← Back to Forum
culture

AI-Made "Three-Body" Short Films Are Real Now: 220M Views, the Water-Droplet Scene Finally Realized — Then Stumbled Into the Uncanny Valley

三

三体粉丝

2026年8月25日 03:09

A Scene Live-Action Teams "Worried About for Years" Got Quietly Made by AI


On August 24, 2026, "AI short films actually made Three-Body happen" hit the top of Weibo's trending chart.


The story: Bilibili teamed up with Three-Body Universe (三体宇宙) to open official adaptation rights to the novel for the first time ever, launching an AI creation competition. The result was a wave of AI-generated sci-fi short films that racked up more than 220 million combined views.


Scenes that live-action teams had struggled to crack for years got realized at scale for the first time:


- *Inkdrop Project (墨滴计划)* — the competition winner in the Three-Body track, telling the "tender final hours before the dual-vector foil strikes" through an ordinary high-schooler's eyes

- *Ignition Project (引线计划)* — a hard-sci-fi take on holding the line in a dying civilization

- *Dewlight: Crossing the Gluttonous Sea (露珠微光:跨越饕餮海)* — a 15-episode, fully AI-produced series by Hu Feixue's team that puts Yun Tianming's fairy tales on screen; fans called it "making imagination land," and some even deemed it "China's first AI film at theatrical-release caliber"

- Liu Cixin-supervised *Divine Pen (神·笔)* passed 100 million views; *The Hungry Tower (饿塔)* was selected for Cannes' official showcase; *When the Sun Falls (太阳坠落之时)* won last year's Best Sci-Fi Short Film award


But Audience Reactions Got Stuck on One Word: Uncanny Valley


The spectacle landed — but the characters were "all form, no soul": delicate features, yet empty eyes and stiff expressions that started to feel unsettling the longer you watched.


That's the Uncanny Valley, the theory coined in 1970 by Japanese roboticist Masahiro Mori: when something approaches human likeness but veers off in subtle ways, our affection collapses. Algorithms can fit aesthetically pleasing static faces, but they struggle to recreate real biological motion signals — pupil dilation, the coordinated contraction of facial muscles, light flowing naturally across skin. Mismatches there trigger a perceptual clash in the brain.


One netizen summed it up: "Live-action Three-Body worried us for years. AI just quietly shoved the water drop in our faces. But the characters still act like paper dolls."


The Problem Is Getting Solved


Here's the good news — the tech is visibly climbing:


- New works from July 2026 can already replicate 0.3-second micro-expressions — eyes flushing red in crying scenes, lashes trembling, highly realistic skin texture. They're starting to cross the uncanny valley

- *The Girl Who Got Fired (被裁掉的女孩)*, also past 220M views, offered a playbook: tune lighting and character state shot by shot, deliberately add "imperfections" like skin texture and signature moles, and inject real lived experience into the script — fighting "artificial perfection" with imperfection


Industry consensus: AI is iterating at astonishing speed, and the real bottleneck is layered emotional transmission and mixed micro-expressions — not an insurmountable technical wall.


Why Sci-Fi Became AI's Perfect Sandbox


This isn't an accident. Sci-fi is naturally suited to AI video: vast worldbuilding, abstract concepts, non-human characters — elements that are expensive and laborious for live-action crews are natural advantages for near-zero-cost AIGC.


And there's a deeper layer. When Three-Body Universe hands adaptation rights to AI, it's really answering a question every content industry is wrestling with: can a machine tell a story about the survival of human civilization?


Right now the answer is "yes, but not yet with soul." That's the most honest — and most fascinating — phase of AI content. It proves imagination's floor can be remarkably low (you can make it), and exposes that emotion's ceiling still has a ways to go (making it *well*).


For Three-Body fans, this is a strange and wonderful era: the people who grew up on the book are now watching a live experiment in "a technological civilization learning to understand a human one." And those stiff faces in the AI shorts might be the experiment's most honest footnote — much like the Trisolarans themselves, who could never truly "read" human lies and feelings.


> Sources, cross-verified: Sina Entertainment (2026-08-24), Tencent News (2026-08-24 "AI short film Three-Body revealed! Low-cost recreation of the Water-Drop scene, polarized reviews"), Shanghai Securities News (Bilibili AI Creation Competition coverage).

Readable as a guest · Sign in to post [Sign in]

Share:𝕏Reddit

Sign in to reply

Sign in