2026 年春节档，真人微短剧的上线量只有 AI 微短剧的五十分之一，总播放量却是后者的 25 倍。换算成单部作品的平均观看量，每部真人剧大约是每部 AI 剧的1250 倍。
During the 2026 Spring Festival season, the number of live-action micro-dramas that went online was just one-fiftieth that of their AI-generated counterparts, yet their total views were 25 times higher. Converted to average views per title, each live-action micro-drama drew roughly1,250 timesthe audience of each AI one.
1250 倍已经不是“观众还需要时间适应”能够解释的差距。它更像市场用脚投票的结果。
A gap of 1,250 times is not something that “audiences just need time to adjust” can explain. It looks far more likethe market voting with its feet.
然而，市场遇冷的另一端，行业的生产端却展现出完全相反的蓬勃信心。同一季度，AI 微短剧上线约十二万两千部，占全部微短剧供给的 95%以上。站在供给侧看，AI 已经是压倒性的主流；站在需求侧看，观众的时间仍然高度集中在极少数真人作品上。供给爆炸与需求冷淡同时发生，暴露了一个深层次的错位关系：AI 让“拍出来”变得极其容易，但并没有让“想看”同步变多。
Yet while the market ran cold, the production side of the industry radiated entirely the opposite confidence. In the same quarter, approximately 122,000 AI micro-dramas went online, accounting for over 95% of total micro-drama supply. From the supply side, AI was already the overwhelming mainstream; from the demand side, viewers' time remained highly concentrated on a handful of live-action titles. A supply explosion and demand indifference occurring simultaneously exposed a deep structural mismatch: AI has made “getting something produced” extraordinarily easy, but has not made “wanting to watch” grow in step.
《左耳》 AI 真人短剧，是目前检验这个错位最好的样本。 2015 年，饶雪漫的同名青春小说被改编为电影，上映两天票房破亿，最终收获约五亿元，捧出了陈都灵、杨洋、欧豪、马思纯一批新面孔。十一年后的今天，饶雪漫抱着"要活下去"的心态转向短剧，又把《左耳》拿出来做 AI 改编——全部角色由演员提供肖像授权、 AI 完成表演生成。她手里握着一般 AI 短剧欠缺的几乎所有筹码：十年前的现象级 IP 、原著作者亲自把关、专业电影导演全程执导、技术公司达瓦科技以项目分红换算力免费。如果 AI 短剧有可能说服市场，这个项目应该是胜算最大的那一个。
The AI live-action micro-drama adaptation ofThe Left Earis, to date, the best specimen for examining this mismatch. In 2015, Rao Xueman's eponymous youth novel was adapted into a feature film that crossed 100 million yuan at the box office within two days, ultimately grossing around 500 million and launching the careers of Chen Duling, Yang Yang, Ou Hao, and Ma Sichun. Eleven years later, Rao Xueman, driven by the need "to survive," has turned to micro-dramas and broughtThe Left Earback for an AI adaptation—every character performed by AI using likenesses licensed from real actors. She held nearly every card that most AI micro-dramas lack: a once-phenomenal IP from a decade ago, the original author personally overseeing the project, a professional film director helming the entire production, and tech company Dawa Technology providing computing power for free in exchange for a share of revenue. If any AI micro-drama stood a chance of winning over the market, this one should have hadthe best odds.
然而它遇到的第一道坎却不是技术。饶雪漫在微博为 AI 版《左耳》发起选角，最终只有四个人参与。她在小红书的日常选角直播观众可达百万，AI 选角直播的人数只有四分之一。这当然不代表《左耳》的知名度远不及他作，而是“AI 版”三个字本身就劝退了大多数人。
Yet the first hurdle it encountered was not technical. When Rao Xueman launched a casting call for the AI version ofThe Left Earon Weibo, only four people participated. Her regular casting livestreams on Xiaohongshu drew audiences of up to a million; the AI casting livestream attracted only a quarter of that. This hardly meansThe Left Earis far less well-known than her other works—rather, the three characters “AI version” were enough to turn most people away.
演员一端的反应更值得细看。饶雪漫透露，片中关键角色黎吧啦先后谈妥了两位有名气的演员，合同都走到了签约阶段，最终却被对方公司叫停。几乎同一时期，爱奇艺高调宣布"百余位艺人入驻 AI 艺人库"，随后张若昀、于和伟等人接连否认签署过 AI 授权。演员并非拒绝赚钱，因为他们心里也有一笔账：肖像一旦交给一个自己无法完全预见用途的生产流程，虽然可能拿到一笔短期收入，但可能丧失的是形象控制权、角色选择的主动权，以及在公众心中"只有我能演这个角色"的稀缺性。稀缺性是明星商业价值的根基；一旦数字分身可以被反复调用，曝光增加未必换来价值增加，更可能是审美疲劳和品牌稀释。
The reaction on the actors' side is even more worth examining. Rao Xueman revealed that for the pivotal role of Li Bala, two well-known actresses were successively approached and agreed to terms, with contracts reaching the signing stage, only for both to be pulled back by their management companies. Around the same period, iQIYI announced with great fanfare that "over a hundred artists had joined its AI Artist Library," after which Zhang Ruoyun, Yu Hewei, and others publicly denied having signed any AI licensing agreements. Actors are not refusing to make money—they are doing their own math: once a likeness is handed over to a production pipeline whose uses one cannot fully foresee, the short-term income may come at the cost of control over one's image, the initiative in choosing roles, and the scarcity of being the only person who can play a given character in the public's mind. Scarcity is the bedrock of a star's commercial value. Once a digital double can be called upon again and again, more exposure does not necessarily translate into more value—more likely, it leads to aesthetic fatigue and brand dilution.
主创团队在制作中的实际经历，又从另一个方向印证了市场信号所暗示的结论。团队最初让 AI 同时生成画面和配音，结果 AI 声音配上 AI 面孔，"真实感大打折扣"。两个角色的配音，技术人员反复"抽卡"比对了一整夜，第二天早上饶雪漫听了成果，用她自己的话说，当时自己"几乎崩溃"——"很假呀，达不到我要的"。最后，影片用的是真人配音。画面交给 AI，声音交给 AI，表情交给 AI，可每到决定作品质感的关键环节，真人都不得不被请回来兜底。一个“不需要真人”的项目，处处依赖真人来维持基本的可信度。 这也恰恰证明，当前技术在连续、细腻、有情感因果关系的表演面前，力不从心的地方比比皆是。
The creative team's actual experience during production confirmed, from yet another angle, what market signals had already implied. Initially, the team had AI generate both the visuals and the voice acting. The result—AI voices layered onto AI faces—left "the sense of realism severely diminished." For two characters' voice-over, technicians spent an entire night cycling through AI-generated takes and comparing them. The next morning, Rao Xueman listened to the results and, in her own words, "nearly broke down"—"It sounds so fake; it doesn't meet my standards." In the end, the film used live voice actors. Visuals were handed to AI, voices were handed to AI, facial expressions were handed to AI—yet at every juncture that determined the work's texture, real humans had to be called back in to hold the line. A project premised on “not needing real people” depended on real people at every turn just to maintain basic credibility.This is precisely what proves that, when confronted with sustained, nuanced performance built on emotional causality, current technology falls short at every step.
面对这些，行业从业者的回应几乎遵循了同一种口径。饶雪漫说，观众厌恶的是流水线式的千篇一律，"技术迭代会迅速解决"；至于 AI 太真引发不适，"纠结这个没意义，大家迟早都会接受"。导演何必认为"沉默的大多数"对影片是否用 AI 其实无所谓，"大家忙了一天，图个乐呵，好看就行"。短剧制作人孙晨说得更干脆：短剧是商品不是作品，"疯狂卷画面多精美，是无效成本"。
Faced with all this, the industry's responses followed almost the same script. Rao Xueman said that what audiences despise is the cookie-cutter uniformity of assembly-line production, and that "iterative improvements in technology will resolve this quickly"; as for AI looking too real and triggering discomfort, "there's no point dwelling on it—everyone will accept it sooner or later." Director He Bi argued that the "silent majority" doesn't really care whether a show uses AI: "People are tired after a long day; they just want something fun—if it looks good, that's enough." Micro-drama producer Sun Chen put it even more bluntly: micro-dramas are products, not artworks—"obsessing over how exquisite the visuals are is wasted cost."
这套解释有一半讲得通。技术会进步，一部分"恐怖谷"的视觉不适确实会随着模型迭代而减轻。大多数观众也不会在每一帧画面上追问"这是不是 AI"。但它忽略了另一半。如果观众排斥的仅仅是脸不够真，那么那些已经做到高度逼真的 AI 作品，播放表现应当大幅领先粗糙的同行——事实上它们远没有拉开这种差距。观众在意的东西远远不是表面上的 “五官假”。
This line of reasoning is half right. Technology will improve, and some of the "uncanny valley" visual discomfort will indeed ease with model iterations. Most viewers will not interrogate every frame asking, "Is this AI?" But it ignores the other half. If audiences were merely put off by faces that don't look real enough, then AI productions that have already achieved ahigh degree of realismshould dramatically outperform their cruder peers—in fact, they have not pulled ahead by any such margin. What audiences resist goes far beyond superficially“fake features.”
一场表演打动人，不仅因为泪流得恰到好处、微表情恰好在第几帧出现，还因为观众知道，或者至少相信，那背后站着一个活人。那个人花了时间揣摩角色，带着自己的阅历进入表演，在现场和对手演员彼此激发，甚至在一条导演没预料到的方向上走出更好的瞬间。观众消费的，部分就是这种“此处有人”的确信。把确信抽掉，画面在像素层面无可挑剔，体验也会变得空洞——像读一篇语法正确但没有作者的文章。
A performance moves us not only because tears fall at just the right moment or a micro-expression appears on precisely the right frame, but also because audiences know—or at least believe—that behind it stands a living person. That person spent time inhabiting the character, brought their own life experience into the performance, sparked off a scene partner on set, and perhaps wandered in a direction the director never anticipated, arriving at a better moment. Part of what audiences consume is this conviction that “someone is here.” Strip that conviction away, and even if the image is pixel-perfect, the experience turns hollow—like reading an essay that is grammatically flawless but has no author.
体育转播能帮助理解这件事。百米赛跑的视觉信息极其简单：十秒之内一个人跑到终点。 AI 完全可以生成一段比真实比赛更丝滑的奔跑画面，但不会有人为此买单。观众看的不是"奔跑"这个动作，而是一个具体的人、在具体的压力下、做到了什么。人被拿掉，比赛就不存在了。影视表演当然比短跑复杂得多，层次也丰富得多，但底层机制相通：观众对“此处有真人在投入、在冒险” 这个事实本身赋予价值。 AI 可以模拟投入的样子，但它自己并不需要承担任何代价。
Sports broadcasting helps illustrate the point. The visual information in a hundred-meter sprint is extremely simple: within ten seconds, a person reaches the finish line. AI could easily generate a clip of running smoother than any real race, but no one would pay to watch it. What viewers watch is not "the act of running" but a specific person, under specific pressure, achieving something specific. Remove the person, and the race ceases to exist.Screen acting is of course far more complex and layered than a sprint, but the underlying mechanism is the same: audiences assign value to the very fact that“a real human being is investing effort here, taking a risk here.”AI can simulate the appearance of investment, but it itself bears no cost.
"迟早会接受"的说法，把一个偏好问题误读成了习惯问题。习惯问题确实会随时间消退。十几年前许多人瞧不起网剧，如今网剧早已是主流消费。但那次跨越之所以成功，是因为网剧和电视剧提供的核心体验是同一种：真人演一个故事，只是播出渠道从电视变成了手机。观众接受了新渠道，没有被要求放弃对真人表演本身的需求。 AI 演员要求观众接受的，是用算法生成的内容去替代人的表演——这是对体验本身的替换，而不只是换了个屏幕。拿网剧的先例来类推，在逻辑上并不成立。
The claim that people "will accept it sooner or later" misreads a preference as a habit. Habits do fade with time. A little over a decade ago, many people looked down on web series; today, web series are thoroughly mainstream consumption. But that transition succeeded because web series and television dramas offered the same core experience: real people acting out a story, with only the distribution channel changing from television to mobile screens. Audiences accepted a new channel without being asked to give up their need for live human performance itself. What AI actors ask audiences to accept is the substitution of algorithmically generated content for human performance—a replacement of the experience itself, not merely a change of screen. Drawing an analogy from the web-series precedent does not hold up logically.
那 AI 短剧是不是注定走不通？也不是。不过前提是行业必须搞清楚卖点。
Does this mean AI micro-dramas are doomed to fail? Not necessarily. But the prerequisite is that the industry must be clear about what it is actually selling.
市场的行为其实已经给出了区分。AI 作为生产工具，包括真人出镜表演，AI 辅助完成场景搭建、视觉特效、调色、补拍、多语言翻译——这条路观众接受度高，因为核心表演仍由人完成，AI 压缩的是后台成本，提升的是制作天花板。 王祖贤授权网易游戏制作 AI 广告短片《倩影》之所以讨论虽多但争议不大，部分也是因为它的框架比较清楚：明星的经典形象加上一段指定用途的广告，观众被唤起的是怀旧情感，评价标准也不是"这算不算好的表演"。 AI 作为表演主体（面孔、声音、动作全部或大部由算法生成）这条路得到的市场反馈，目前就是那 1250 倍的冷遇。
The market's behavior has, in practice, already drawn a distinction. AI as a production tool—where real actors perform on camera while AI assists with set construction, visual effects, color grading, reshoots, multilingual translation—enjoys high audience acceptance, because the core performance is still delivered by humans, and AI compresses back-end costs while raising the production ceiling.The AI advertising shortQianying, featuring Joey Wong's licensed likeness for NetEase Games, generated plenty of discussion but relatively little controversy, partly because its framework was clear: a star's iconic image plus a designated advertising use. What it evoked in audiences was nostalgia, and the standard of judgment was never "does this count as good acting." AI as the performing subject—where face, voice, and movement are entirely or predominantly algorithm-generated—has received market feedback that currently amounts to that 1,250-fold cold shoulder.
行业的困惑在于，许多人把这两种产品当成同一种东西的"低配版"和"高配版"，认为技术成熟之后观众自然会从前者滑向后者。但数据暗示它们可能根本是两种生意。前者的卖点仍是真人表演，AI 只是幕后降本；后者需要建立一种全新的、不依赖“假装是真人”的价值主张。动画片从不因为"不是真人"而被瞧不起，游戏 CG 从不因为"没有演员在现场"而失去市场，它们成功的关键恰恰在于从未伪装成真人实拍，而是和观众建立了另一套审美契约。 AI 短剧最大的尴尬，是卡在两者中间：想获得真人表演的情感分量，却不愿承担真人表演的成本和约束；想利用生成技术的效率，却不愿面对观众对机器生成内容天然的预期落差。
The industry's confusion is that many treat these two products as the "budget version" and the "premium version" of the same thing, assuming that once the technology matures, audiences will naturally slide from the former to the latter. But the data suggests they may be two entirely different businesses. The former still sells human performance, with AI merely cutting costs behind the scenes; the latter needs to establish an entirely new value proposition that does not depend on “pretending to be a real person.”Animation has never been looked down on for "not being live-action." Game cinematics have never lost their market for "having no actors on set." The key to their success lies precisely in the fact that they never pretended to be live-action footage—they established a different aesthetic contract with their audience. The deepest embarrassment of AI micro-dramas isbeing stuck in between: wanting the emotional weight of live performance without bearing its costs and constraints; wanting the efficiency of generative technology without confronting the gap between what audiences naturally expect from machine-generated content and what they expect from human work.
最终决定《左耳》 AI 版成败的，不会是预告片的讨论热度，而是正片上线后的完播率、付费转化和用户留存。它能不能走通"老 IP 加明星授权脸加电影级 AI 画面"这条路，还需要实际数据来回答。但即便它成功了，也只能证明这个极其特殊的组合可以成立，而不能替那十二万部既没有强 IP 、也没有明星面孔的普通 AI 短剧背书。
What will ultimately determine the fate of the AI version ofThe Left Earis not the buzz around its trailer but itscompletion rate, pay-to-view conversion, and user retentionafter the full release goes live. Whether the formula of "legacy IP plus celebrity-licensed faces plus cinema-grade AI visuals" can work still needs real data to answer. But even if it succeeds, it can only prove that this highly particular combination is viable—it cannot vouch for the 122,000 ordinary AI micro-dramas that have neither a strong IP nor a famous face.
行业真正应当警惕的，是一种更隐蔽的路径依赖。因为 AI 让生产几乎没有门槛，平台和制作方持续向一个并不买单的市场倾倒海量内容，再用供给端的繁荣数据安慰自己"大势仍在"。十二万部 AI 短剧里，有多少是因为确信观众会看才被做出来的，又有多少仅仅是因为做它们已经不花什么钱？当“做出来”的成本趋近于零，“能做”和“该做”就极容易被混为一谈。
What the industry should truly be wary of is a more insidious form of path dependence. Because AI has reduced the production threshold to near zero, platforms and production companies keep flooding a market that is not buying with massive volumes of content, then console themselves with supply-side prosperity metrics that "the trend is still going strong." Of those 122,000 AI micro-dramas, how many were made because someone genuinely believed audiences would watch—and how many simply because making them costs next to nothing? When the cost of “getting something made” approaches zero, “can make” and “should make” become dangerously easy to conflate.
1250 倍的落差不是等待技术去填平的沟壑。市场已经在此一锤定音：我愿意花时间看一个人演的故事，暂时不太愿意花同样的时间看算法算出来的结果。 这话也许不好听，但绝对算得上是清晰的信号。行业可以继续赌观众会改变主意，不过在此之前，至少应当认真考虑一种可能性——观众也许不需要被说服，他们早已明白自己想要什么。
A 1,250-fold gap is not a chasm waiting for technology to bridge. The market has already spoken, loud and clear: I am willing to spend my time watching a story performed by a person; for now, I am far less willing to spend the same time on a result computed by an algorithm.That may not be pleasant to hear, but it is an unambiguously clear signal. The industry can keep betting that audiences will change their minds—but before that happens, it should at least seriously consider one possibility: perhaps audiences do not need to be convinced. They already know what they want.
10个月宝宝每天需要喝多少奶粉？
