Xiaohongshu's Youngest Business Unit Head Strikes Out on Her Own, Using AI to Reimagine "Grass-Seeding"丨début

@Chen Zhiyan**

début is our new column. The word originally refers to the beginning of an action, work, or speech, and also to a person's first appearance on stage. début will introduce you to a group of entrepreneurs you've never met before.

We promise this is a completely free, public-interest column. Welcome to début, where you make your first appearance in the public eye.

Feel free to reach out via message through our official account backend, or email us at info@elsewhere.news.

Below is the second installment of this column.

A Ban, once the youngest head of an innovative business unit at Xiaohongshu, has embarked on a new venture.

Dreamova — a native AI platform attempting to use multimodal video and Proactive Agent to let users "experience their desired lifestyle with one click" — has completed its angel round of funding.

From ByteDance and Kuaishou to Xiaohongshu, over ten years at major tech companies, A Ban has walked the complete path "from video tooling, to tool-based community, to community commerce."

In 2016 at ByteDance, A Ban was among the first product managers pioneering innovative businesses, building a filming tool with millions of DAU and collaborating with the AI Lab to hone ByteDance's first-generation computer vision and image capabilities that became breakout hits. He then moved to Kuaishou, where he built MV Master — from 2018 to 2020, the largest self-growing hit album tool globally, achieving tens of millions of DAU and completing the validation of tools evolving into community and commercialization. In 2020, A Ban joined Xiaohongshu, becoming its youngest head of innovative business.

During nearly five years at Xiaohongshu, A Ban reported directly to Charlwin Mao for more than three years. successively serving as head of innovative business, responsible for standalone video business, vertical interest communities, and standalone community commerce, driving the upgrade from image-text to video ecosystem, demographic and category expansion, community commerce development, and core organizational strategic restructuring.

In the second half of 2025, A Ban left Xiaohongshu. The reason: he had spotted a crack in a new world — users were no longer satisfied with "watching other people's lives"; what they craved was "confirmation through immersion."

And so, he founded Dreamova.

Through Dreamova, users can directly transform content they've seen on Instagram or Pinterest into a personally experienced video, with the freedom to modify it until they confirm whether it truly suits them.

Beyond this, Dreamova also supports a 7x24 online "avatar Agent" that acts on behalf of users to browse the entire web for inspiration. As the Agent continuously learns the user's aesthetic preferences, it prepares various "ideal moments" and pushes them to the user.

"The user is no longer a spectator, but the creator and confirmer of their own life."

Currently, the Dreamova team has only five people, with backgrounds from ByteDance, Kuaishou, and BAAI. Among them, its model lead Li Xiang is the youngest associate professor and doctoral advisor at Nankai University, with over 20,000 academic citations.

They are attempting to start from the US market, using AI to shatter the old world built on "imagination" and "spectatorship."

Watch at the Last Stop, Experience at the Next

elsewhere: What is Dreamova?

A Ban: **The next-generation "experiential" lifestyle platform.

If Xiaohongshu's core is "watching other people's lives," then I want users to see the life they aspire to and first experience it inside Dreamova.

Through multimodal generation technology, we transform life scenarios that previously required "mental imagination" — such as wearing a certain piece of clothing to a particular occasion, or moving into a home of a certain style — into an "experience video" that users can generate with one click.

The user is no longer a spectator, but the creator and confirmer of their own life.

elsewhere: "Fit visualization," "real home scenes" — these have long been achievable on e-commerce platforms, right?

A Ban: **Traditional tools cut into the "fitting room" segment downstream in e-commerce, solving the matching problem for specific SKUs. Dreamova wants to cut into the most upstream "discovery" segment. Users aren't trying because they want to buy; they're inspired by a lifestyle they've seen and want to confirm whether it suits them.

Meanwhile, videos generated on Dreamova are linked to real scenes and real products. What we offer is a new logic of "experience first, then choose, then transact."

elsewhere: Is watching your own virtual experience video really more interesting than watching other people's real lives?

A Ban: **This is actually based on my observation of Gen Z. What I see is that they are no longer satisfied with watching others — they are constantly trying to project others' lifestyles onto themselves.

A phenomenon has become very clear within communities: users are shifting from "spectating" to "projecting." Under a fashion blogger's post, users ask in comments: "I'm 165cm, 128 pounds — what would this look like on me?" Under home decor sharing posts, there are many questions like "Would this sofa make my small apartment look cramped?" On the surface, users are consuming content; in essence, they are using others' content to complete "projection" of themselves.

This trend also shows up on the commerce side. Starting in 2025, a wave of top merchants and buyers who broke out are no longer polished, distant showcases — they are more like the users' own bloggers: curvy, pear-shaped, petite. More critically, the core demographic profile behind them is uniformly Gen Z. Users don't simply want to see more real people; they are looking for a sample that resembles themselves more, that can help them project.

But obviously, this kind of projection relying entirely on "mental imagination" is too painful. I realized that the qualitative change of the next era lies not in providing more content, but in letting users more easily complete "project, adjust, confirm." Dreamova wants to take the lifestyle you like and generate your own experience video with one click.

elsewhere: So the virtual videos Dreamova generates are also adjustable?

A Ban: **Yes, projectability and adjustability are what we believe are key to achieving the Aha Moment.

Projectability is essentially generation centered on me. The video isn't generated aimlessly, nor is "looking good" enough — it must satisfy your aesthetic preferences and hold up under your real-life constraints (such as body type, apartment layout). Adjustability is "fine-tune based on your own taste + generate to verify." Users can achieve "click anywhere to change anywhere" through the Remix function — switch scenes, colors, materials, lengths. This "continuous experience to confirmation" is what matters.

elsewhere: How do you ensure these videos aren't one-time gimmicks, but can drive sustained retention?

A Ban: **What truly determines retention isn't just "the first generation is stunning," but whether it increasingly understands you and can continuously help you do better lifestyle experiences and confirmation.

Dreamova's core moat isn't the video itself, but the user's Taste-Decision Graph.

Every real choice users make in generation, Remix, Pick, Nope — all continuously accumulate Life-Long Context in lifestyle scenarios. Eventually this precipitates into the Taste-Decision Graph. It determines that the next generation is more accurate, experiences hit faster, and ultimately can become the user's true "avatar Agent."

Many AI products ask you 20 questions upfront — this is helpless, but true preferences aren't obtained by filling out questionnaires. They grow out of real user interaction in authentic demand scenarios. Dreamova is positioned in real lifestyle scenario experiences, where natural user interaction precipitates Context. This is the core of what can continuously drive retention.

elsewhere: There's also a Proactive Agent?

A Ban: **This Agent is the user's "7 x 24h dedicated lifestyle experience officer." It helps users browse the entire web, surf Instagram and Pinterest, proactively prepare inspiration guides for users, and directly generate experience videos belonging to them, which can be seen in Dreamova's My World section.

And the strong interaction and strong calibration that users engage in during this process form the Taste-Decision Graph moat, which no general-purpose large model or pure tool product can replicate in the short term, and which is also the core of the recommendation and distribution engine in the AI era.

elsewhere: That token consumption would still be pretty staggering.

A Ban: **If we were just making AI-generated video as digital snack food, the compute cost would be almost insurmountable. But Dreamova cuts into "useful" lifestyle experiences. Only when a product truly participates in users' life decisions does it possess extremely high commercial value-added space, enough to offset early compute leverage.

The current plan is to start from the US market. From day one, there was no intention to rely on subscription model as the core model to hard-carry compute costs. By connecting to product databases from Amazon, Shopify, and independent sites, we've built a complete Affiliate chain. If users make purchases after experiencing videos, the commission sharing behind it can fully cover the compute costs generated, and by our calculations there would even be surplus.

Community Needs Soul

elsewhere: You left Xiaohongshu in 2025 — did it take the better part of a year from idea to execution for Dreamova?

A Ban: **There was a pivot in between, but the endgame didn't change. When I first started the business, I first made a "consumption decision Agent" — using "LLM + mixed clips as primary, multimodal generation as supplementary" to help users summarize shopping guides. This was the optimal entry point at a stage when multimodal generation capabilities weren't mature. But in execution, we found a problem: the path from Chatbot to community is counterintuitive — users left after using it, making retention difficult.

At that time, I judged that multimodal and Agent capabilities were approaching an inflection point. It wasn't until the beginning of this year that a qualitative change occurred, and we found a superior product form.

elsewhere: It seems Dreamova still has very strong tool attributes — users come with strong intent rather than content consumption willingness. How does it form community?

A Ban: **No one can say on Day 1 that what they're building is community. Community is never "positioned" into existence; it grows out "ex post." But we know what kind of soil makes community more likely to grow.

In its early years, Xiaohongshu didn't know it was a community either. It was originally just a PDF of shopping guides. The underlying logic of community formation has only two parts: one, whether you've chosen the right demographic with momentum; two, whether you've chosen the right category with interactive value.

elsewhere: How did you choose this time?

A Ban: **On demographics, skip the post-80s and post-90s, go straight to post-00s. Post-80s and post-90s are "digital natives" of mobile internet. Their stage of coming online was one of content scarcity, so they are now too tightly bound to existing paths of scrolling feeds and watching blogger recommendations. But post-00s are different. They are aesthetically confident, with strong personalized expression. From their first time online, they haven't lacked content. What they truly lack is "sense of experience, sense of confirmation." Not satisfied with being spectators, what they want is "more experience, more confirmation, fewer mistakes."

On category, from lifestyle categories we focus on visually aesthetic-driven scenarios. The first choices are fashion and home. Because these two categories naturally possess extremely strong visual impact and high-frequency "hallucination scenarios." For example, when you see a dress on Instagram, you fantasize about wearing it to dance; when you see a mid-century modern sofa, you fantasize about placing it in your living room. This is "life at a 45-degree angle" — a life you can realistically obtain by stretching a bit, not pure fictional fantasy like Miracle Nikki.

elsewhere: Over the past year, AI entrepreneurship and funding, especially in the agent direction, has been extremely abundant. Dreamova plans to launch in June — are you worried it's a bit late?

A Ban: **In the mobile internet era, being half a step early could mean the difference between life and death. But in the AI era, coming out early isn't necessarily right.

I think 2026 is a critical inflection point: multimodal generation capabilities become commercially usable, physical simulation and subject consistency issues are solved at scale, and Agent proactive execution capabilities also undergo qualitative change. Meanwhile, major companies must protect their existing interests. It's a gap where the giants haven't caught their breath and the technology can just support new forms.

elsewhere: Then if Xiaohongshu and ByteDance follow suit, do you think you'd still have a chance?

A Ban: **I don't think this is a question of "whether the giants will do it," but that they would find it difficult to do this with their existing distribution and commercial logic.

Xiaohongshu's core is built on UGC content, influencer recommendations, and community trust — essentially completing decisions through "others influencing me." Its commercial infrastructure, distribution mechanism, and organizational capabilities all grew around this logic. Dreamova is "user-centered confirmation," not others-centered influence. This isn't adding a feature; it's reconstructing the system.

Douyin is purely efficiency-driven kill-time logic, while our logic is "user-centered confirmation" — the underlying distribution logic is completely opposite. The giants were too strong in the last era, and this反而 becomes their shackle for entering the next generation.

So I'm not betting that they won't do it, but that the new paradigm requires a new starting point. It also requires a team completely matched to this task. We only have five people, but we are assembled around "multimodal technology + content community/commerce cognition + global growth commercialization capabilities."

elsewhere: You spent several years as a reserve executive at Xiaohongshu, even at Charlwin Mao-minus-one level. Was leaving to start a business a difficult decision?

A Ban: **Before I joined, he (referring to Charlwin Mao) asked me what I ultimately wanted to do. I was very honest with him at the time — I told him I would definitely start a business in four years. Two years ago, during Xiaohongshu's management training co-creation, I also openly shared where my heart truly lay.

Over the past 10 years at major companies doing 0-to-1, I have indeed achieved results, but I always saw myself as an entrepreneur "borrowing the path of major companies."

Actually, I started businesses from university. To my past work doing 0-to-1 innovation at platforms like ByteDance, Kuaishou, and Xiaohongshu, along the path of "tool-community-community commerce" — I made some products explode, and also personally shut down quite a few. Entrepreneurship has always been a certainty for me; it was only a matter of timing, based on what opportunity, doing what.

The reason I chose now is that I believe every wave of technological qualitative change rewrites a generation's lifestyle experience.

In the PC era, information digitization and search moved the transaction experience from "offline malls" to "shelves," with platforms solving for abundance and efficiency. In the mobile internet era, HD cameras + recommendation algorithms pushed the experience from "looking at shelves" to "watching other people's lives," with platforms solving for inspiration discovery and planting seeds. And in the AI era, another experience leap occurs, transforming "distributing reality" into "generating experience" — that is, taking "watching other people's lives" and for the first time bringing it to "letting me experience it in advance."

Dreamova's opportunity lies in seeing this technological inflection point.

elsewhere: You've worked at both ByteDance and Xiaohongshu — did they feel very different?

A Ban: **ByteDance is like a massive experimental field. Early strategy went from content to community to social, conducting networked, large-scale trial and error. In ByteDance's logic, what matters most is rapid validation and data-driven decision-making. What I learned there was that extremely formidable product and growth methodology.

What Xiaohongshu gave me most strongly was understanding of community, and judgment of self-opportunity. Community needs soul. This manifests in understanding users to leading them, not just efficiency itself. Of the past 10 years, I spent the longest at Xiaohongshu because I am fundamentally aligned with Xiaohongshu's Taste, but I also know how formidable ByteDance and Kuaishou's war-fighting methodology is.

Dreamova's core logic is the same: based on understanding of content/community value, using the hardest technology and most efficient growth methods, to serve a group of post-00s players with the strongest aesthetic presets.

Cover image source: Berthe Morisot, $2, 1876, Museo Nacional Thyssen-Bornemisza