I designed and led a 0-to-1 AI product that lets fans keep living inside the shows they love: talking with the characters, unlocking new chapters and shaping their own endings.
Opening films for two original stories built on the platform
Tencent Video runs on premium IP and membership revenue, much like Netflix. Every show has a finale, but fans' feelings carry on, and they take them to Weibo and Xiaohongshu. In 2024 we asked: beyond watching, what can a streaming platform give its fans?
Illustrative. Our mission: use AI to fill this gap and extend the life of every IP.
Character: AI agents that feel real. Plot: open storylines that break past the script.
Writers own the world and the key beats. Users choose the path.
We launched with characters from the platform's top shows.
PMs and scriptwriters set characters, rules and quality bars together.
Every capability in the foundation is a reusable prompt module, so a new IP never starts from zero.
A story feed where each world is one tap away, and recommendation puts the right story in front of the right fan.
A real session in Galaxy Hunter







We profiled the fans chatting with Lost You Forever characters to tune each agent's voice, timing and story hooks.
Approximate, from Tencent Video age tags of AI chat users (97.7% tagged).
Share of users by number of private-chat turns with the Xiangliu agent.
Chats per hour of day, thousands. Hover a bar for its value.
Most frequent messages, from chat keyword analysis.
Every two-week sprint was judged on two numbers.
Plots moved from one main line to many branches, driven by stats. Lighter presets let user choices decide where the story goes.
Three levels of AI continuation, each handing AI a bigger share of the writing while creators keep the frame.
Female-oriented IPs like Lost You Forever have huge fan-fiction communities. AI tools let fans build new episodes around official characters, and players turn into creators.
Before: script → mechanics → art → audio → QA
Building a story end to end: idea, branching canvas, AI-written scenes, art and music.
18.37% is post-finale retention: unique users still chatting after the finale ÷ all chat users, across two IPs, through December.
Users stayed because chapters, branches and outcomes gave conversation a purpose.
Story quality has to be defined by people who understand stories, then encoded into prompts and evaluation.
Next time I'd build the automated evaluation loop before launch, not after.