朋友们，今天来聊一个我早就想吐槽的赛道：AI微短剧。
先报几个数，你们感受一下：2026年第一季度，全行业上线微短剧约12.8万部，其中AI短剧占比超过95%。也就是说，12万多部里，只有几千部是真人拍的。
再看另一个数：今年4月至今，全网AI短剧下架总量已超7万部，新剧过审率骤降到30%以下。
12万部上线，7万部下架。这个合格率，大概也就比你写代码第一遍就跑通的概率高那么一点点。
AI短剧到底有多离谱？
我看过几部，怎么说呢……看完你会觉得「这也行？」。
有个叫《桃花簪》的古风AI短剧，直接把人家汉服博主的写真照扒下来，用AI换脸生成反派角色。博主发文维权，平台72小时内没等到出品方拿出授权证据，直接全线下架，还把出品方禁更了15天。
这还不是最离谱的。有些剧为了蹭流量，什么奇葩剧情都写得出来：「家具渡河」——普通家具在河里漂着就渡了河；「床垫挡炮」——一张床垫挡住了炮弹；什么婆媳矛盾、贫富羞辱、外貌焦虑，怎么狗血怎么来，怎么刺激怎么编，AI写起来根本没有道德底线这种概念。
更魔幻的是生产效率。以前一个团队一个月能出两三部短剧就不错了。现在呢？一个人、一台电脑、几个AI工具，一周能产几十集。质量？不重要。数量堆上去，总有那么几部能跑出来。
这不是内容创作，这是内容挖矿。疯狂试错，靠概率赚钱。
监管的80万红线，到底在卡什么？
于是监管出手了。AI微短剧按投资额度分三档：80万以上重点管理，30-80万普通管理，30万以下其他管理。很多人第一反应是：「AI短剧成本这么低，谁会投80万？」
这个问题问反了。你应该问：「为什么AI短剧的成本会这么低？」
因为AI干掉了几乎所有的生产环节成本——没有演员、没有场地、没有摄像、没有后期剪辑，甚至可能连编剧都是AI。一部剧的成本可能就是几百块算力费+一点电费。这种成本结构下，质量自然没有保障——反正便宜，大不了下架再换一部。
80万这条线，本质上是在说：你要真想认真做剧，那就投入点像样的成本，接受像样的监管。你要只想低成本批量造垃圾？那对不起，垃圾就该待在垃圾桶里。
而且特殊题材不管投资多少，一律按重点管理。意思很清楚：你AI胡编乱造点霸总甜宠也就算了，想碰政治、军事、国家安全这些东西，门都没有。
给创作者的三个提醒
最后给还在做或者想做AI短剧的朋友们提个醒：
第一，素材版权要干净。不要随便从网上扒图扒视频来训练或者生成。盗脸盗声这种事，以前可能没人管，现在是真的会被告、会下架、会赔钱。AI生成的内容，版权责任在你，不在工具。
第二，别踩内容红线。广电总局列的那一堆特殊题材，能不碰就不碰。AI生成的东西有时候自己都不知道在说什么，万一输出了什么不该有的，锅是你的。发布前自己多过几遍，别偷懒。
第三，早点做精品化转型。野蛮生长的窗口期已经结束了。靠批量制造垃圾赚快钱的日子，越来越短了。当监管收紧、平台提高门槛，真正能活下来的，一定是那些有内容能力、有合规意识的团队。
说起来也挺好笑的——AI把内容创作的门槛砸到了地板上，最后发现真正的门槛不是工具，是审美和判断力。
今天就槽到这里，明天继续。
Folks, let's roast a sector I've been dying to talk about: AI microdramas.
Start with some numbers to set the scene: in Q1 2026, the industry launched roughly 128,000 new microdramas. AI-generated ones accounted for over 95%. Out of 120K+ titles, only a few thousand were actually filmed with real people.
Now the other number: since April this year, over 70,000 AI short dramas have been taken down nationwide, and the pass rate for new shows has plummeted to under 30%.
128K launched, 70K removed. That success rate is barely better than your code compiling on the first try.
How Bad Are AI Microdramas, Really?
I've watched a few and… honestly, you finish one and go, 'this actually got approved?'
There was this one called 'Peach Blossom Hairpin' — a period AI drama — that straight up lifted a Hanfu blogger's personal photos and AI-swapped the face to make a villain character. The blogger called it out, the platform gave the studio 72 hours to produce licensing evidence, they couldn't, so the whole show got pulled and the production company was banned from uploading for 15 days.
And that's not even the wildest stuff. For traffic, some shows will crank out any absurd plot: 'furniture crossing a river' — regular household items just float across a river; 'mattress blocks cannon fire' — a bed mattress stops artillery. Mother-in-law conflicts, wealth shaming, appearance-based humiliation — the more melodramatic, the more AI generates it. It has zero concept of a moral floor.
The production speed is even more absurd. Back in the day, a team might put out two or three shows a month and that was solid output. Now? One person, one computer, a few AI tools — you can pump out dozens of episodes a week. Quality? Doesn't matter. Stack enough quantity and probabilistically, something will hit.
This isn't content creation. This is content mining. Spray and pray, live on probability.
What's the 800K Regulatory Line Actually Blocking?
So regulators stepped in. AI microdramas get three tiers based on investment: 800K+ gets key-level management, 300K–800K gets standard management, under 300K gets platform-level review. A lot of people's first reaction: 'AI dramas cost so little, who would invest 800K?'
You're asking the wrong question. Ask instead: 'Why are AI microdramas so cheap to make?'
Because AI eliminated almost every production cost — no actors, no sets, no cameras, no post-production editing, sometimes not even a real screenwriter. A show might cost a few hundred bucks in compute plus some electricity. With that cost structure, quality naturally doesn't matter — it's so cheap, if it gets taken down you just make another one.
The 800K line is essentially saying: If you're serious about making a show, put real money into it and accept real oversight. If you're just churning out garbage at scale… well, garbage belongs in the trash.
And sensitive topics get key-level management regardless of budget. The message is clear: you can AI-generate your CEO rom-coms, but political, military, national security stuff? Not on your life.
Three Tips for Creators
Three quick warnings for anyone still in or thinking about getting into AI microdramas:
First, keep your source material clean. Don't scrape random images and videos from the internet to train or generate content. Stolen faces and voices used to fly under the radar. Now people actually get sued, shows get taken down, and judgments happen. You're liable for AI-generated content, not the tool.
Second, stay off content red lines. That whole list of sensitive topics the NRTA published? Just don't go there. AI sometimes generates stuff without you even realizing it. If it outputs something it shouldn't have, the bucket stops with you. Review before publishing. Don't be lazy.
Third, shift to quality while you can. The wild growth window is closing. The fast-money era of flooding the market with low-effort content is running out of runway. When regulation tightens and platforms raise the bar, the teams that survive will be the ones with real content ability and compliance awareness.
It's kind of funny when you think about it — AI dropped the barrier to content creation down to the floor, and it turns out the real barrier was never tools. It was taste and judgment.
That's all the roasting for today. More tomorrow.
