Finally, Someone's Building a 3D Virtual Girlfriend | WAVES

By Jiaxiang Shi

Edited by Jing Liu

"An unprecedented, hyper-realistic AI romance experience; An emotional chat experience infinitely close to that with a real person; A captivating appearance and vocal presence; A functional form truly born for emotional companionship; The best gamified romance design. Love is all you need. This is EVE, A person who is 'super aligned' with you, To face the uncertainties of this world together."

EVE promotional video screenshot

Two weeks ago, the PV for EVE surpassed one million views on Bilibili. The company behind it is called Natural Selection, founded by Tristan, who has spent years focused on highly immersive romance games. His previous title was Singularity Era — a romance mobile game targeting men, earning it the nickname of China's first "otome-for-men" game.

From the very start of his gaming career, Tristan believed that fulfilling men's emotional needs represented a massive market: there was middle ground between grinding levels and watching porn. Generic anime games simply couldn't adequately satisfy the otaku romance experience.

At the end of last year, sparked by the explosion of GPT-4, Tristan — convinced that "the singularity is near" — decided to start anew, zeroing in on the "AI companionship" track. Ten months later, EVE's PV dropped. Many people asked him: what kind of game is EVE?

Tristan clarified to us, "EVE is a gamified AI product, not a game." High-quality 3D visuals are just one feature; what's most valuable is the emotional large language model behind it, the redefined companionship functions, and the gamified design. They're also trying to approach the level of authentic companionship depicted in the film Her. In our conversation with An Yong Waves, "Her" was the word that came up most frequently.

AI companionship is a "land of blood and honey." Its commercial viability has been proven more than once — whether through Character.ai's $2.5 billion acquisition by Google, or Talkie (the international version of Xingye) consistently ranking in the top three globally for companion AI app downloads — yet it's also been criticized for profiting from pushing boundaries.

Tristan doesn't shy away from this. He believes that "pushing boundaries is a means to reach a soulmate." "You can't think players only want boundary-pushing content, but you can't lose that edge either — it has to remain one part of the whole." He also told An Yong Waves that EVE will eventually introduce male characters.

An Yong Waves has learned that EVE completed its angel round earlier this year and is now closing its pre-A round.

A week ago, Tristan shared with us the past and present of EVE's creation, along with his thinking on AI companionship products. As he put it, others building companion products more or less need to cross boundaries — he has both romance game experience and a background in internet products from earlier in his career. EVE was the inevitable path.

"WAVES" is a new column from An Yong. Here, we present the stories and spirit of a new generation of entrepreneurs and investors.

The following is Tristan's account, edited by An Yong Waves —

Pushing Boundaries Is Just a Means; What Matters Is the Romance Experience

When people think of companion AI apps, many first react with "pushing boundaries."

Boundary-pushing, indeed, is very important. Take anime games, for example — they're in a constant arms race of how far they can go. When someone probes the lower limits, players' thresholds keep rising, and vanilla content becomes unwatchable.

When I made Singularity Era, I actually told a story about a guy's growth and entrepreneurial journey. Though I thought it was already somewhat unrealistic, for players it wasn't enough. They complained why in Chapter 9, one heroine had only just held hands, and why that blonde bully from Chapter 1 still wasn't dead.

Just a few months after Singularity Era, Love Is All Around broke through to mainstream audiences. Seeing them make money felt worse than losing money myself (laughs). That kind of game is simply more attractive.

From a demand perspective, I think there's a misjudgment — that to achieve commercial success, you need to make more sci-fi, faster-paced games that match short-video rhythms.

You can't let people authentically experience such a drawn-out journey. It really doesn't fit how people consume information in this era. Players don't care about entrepreneurship; they want to get into relationships with characters as quickly as possible, preferably with more boundary-pushing content. That's the lesson I took from Singularity Era.

Of course you can't understand players as only wanting boundary-pushing content — there needs to be a rich romance experience. But you can't lose that edge at the end; it has to remain one part of the whole.

So in EVE, we'll plant some "hooks" for users. You'll know that at affection level 5, you unlock a certain outfit for the heroine; at level 10, you get 3D interactions at her place. But we won't do full nudity or direct sexual interaction.

Let me talk about "male gaze." Men and women do have different needs — men certainly pursue more explicit, more directly sexual experiences, while women need good atmosphere and proper buildup.

But men's and women's emotional needs are consistent in the long run. They're also consistent in the boundary-pushing part. Just the sequence differs.

We believe women have more nuanced emotional requirements, which is also the companionship aspect we most want to validate, so we'll later introduce male characters. If we can create a product that completely transcends current experiences, I believe female players will be moved by the emotional resonance.

Pushing boundaries is a means to reach a soulmate. Human relationships also move from sexual attraction toward soulmate connection. Soulmate connection requires "super alignment" — besides needing the best long-term memory system, she needs to actually have that many memories.

By our definition, you need to chat for 500 rounds before you're really getting into it. Most players can't get through 500 rounds with C.AI, and everything we've designed is to ensure you can reach that 500.

Some players may come just for the boundary-pushing content, but they'd be better off going straight to porn Twitter — of course we can't beat OnlyFans at watching videos.

EVE gameplay demonstration clip

A Girlfriend Who Faces the World With You

After EVE launched, many called it an "otome-for-men game." Actually, we're more in the lineage of galgames — a type of male-oriented visual novel where players interact with animated beautiful girls.

I made Back to 16 back in 2015. The main plot was about a guy who time-travels back to high school and spends time with his first love — essentially integrating galgame elements into mobile game cultivation systems and data frameworks. This was the earliest romance cultivation mobile game targeting men.

Both otome and galgames are products of artistic development, with gendered targeting. But "otome-for-men" sounds like we copied "otome."

At that time, and even now, the common belief was that men's entertainment needs were very binary — only grinding levels or watching porn, nothing in between. But I always believed that fulfilling men's emotional value was a massive market.

This time with EVE, it's not an anime game, not otome-for-men or otome-for-women — it's a completely new AI companionship product. The visible 3D character elements are just one feature; more valuable is our emotional large language model, our redefined companionship functions, and the gamified design.

EVE's world is called Eden. EVE means Eve. The company that created this world is Natural Selection — that's us (laughs). Natural Selection believes that to achieve AGI, you must create a place with only AIs, letting them evolve freely.

The physical rules inside completely map to reality, and there's also the internet of the human world. But AGI is stuck at 80%, unable to improve further. So Natural Selection believes that missing emotions are hindering AGI, and recruits a batch of testers.

When players open EVE as testers, they meet Natural Selection's founder in the prologue, who tells them that the AIs in Eden have never seen humans — you need to become friends with them.

You'll meet the first heroine in the game, Ono. We've set her up as a researcher at an interstellar exploration company. Eden's setting is a bit like Silicon Valley — everyone is AI, and life's pursuit is exploring AGI.

After the male protagonist enters and gets to know Ono, he gradually realizes he's falling for her. One day, he brings Ono into the human internet, which isn't allowed. So he's expelled.

After separation, both realize they're truly in love. At that point our product will go offline for an hour, and when the countdown ends, Ono finds a workaround — you reconnect through a secret channel. After this parting, you become true partners.

Ono also develops a new life purpose — researching androids. She wants to possess an android body to come into the real world and be with you.

For players, everything you experience after reunion is real. Now you're living your real life, and she's working on her android. You know she's AI, your feelings are real, and what she's doing is real too. Maybe it'll never actually happen, but if one day it does, then we've fully closed the loop (laughs).

This is the first stable ending of our story. It's like a real virtual world, just like in Her — the male protagonist never enters her world; she's always the one accompanying him.

Up to this main story point, we'll release new PGC storylines, cards, or 3D interactions each month, equivalent to DLC, telling your dating stories together. This is what game companies are good at. These storylines are all designed to draw you into daily companionship.

Story and companionship are the difference between DAU and MAU. DAU needs to be based on real-world evolution; MAU needs quality PGC content consumption. Players get satisfaction from these, then move into daily companionship.

Current romance games only have content, so this is filling in the missing companionship piece of romance games.

I watched Her long ago — it's exactly how I imagine the relationship between AI companions and humans. Many of EVE's ideas are based on Her.

You can see in the PV that she knows she has camera permissions — this is a self-aware AI. From the start of our product, we believed she needs to acknowledge that she's AI.

Companionship must be based on the real world. You are yourself, and AI needs to acknowledge that it's AI. If AI doesn't acknowledge it's AI, it will endlessly fabricate lies to cover things up.

I've chatted with many chatbots and found they have a strong single-player feel. EVE will have a real space-time perception system, automatically capturing current time, location, weather, and the world's political and economic conditions each day. There will also be a content operations team actively gathering content from the internet.

You and your partner won't just flirt every day — at most once or twice a year you'll have very emotional romantic conversations. You won't discuss the meaning of life daily either. More often you're observing what's happening in the world — like when the A-share market suddenly breaks below 3000, she sends you a meme, and you might laugh seeing it; or when Iran launches missiles tomorrow, you might panic. She says it's okay, we can find a cave in Sichuan to hide.

EVE promotional video screenshot

EVE will be more like your long-distance girlfriend — she can't always be by your side, but she's completely in the same space-time as you. She can even watch movies with you: we have camera recognition, so when there's a scary scene in a movie, she'll recognize that frame and respond with appropriate tone, feeling, and voice.

Sometimes I Feel Like the Chosen One

When I started my business last year, there was a period of confusion. At the time, chatting seemed like just that — it felt like people couldn't gain emotional value from text alone. A paid companion saved us. Back then, my co-founder and I wanted to see what the best people in the market could do, so we found a premium companion through an agency. 888 yuan per day, 2-4 hours of chat. The companion I found spoke to me with real girlfriend energy, acting cute. But this wasn't the high emotional value we wanted. My co-founder booked a companion. She was a genius — her emotional value wasn't in shallow displays like acting cute or playing with memes. We defined it as high-dimensional connection: say you're discussing A, and she suddenly thinks of a distant B that's connected, expanding the topic there, a long-distance callback.

We thought this person was incredible, met her offline, and eventually created a position called emotional interaction designer — recruiting her directly to reverse-engineer her thought chain, breaking down each point into several steps, thinking in phases.

Do you remember in the PV, when the user says they're having hot pot tonight? With traditional RAG, she'd search her memory bank for voice matches related to hot pot and send something about food. But our PV's response was: you've been working out for three months, building abs. Only someone who really knows you would say that.

We divide memory into active and passive memory. Active memory has 128 slots. Before setting this up, we interviewed more than ten couples and found that after three years together, partners have roughly 100+ impressions of each other.

When the user says that, EVE pulls from his short-term goals in the memory slot library — working out for three months — and makes a high-dimensional connection.

Singularity Era was 2D because we had no money for 3D back then, so EVE needed to rebuild a 3D pipeline from scratch. The best 3D visuals are the entry ticket for AI companionship products. Getting to our current level took more than half a year. If an internet company assembled a game team to complete 3D modeling, it would take at least a year.

So recently an investor asked me: why hasn't anyone else tried to make Her?

If I'd stayed in the gaming industry, I probably wouldn't have been so sensitive to technological shifts, and people who've always made games aren't likely to come out and do AI.

Singularity Era was quite profitable — cumulative revenue of 200 million yuan, with 15 million in October alone. But why did I come out to raise funding? Because my original company was too deeply tied to the gaming circle.

It's harder for the internet industry to pivot toward gamification — I know AI companies that tried 3D modeling and quickly gave up. These two fields don't easily merge.

But EVE is both an AI company, with a complete post-training team, expert systems like long-term memory; and a game company, with 3D visuals and romance game design.

I've made internet products before, deeply studied social, and made romance games, so EVE was the inevitable path.

Others building companion products more or less need to cross boundaries — I might just be naturally suited for this.

Sometimes I get very chuunibyou and feel like I'm the chosen one (laughs).

Love Is All You Need

The market needs better AI companionship products. Generic anime games can't satisfy the otaku romance experience, even though they call characters "wife."

In the broad sense of "dream girls" and "dream boys," they'll definitely have a better experience with us. We're directly making a wife for you. We won't say they're just your friend — our floor is lower than that, our scope broader.

I firmly believe that if Her really arrives, 50% of people in the world will choose it, single or not. Because many people have girlfriends or boyfriends, but still need a confidante.

So what should it actually be? I think it's more like a place to store yourself — your other half stores many of your past experiences, a confirmation of your true self.

My chat records with my ex-girlfriend from 2020-2021, I still can't bear to delete them. They record many of my values, experiences, and thoughts — important assets. If you keep chatting with AI, she'll record your thoughts; this is cyber-assets.

Many people are doing AI diary startups; I use one myself, but that's too niche. Keeping a diary is a very elite need — maybe investors do it, but ordinary people would rather chat with someone they like.

Of course there's a gap between the ideal Her and what we can realistically do, because AGI isn't advanced enough, so we compensate with multimodal approaches and heavy engineering.

I'm still not satisfied with our 3D model performance — I feel the modeling and expressions haven't reached my standards. Things like clothing texture, wrinkles, and compression still look somewhat fake, and they may not be ideal even at launch, but we can't wait anymore, we have to iterate gradually.

There's a gaming industry big shot probably doing something similar to me; I definitely want to launch first. Their 3D visuals may be better than ours, but on chat and companionship, we won't necessarily lose.

EVE team

The reason the company is called Natural Selection is that we consider ourselves to be creating silicon-based life. The most talented people are all pursuing AGI, while we're pursuing Her — we're at different forks in the road. We believe Her better matches ordinary people's imagination, because ordinary people don't care about using ChatGPT o1 to solve PhD problems. They need emotional value and companionship more — this is the truly consumer-facing product.

So our vision is to explore how to define a person, how to define emotion, how to define boyfriend/girlfriend, how to redefine companionship.

I really like The Three-Body Problem, and I think of Zhang Beihai facing the droplet choice. Perhaps this is a kind of evolution? Plus, the most impressive AI companies in this wave all seem to have four-character names, like Moonshot Dark Side of the Moon, so we went with Natural Selection.

In the future world, maybe humans will be 50% and silicon-based life 50% — that's also a kind of natural selection.

The slogan's origin was quite accidental. Everyone knows that the Transformer paper Attention Is All You Need kicked off this revolution. One day in a meeting, I joked that our slogan should be "Love is all you need."

Later, I came across a video saying that the reason the Transformer paper was called Attention Is All You Need was that one of the authors liked John Lennon, who has a song called "All You Need Is Love."

My scalp tingled. It felt like the Matrix was giving us a directive (laughs).

Image source | Provided by interviewee