内容来源：https://www.wired.com/story/made-in-china-the-chinese-creator-who-imitates-ai-slop/

内容总结：

近日，一段模仿人工智能生成视频风格的创意短片在国内外社交平台引发热议。中国青年创作者穆天澜通过对AI视频缺陷的精准模仿，以幽默形式展现人类表演的不可替代性，同时引发行业对技术发展的思考。

这段发布于2024年7月的短视频中，两位看似即将发生冲突的男子突然跳起机械感十足的探戈，随即凭空变出酒杯和面条。该片段因完美复刻AI视频的典型特征——僵硬的肢体动作、空洞的眼神和违反常理的剧情展开，在海外平台获得超1100万播放量，但创作者穆天澜直至接受采访时才知晓其作品已在国际走红。

29岁的穆天澜毕业于影视相关专业，自2019年起在国内社交平台创作喜剧短剧。为精准呈现AI视频的“诡异感”，他系统研究了AI生成内容的常见缺陷：物体用途错乱（如衣架突然从惩戒工具变为晾衣工具）、画面连续性缺失（由两位演员交替饰演同一角色），以及最核心的“飘忽眼神”——“AI的视线总在游移，要演得不像人类才能传神”。

值得注意的是，这支爆火视频为他带来了国内某AI公司的商业合作机会。但在为赞助方制作的视频中，甲方最终选择了嵌入AI生成画面的版本，这个选择让穆天澜感到些许讽刺：“这似乎正在抢夺人类演员的工作机会。”

随着OpenAI发布新一代视频生成模型Sora，穆天澜近期推出了新的模仿作品。他发现新模型虽大幅提升了画面真实感，但仍存在夸张失真的笑声、违反物理规律的头发运动等细节破绽。不过他也坦言：“按照技术发展速度，明年此时我们可能已无槽可模仿——因为AI将无限接近人类表现。”

面对“AI会否取代演员”的提问，这位怀揣奥斯卡梦想的创作者流露出复杂心境：作为导演，他看好AI赋能影视制作的潜力；作为演员，他已在现实中感受到就业压力。尽管通过模仿AI视频获得关注，穆天澜始终坚持在自主创作中展现纯人工表演，其微信头像仍保留着与奥斯卡奖杯合影的生日蛋糕——这个细节或许正揭示着在技术洪流中，人类创作者对艺术本真的执着守望。

中文翻译：

常年泡在网上的朋友或许都看过这个源自中国社交媒体的视频：两名看似即将拳脚相向的男子，突然跳起浪漫又略带机械感的探戈，转眼间竟凭空变出酒杯和面条。这段浑然天成的表演看似AI生成，实则是百分百的人类智慧。

视频中端面条的创作者是29岁的中国小伙慕田然。当我联系到他时，这位年轻人对自己模仿AI的短视频在全球的传播范围知之甚少——在X平台，两条转载他视频的推文浏览量合计超1100万，Reddit、TikTok和Instagram上的转发也获赞数万。从未入驻这些平台的慕田然，对自己在海外引发的病毒式传播毫无概念，更不曾接受过国际媒体采访。

虽非唯一尝试模仿AI视频风格的创作者，但慕田然精准抓住了所有精髓：笨拙的肢体动作、放空的面部表情、难以预测的情节走向。无数观众（包括我在内）都震惊于他对AI视频混沌特质的精准复刻。

这位年轻人告诉我，迄今拍摄的六条AI模仿视频仅是他演员生涯的冰山一角。从大学立志成为演员起，他曾在全球最大影视基地横店寻找群演机会，2019年开始在中国社交媒体创作喜剧短剧，如今内容创作已占据他大部分时间。

AI模仿视频的成功为他赢得了国内某生成式AI公司的赞助——8万元人民币（约1.1万美元）酬劳用于制作两条推广该企业视频模型的短剧。这份收入虽不俗，但以其全球影响力而言，本应获得更多机遇。

根据赞助要求，慕田然为每条短剧拍摄了两个版本：含AI生成片段与纯人工表演版本。他私下更期待广告主选择后者，因为那能将人类演技置于聚光灯下，但事与愿违。"这感觉就像AI开始抢演员饭碗了。"他苦笑道。

上周，随着首部AI模仿系列续作的发布，慕田然再次闯入我的视野。新作瞄准OpenAI最新视频生成工具Sora的输出特质，虽模仿痕迹更隐晦，却依然精准捕捉到那种伴随AI技术进阶却始终存在的诡谲感。

"随着AI加速进化，永不停歇的角逐并非人机对抗，而是人类与AI模型开发者之间的博弈。"慕田然洞见道，"我们在戏谑AI的诡异荒诞，而开发者正在修正这些缺陷。你看，今年AI生成的内容已更具人性。"

【人类模仿AI演技手册】
在2024年7月创作首支AI模仿视频前，慕田然系统研究了大量AI视频的共性，着重分析其常见错误并融入剧本构思。

比如AI常误解场景中物体的存在意义：衣架既能悬挂衣物，也可能成为中国家长教训孩子的工具。这种双重属性激发他去年的创作灵感——在假装用衣架教训"儿子"时，男孩短裤离奇消失，而他突然忘情地将衣架挂上短裤。

AI视频的另一常见问题是连续性缺失：物体变形、人物容貌服饰突变。在他最火的视频中，多数观众甚至未察觉同一角色由两位演员交替出演，以此复现AI的不连贯性（慕田然的大学同窗参演其所有作品，这些白天各有本职的伙伴总愿抽空回归他的片场）。

但AI视频最令人不适的，当属人物诡异的面部表情。无数观众惊叹慕田然对空洞眼神的精准把控，他表示这源于大量观摩AI视频及其他创作者模仿后的心得："AI视线总是飘忽不定，此刻凝视这里，下一秒就转向别处。就像我和你对话时，目光可能游移到他人身上——简而言之，只要演出不自然的状态即可。"

上月OpenAI发布Sora后，慕田然在模仿该模型生成的视频时发现难度陡增，但仍捕捉到新瑕疵：除了游移的目光和时间线错乱，Sora生成的人物常伴夸张骇人的大笑，发丝在空气中疯狂颤动。他在新作中复现了这种笑声，再次引发中国粉丝的强烈共鸣。

目前他暂无继续模仿Sora的计划，部分因为视频质量已臻化境， parody变得近乎不可能："待到明年此时，我们或许再无模仿对象。若硬要表演，恐怕只能演回人类自己了。"

【AI会取代演员吗？】
尽管众多评论认为慕田然的视频印证了AI无法取代人类，但他本人对此远未乐观。作为导演和内容创作者，他清醒认识到AI变革电影制作的潜力，预计两年内全球影坛将涌现大量活用AI的导演制片人。

但作为演员，他感到寒意。除社交媒体短剧和电视广告小项目外，慕田然已许久未获真正演出机会。与刻苦的同业竞争已属不易，而今AI演员亦将加入战局。

除赞助内容外，他始终坚持在作品中使用纯人工表演，希望以演技立身。他的微信头像定格在凝视奥斯卡奖杯造型生日蛋糕的瞬间——摘取小金人是他毕生梦想。"最终目标是自编自导自演一部电影，并因此获奖。"慕田然郑重说道。

（本文系Zeyi Yang与Louise Matsakis联合撰写的"中国制造"时事通讯专稿，过往内容可通过此处查阅）

英文来源：

Our fellow terminally online readers probably have seen this video, which originated on Chinese social media. In it, two guys who look at first like they are about to get into a fistfight suddenly break out into a romantic, yet slightly robotic tango dance routine. The next second, they pull a wine glass and a bowl of noodles out of nowhere. It really looks like it’s generated by AI, but it isn’t. It’s 100 percent human intelligence.
I recently talked to Tianran Mu, the 29-year-old Chinese creator behind the video (he’s the one holding the noodles), who said he was largely unaware of how far his AI imitation sketch had traveled. On X, two users who shared his video got over 11 million views combined. Other reposts racked up tens of thousands of likes on Reddit, TikTok, and Instagram. Mu, who has no presence on any of these platforms, had little knowledge about his virality outside China and had never been interviewed by international media.
Mu is not the only comedian who has tried to imitate the style of AI-generated videos, but he really nails all of the elements: The clumsy bodily movements, the spaced-out facial expressions, and the unpredictable plot development. Many viewers, me included, were shocked at how accurately he captured the essence of AI slop videos.
Mu tells me that the half-dozen AI imitation videos he has filmed represent only a small part of his acting career. He has wanted to be an actor since college and spent the summer after his freshman year at Hengdian World Studios—the world’s largest film studio—looking for background acting opportunities. He started making comedy sketches on Chinese social media in 2019, and content creation now takes up most of his time.
The success of his AI imitation videos earned him a sponsorship deal from a Chinese generative AI company, which paid him 80,000 RMB (about $11,000) to produce two more sketches promoting the company’s video model. That’s not a bad gig, but I honestly expected Mu to have received more opportunities through his global virality.
As part of the sponsorship, Mu shot two versions of the sketches, one that embedded AI-generated footage and one without it. He was secretly hoping that the advertiser would choose the latter, because it showcases human acting skills front and center. But the advertiser chose the one with the AI. “That kind of feels like it’s starting to steal jobs from human actors, doesn’t it?” Mu says.
Mu popped up on my timeline again last week when he released a sequel to his first AI imitation series, this time mimicking the videos created by Sora, OpenAI’s latest generative video tool. His new video is much more subtle but still manages to nail that unexplainably unsettling feeling that has endured even as AI videos become more advanced.
Mu says there is a perpetual battle underway as AI accelerates, but it’s not man versus machine. Rather, the clash is between humans and other humans who make AI models, and each side is constantly trying to one up the other. “We’re poking fun at some of AI’s flaws, its eeriness and absurdity, but the AI creators are probably improving those, too. You see, this year’s AI already looks much more human,” Mu says.
How to Act Like AI
Before he made his first AI imitation sketch in July 2024, Mu watched a lot of AI slop videos to study their common traits. He wanted to understand the kinds of mistakes AI often makes and then re-create them in his own scripts.
For example, when an object appears in the frame, AI often misunderstands its purpose for being there. For example, a hanger can be used to hang clothes, but it’s also often the weapon of choice when parents in China physically punish their children. That dual use inspired another one of Mu’s videos last year, where midway through pretending to hit his “son” with a hanger, the boy’s shorts mysteriously come off, and Mu looks like he suddenly forgot what he’s doing and decided to hang up the shorts instead.
Another common issue in AI-generated videos is the lack of continuity—objects change shape, and humans change their appearances or outfits. In Mu’s most popular video, many people didn’t even notice that two different actors played the same role to imitate that same inconsistency. (Mu’s college friends act in all of his videos. They have day jobs but are willing to come back to star in his sketches.)
But perhaps the most eerie aspect of many AI-generated videos is the strange expressions on people’s faces. I saw lots of people commenting on how Mu mastered the empty focus of his eyes. He says that’s indeed another observation he made after watching copious amounts of AI slop and seeing how other creators mimicked it. “The AI’s gaze tends to wander—one moment it’s looking here, the next it’s looking there. When I’m talking to you, I might actually be looking at someone else,” he says. “Simply put, it’s unnatural, so just act unnatural.”
Late last month OpenAI released Sora, a social media app that allows users to generate and share short AI videos. Mu says that when he set out to parody the clips people were making with Sora, he found it a lot harder to nitpick their flaws but still discovered a few. On top of the wandering gaze and timeline inconsistencies, Sora-generated human characters often laugh in an exaggerated, offputting way, with their hair shaking frantically in the air. Mu imitated that laugh in the new video he released last week, and once again, it really struck a chord with his Chinese followers.
Mu says he doesn’t have any immediate plans to make another Sora imitation, partially because the video quality has gotten so good that it’s now almost impossible to create parodies. “By this time next year, we might honestly have nothing left to imitate. If I try to act it, I’d just be acting like a human,” he says.
Will AI Replace Actors?
Many commenters have said that Mu’s videos reassure them that AI can't replace people. But Mu himself is a lot less certain about that. As a director and content creator, he says he definitely sees the potential for AI to transform moviemaking. He thinks it’s likely that the world will see movie directors and producers using AI liberally within the next two years.
But as an actor, he’s scared. Outside of his social media sketches and small gigs doing TV commercials, Mu hasn’t gotten any real acting roles in a while. It’s already hard to compete with other hardworking actors, and soon AI actors may be coming to take his job, too.
Mu hasn’t used any AI-generated footage in his own content (except for the sponsored posts) because he wants to be known for his acting skills. His WeChat avatar is him looking at a birthday cake with an Oscar trophy replica on it, because winning an Oscar is the dream of his life. ”My ultimate goal is to write, direct, and star in my own movie—and win an award for it,” Mu says.
This is an edition of Zeyi Yang and Louise Matsakis’ Made in China newsletter. Read previous newsletters here.