacing | The hook in the first few seconds, the cliffhanger at the end of each episode, the rhythm of twists, where to put the paywall | Micro-dramas make money by keeping viewers watching, so pacing is itself the business model |
| Storyboarding and visual language | Shot sizes, camera positions, transitions, and how to make a point clearly with images | AI generates the shots, but which shot to use is still a human decision |
| Character consistency | Character sheets and reference material that keep the same person looking the same from shot to shot | A character who drifts off-model is what most easily pulls viewers out of the story |
| Sound | Voice-over, voice timbre, music and sound effects | Sound sets the emotion; a podcast relies on sound alone |
| Platforms and distribution | Each platform's recommendation system, completion rate, paid traffic | Platforms hold the attention; content has to be seen first |
| Monetization | Platform revenue share, paid content, brand placement, live-stream e-commerce (Southern Finance) | Work out how you will make money first, then decide what content to make |
| Compliance | AI labeling, filing, portrait rights and copyright | Non-compliant work is taken down outright; see below |
Compliance deserves a section of its own. These are the rules in mainland China (The Paper):
- Dual labeling: Since September 1, 2025, 《人工智能生成合成内容标识办法》 (Measures for Labeling AI-Generated and Synthetic Content) has required audio and video that AI generated or helped make to do two things at once: carry an explicit label in a prominent position on screen, and carry implicit traceability information in the file's metadata.
- Opening-credit label: The NRTA's AI 微短剧分类分层标准 (classification and tiering standard for AI micro-dramas) has applied since July 1, 2026. If the share of AI reaches the red line in any one of the three stages of script, visuals and voice, the opening must carry the label "AI 辅助制作" ("AI-assisted production").
- Tiered filing: A project with an investment of RMB 800,000 or more, or with a special subject, reports to the NRTA; one of RMB 300,000–800,000 files with the provincial authorities; one below RMB 300,000 is self-reviewed and reported by the platform that airs it.
- Portrait rights and copyright: In April 2026, Hongguo took down a period-style micro-drama with a popularity score above 40 million, because it carried no AI label and had used AI to misappropriate a blogger's portrait photos to create its characters.
Platforms in Japan or elsewhere outside China have different rules, which need to be checked separately.
Tool map
The gap between the leading video models is now small, so choose tools by whether the workflow suits you, not by chasing the strongest model.
| Stage | Representative tools | Notes |
|---|---|---|
| Script and ideas | Large language models such as Claude and ChatGPT | Breaking down the outline, writing episodes, revising dialogue; the direction and the trade-offs are still decided by people |
| Video generation (international) | Google Gemini Omni / Veo 3.1, Runway, Sora | Strong image quality and camera control; Gemini Omni Flash ranks near the top of the text-to-video leaderboard |
| Video generation (China) | Seedance 2.0/2.5, Kling 3.0, MiniMax H3 (Hailuo), Tongyi Wanxiang (通义万相) Wan 2.7 | Seedance is already built into Doubao (豆包), Jimeng (即梦) and Jianying (剪映); MiniMax H3 and Wan 2.7 are open-source models |
| All-in-one micro-drama platforms | LibTV, AniShort, 帧赞 (Zhenzan), 万兴剧厂 (Wanxing Juchang) | Script, storyboard, assets and final cut in one pipeline; good for getting a first piece done end to end |
| Voice and sound | ElevenLabs | Natural voices, and it can clone your own voice; suits narration and storytelling |
| Editing and post-production | 剪映 / CapCut | Subtitles, editing and publishing in one place; Seedance is already integrated |
Three principles for choosing:
- Look at the workflow, not the model. On the blind-test leaderboard of Artificial Analysis, the confidence intervals of the top few video models overlap, and there is no significant difference in viewer preference (Southwest Securities).
- Get one piece done on an all-in-one platform first, then take it apart and optimize. The goal of a first work is to go through the whole pipeline and learn where each stage is hard. Tools that are just a wrapper around a model will be worth less and less (The Paper).
- Budget by the second. MiniMax H3's API price is RMB 0.8 per second at 2K resolution and RMB 0.5 per second at 768P; Seedance 2.0 is about RMB 1 per second (Southwest Securities). On top of that, count the cost of generation retries ("gacha" waste): a single shot often has to be generated several times before you get one you can use.
Podcasts: a cheaper way in
If competition in video has already moved to the content itself, podcasts are a cheaper starting point, and one that shows more of your own character.
Low cost. There are no visuals to generate and no "gacha" waste; a microphone and a computer are enough to start. AI can cover most of the post-production work (AI Studios review):
- Adobe Podcast's speech enhancement removes noise and makes voices clearer
- ElevenLabs generates natural narration and can also clone your own voice
- NotebookLM turns uploaded material into conversational audio between two hosts
- Wondercraft makes it easy to add music and sound effects, which suits storytelling shows
The host is the asset. A 2024 podcast industry report found that the top three reasons listeners subscribe to a show are informative content (57.0%), a light and humorous tone (54.4%), and liking the host or guests (53.8%) (前瞻产业研究院 (Qianzhan Industry Research Institute)). The visuals can be handed to AI, but a host's views and personality are hard to replace.
Growing in both China and Japan:
- Chinese podcasts: In 2025, Xiaoyuzhou (Chinese podcast app) gained 64,000 new shows and 696,700 new episodes; the fastest-growing category was AI startups, up 310% year on year (Qianzhan Industry Research Institute). In the same year, AI-related creators grew 187% and related shows 239% (凤凰网 (Phoenix New Media)).
- Japan: 18.2% of people listen at least once a month (14.2% in 2020); the figure reaches 40.5% among 15–19-year-olds and 32.3% among 15–29-year-olds, higher than TVer, TikTok and Prime Video. 80.8% of users listen while doing something else, and 76.2% watch video podcasts every month (Weekly ASCII).
A storytelling show can start as audio only and later add visuals to become a video podcast. Two things to keep in mind: AI-generated voices must be labeled as the rules require, and cloning someone else's voice requires that person's permission first.
Where small teams can break through
My view: small teams cannot compete with industrial-scale output, but they can find a place in three areas: vertical niches, cross-language work, and "tools plus content."
- Go for vertical long-tail niches, not mass-market hits. Zhongtai Securities expects AI film and TV content to split into three tiers: standardized basic content, vertical long-tail subjects, and creative hits (Southern Finance). Standardized content is a contest of production capacity; the opportunity for small teams is in the second tier, a subject that is specific enough and that you truly understand.
- Working across languages is a natural advantage. AI has made multilingual remakes very cheap. Overseas AI live-action-style and animated dramas are expected to exceed US$4 billion, and Japan is one of the key markets for overseas micro-drama platforms (Southwest Securities); Japan's domestic micro-drama market is expected to reach JPY 153 billion in 2026 (CREAVE). I live and run a company in Tokyo, and Chinese is my native language: the subjects and perspectives that sit between China and Japan are a differentiator in themselves.
- Keep each try cheap, and try many times. Zhongtai Securities believes film and TV investment will shift from betting on single projects to a low-cost, multi-project "portfolio of options" (Southern Finance). Small teams are well suited to this: test subjects with low-cost works first, then invest more in what takes off.
- Build the tools and the content together. Industry observers believe the tools that win in the end will be those that build up a library of subjects, a library of characters and a reusable content method (The Paper). Creators with an engineering background have an edge here: they can be their own first users and turn the pain points they hit while creating straight into tools.
Closing: start doing
The barrier to content creation hasn't disappeared; it has only moved from "can you make it" to "can you tell it well, and can it be seen." My next step is to get started at the lowest possible cost:
- Pick a topic for a storytelling podcast and make 3 trial episodes with existing tools.
- Pick one video tool and make a complete 1-minute short, going through the whole process from script to publishing.
- Write down the problems I run into along the way, as material for the next article.
— Tony Cai, SolanaLink Co., Ltd.
References
- 《六十岁山野老樵夫一朝觉醒六十倍修炼速度》 (YouTube)
- 《我这一生最大的罪，是把人写成了妖……》 (original on Bilibili)
- unwire.hk: the viral success of 《把人写成了妖》 (Writing People as Demons) and the tools behind it (2026-08-24)
- HK01: 10 key points about 《把人写成了妖》 (2026-08-28)
- Jiemian News: AI film and TV competition shifts from tools to content (2026-09)
- Southern Finance: the AI micro-drama price war and the RMB 40 billion market (2026-09-06)
- The Paper: the AI micro-drama tool market and its funding (2026-06-24)
- Southwest Securities: media sector investment strategy for Q4 2026 (2026-09)
- The Paper: new rules on labeling and filing AI micro-dramas (2026-07-23)
- Qianzhan Industry Research Institute: analysis of podcast content consumption in China, 2026 (2026-05-28)
- Phoenix New Media: Xiaoyuzhou's 《2026 AI 趋势报告》 (2026 AI Trends Report) (2026-08-22)
- Weekly ASCII (週刊アスキー): survey of podcast use in Japan (2026-04-17)
- Web担当者Forum (Web Tantousha Forum): CREAVE seminar on the micro-drama market (2026-01-06)
- AI Studios: review of AI podcast generators, 2026 (vendor article, updated 2026-08)
Comments
Comments (0)