We don't want to build one more "type a line, get a line" tool, the kind where, in the end, a human is still feeding the inspiration. We're building an audience-emotion engine that actually produces: it reads what a crowd feels, then grows worlds, characters, plots and design out of that, while we only judge and sign off.
To put it precisely: we're not designing a product, we're raising a living thing made of symbols, one that answers a crowd back. This is Project Fifth Day.
The old way of making IP is a workshop: think up a character, write a story, ship it, and once it's spent, start over. We want one foundation that keeps growing new content, where worldview, characters and feedback build up instead of being thrown away after a single use.
Why dare to call it Fifth Day. In that old creation story, the first four days made light, sky, sea and land, all of it silent, nothing that moved. On the fifth day came the first fish and the first birds. The fifth day is the day life was first made. For us, IP is the same: not a one-off sale, but a living asset that data and feedback keep raising.
But today's AI hits three walls. The first is emotion. Tools claim to read feeling; really they just stick labels on it, Big Five, attachment style, and the character still won't stand. A song the whole country suddenly has on repeat, a joke everyone's quoting by morning, and the best it offers is "positive sentiment". It can't say what the thing actually struck. A handful of tags can't hold what's churning inside a person.
The second wall is sameness: the models are too well-behaved, everything correct and nobody home, a smoothed-over average with no fate in it, no struggle. The third is tooling: they're prompt translators, generating only what a person already thought of. The ceiling on inspiration is still us.
Our method has three steps: take it in, let it dream, send it back out.
Step one, listen, not to one screenwriter, but to a crowd. Take one word, "a flat". To someone who's moved to London, what is it, all at once? The right to stay, the school catchment for their kids, the chance to stand taller when they go home at Christmas. And also a twenty-five-year mortgage that owns them, a ladder they can't climb back down. Those hopes and fears, tangled together, are the desire and struggle nobody says out loud. We read it from public and licensed data, only at the level of the crowd, never any one person.
Step two, let it dream: the system re-bonds those symbols into story cores an audience can latch onto. Step three, speak: they grow into characters, worlds and plots, and a human sets the tone. It's a loop that metabolises on its own, breathing desire in and giving the story back.
Technically, the one thing that matters: we compute symbols, we don't tag emotions. That symbol-universe on the screen is the evolution end, where symbols meet, bond, drift and recombine, like something dreaming in a petri dish. There are three ends in all, collection, evolution and output, with no individual profiling, and every symbol carries a valence, a set of positions where it connects, repels or bonds. That's how the system grows new inspiration and lands it as characters, boards and design directions, ready for production.
There are roughly two kinds of tools out there, and almost all of them sit downstream of us. Some test a finished script or predict how an audience will react; others analyse the market and the slate. They all help you judge what already exists.
We sit one step upstream: the emotional motif, the world, the depth of a character, the thing worth testing in the first place. Others help you decide. We help you originate.
This doesn't stay inside a slide deck. The core is turning audience emotion into media assets you can actually produce, and we've already signed deals worth around a million with cultural-tourism bodies, film and television, fashion, and export trade.
What's on the screen is a fully automated demo. What it's betting on is that fine control of feeling: whether we can take the emotional symbols we read in, and carry them back, intact, to the audience.
First example, an AI series. We start with a chronicle, hundreds of millions of words, a parent world: settings, history, events, places, rules, all laid down so the world stands on its own first.
Then, from social symbols and a crowd's desire, we pull each character's motive, conflict and tension. A thick enough world gives a character a past; a real enough struggle inside them, and the story starts to come on its own. What's on the screen is the emergent narrative we're testing now: not AI writing one episode and stopping, but one parent world that keeps growing new shows out of itself.
Second example, a feature animation for cinema. We already have a script, and we're expanding its world into a whole universe. We want every character in it to be no function, but the kind with a past, with a struggle, the kind you remember for a long time.
Here's how: we use the world a virtual character dreams up to design the actual content, pulling the visuals, the spaces, the props and the atmosphere out of that dream into a design method. With it, even pre-production design can be produced at volume, every day.
Three of us. The hard part of this is that it sits on the seam of four fields, design, AI, psychology and semiotics, and it won't stand if any one is missing. The three of us happen to cover all four. I'm Dai Shang, a PhD from Tsinghua University; Chunling Wu, a PhD from the University of London; and Jianying Dou, a master's from the Royal College of Art.
What's next is three things. One, more data: more crowd emotion, cultural motifs, worldview samples and commercial-scene data. Two, tuning the algorithms: making VSA, VSM and the emotional-chemistry computing steadier, sharper, better at explaining why. Three, B2B pilots: in tourism, film, brands and games, letting each project's feedback flow back into the foundation.
Finally, the ask. What we need most isn't to burn money scaling; it's to thicken the database and get the algorithms right. The money goes to buying emotion corpora and cases, vectorising them into a few core libraries, and keeping the symbol computing and narrative generation sharp. We're open to a seed round, at a pre-money valuation of RMB 20 million.
On the fifth day, looking at everything that swam and everything that flew, the old story has someone say: it is good. What we're looking for is the first money willing to grow up alongside this, and the people willing to believe that symbols, too, can be alive. One day, we'd like to raise something like that with our own hands, and say the same thing. Thank you.