---
title: "While Others Are Still Asking AI for Images, They Have Started Asking It to Build Worlds"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-03-21T05:11:04+00:00"
canonical: "https://ffcap.cn/en/research/src-20260321-03html"
source: "https://uniqueresearch.substack.com/p/src-20260321-03html"
language: "en"
---

# While Others Are Still Asking AI for Images, They Have Started Asking It to Build Worlds

_Original · Unique Research · 2026-03-21 · Shanghai_

_English edition note: This complete translation preserves the source author's analysis and the full Q&A. The source introduces Zhao Lu as president of Xufeng Technology's China business but later attributes the technology discussion to Yu Hao without explaining the connection; both names are retained rather than silently harmonized. Their spellings and the company name are romanizations, not confirmed official English names. Statements about controllability, transactions, supply-chain advantages, world-model development and prospective robotics cooperation reflect the source's claims and outlook as of publication, not independently verified commercial or robotics outcomes._

Unique Awards · Guest Interview

While Others Are Still Asking AI to Draw Pictures,

They Have Already Started Asking It to Build Worlds

"AI is moving from generating a result to generating a world."

Many people's understanding of AI remains quite shallow.

Enter a prompt and generate an image.

If it looks good, share it. If it does not, revise it again.

There is plenty of excitement, but ultimately this is still AI inside a screen.

Look into industry, however, and the really big change is no longer whether AI can draw. It is:

Can AI understand how humans imagine the real world,

and turn that imagination directly into a

space people can enter, modify, and transact around,

and even use to train robots?

This is no small upgrade.

It is AI moving from content generation toward modeling the physical world.

It was also my strongest impression after reading the interview with Zhao Lu, president of Xufeng Technology's China business:

AI's next tough battle is not about images, but space.

Images can only be seen.

Spaces can be used, transacted around, built, renovated, and understood by robots.

01 It Is Not Just Better Image Generation: AI Is Beginning to Take Charge of Space

Over the past two years, generation has been AI's most visible capability.

Generating text, images, video, and music.

It seems able to generate anything.

But the problem is equally obvious: most generation stops at something that can be viewed.

The results suit sharing, presentation, and content creation.

Enter a real industry value chain, however, and merely being viewable is nowhere near enough.

A home-renovation customer is not simply looking for a beautiful rendering.

What the customer really wants to know is:

• What will this space actually look like?

• Can it be changed?

• Which cabinets, sofas, lights, and flooring does it use?

• Can they be matched directly to the supply chain?

• Can the deal be closed quickly?

• Can the result then be built, delivered, and reused?

At that point, an image is no longer enough.

What industry truly needs has never been a picture,

but a complete system organized around space.

"

AI should not merely generate a static image. It should construct an interactive, perceptible digital world—that is the true meaning of spatial intelligence.

On the surface, this defines spatial intelligence. In reality, it draws a dividing line.

On one side is AI as a visual effect.

On the other is AI as an interface to the real world.

The former can become popular easily.

The latter has a chance to grow into industrial infrastructure.

02 The Real Value Is Not Drawing Convincingly, but Turning Abstract Needs into Images

Why is spatial intelligence worth watching?

Because it did not begin with a technical showpiece. It began with a concrete, commercial pain point:

People's requirements for spaces

are inherently vague.

In home renovation, building materials, and furniture in particular, customers often cannot explain what they want.

They say they want something more sophisticated.

You do not know whether they mean walnut, minimalism, Italian styling, a soft cream-colored look, brighter lighting, more storage, or simply something that does not look like a rental.

They say they want to remodel the kitchen.

Give them a floor plan, and they cannot understand it.

Give them specifications, and those mean even less.

But show them a rendering, and they can immediately say:

"This works."

"That color does not."

"Move the cabinets a little farther left."

"Do not make the island so big."

So what Xufeng Technology first addressed was not the buzzword AI image generation, but a harder problem:

How to take people's vague ideas

and quickly turn them into visual results that can be discussed, modified, and used to close a sale.

Once that works, the value is not merely being cooler. It is being better at converting sales.

"

When we first applied highly controllable, scalable AI image generation commercially, the critical problem we solved was turning abstract requirements into images.

Notice those two terms: highly controllable and scalable.

They already separate many AI products from industrial-grade AI products.

Consumer AI asks whether something can be generated.

Industrial AI asks whether it can be generated reliably within a business process.

The former competes on delight.

The latter competes on completing the loop.

03 While Many AI Companies Still Sell Tools, This One Is Moving into Transactions Around Space

I have always felt that a simple way to judge whether an AI company has substance is to look at where its money actually comes from.

If the revenue logic stops at subscriptions, call volumes, or generation counts, it may still be at the tool layer.

But if it can embed itself in customers' sales, supply chains, and organizational efficiency, the situation is different.

This is what makes Xufeng Technology interesting.

It does not stop at helping you generate a rendering.

It connects another layer behind it:

• Products shown in the rendering can become recommendations.

• Recommended products can connect to real supply chains.

• Beyond the supply chain come orders, delivery, and industrial clusters.

This is no longer a picture.

It is a complete design–recommendation–supply-chain chain.

In other words, the value it captures is not image-generation fees, but transaction value released when AI changes spatial decision-making.

This logic matters greatly.

Many people's understanding of AI commercialization still stops at how much efficiency AI improves.

But efficiency does not necessarily turn into revenue.

The truly large opportunities often appear when AI enters the transaction chain.

This is especially true in home furnishings.

Customers previously struggled to explain what they wanted, understand what they saw, and decide quickly.

When AI shortens that process, upstream traffic, downstream supply chains, and the entire sales cycle change.

This is why Zhao Lu summarizes their path as G to B to C.

They are not directly building a hit App for every consumer.

Instead, they first embed themselves in industries, platforms, leading enterprises, and state-owned and centrally administered enterprise projects, then work downstream toward consumers.

The route is not glamorous, but it is distinctly Chinese and practical.

In China, many truly large businesses do not first explode in the consumer market. They grow out of value chains, organizational systems, and infrastructure.

04 The Company's Real Bet Is Not Renovation, but Space as a Major Entry Point

If you see this company only as doing AI for home renovation, you underestimate it.

Home renovation was merely its first entry point.

The underlying capabilities are spatial understanding, spatial generation, spatial interaction, and the accumulation of spatial data.

What does that mean?

It means it can work on home renovation today, then commercial fit-outs, restaurant chains, age-friendly modifications, hotels, commercial spaces, and eventually robot training and embodied intelligence.

These scenarios look different, but all depend on one thing:

Understanding space

and making it computable.

When discussing technological evolution in the interview, Yu Hao described a clear path:

From using ANN, CNN, GNN, and other techniques to optimize interior-design workflows,

to using the Diffusion model to improve image-generation quality,

to large models providing stronger semantic understanding and tool-calling capabilities,

and finally to 3D spatial generation.

The key is not that model names have changed, but that AI's ability to represent the world is changing.

Previously, AI produced a two-dimensional result.

Now, AI is beginning to produce a three-dimensional environment.

Previously, AI generated images.

Now, it generates scenes people can enter and machines can understand.

"

As technology evolves from two dimensions to three, it no longer merely generates static images. It generates a 3D space in which users can immerse themselves. This is a fundamental experiential upgrade from images to spaces.

At this point, many readers may think: isn't that just a more advanced rendering?

No.

A rendering is fundamentally about presentation.

Spatial intelligence is fundamentally about modeling.

Presentation can affect how people feel.

Modeling can affect every subsequent process.

Once a space becomes digital, structured, and interactive, its possibilities change completely.

05 The Bigger Ambition Is Not in Design, but in Robotics

The easiest part of this interview to overlook—and perhaps the most valuable—is not home renovation or cooperation with state-owned enterprises, but the next step the company sees:

World models and embodied intelligence.

Why does this matter?

Because spatial data is naturally one of the fuels robots most need.

Why have robots so often looked intelligent but acted clumsily?

An important reason is that their understanding of real environments remains insufficient.

• They know actions but do not understand the scene well enough.

• They recognize objects but do not understand spatial relationships well enough.

• They plan paths but still fail easily in complex reality.

If you have accumulated spatial data across interior settings over time, established 3D scenes, materials, object relationships, and usage logic, and can eventually simulate human behavior within spaces, the value of that work for robot training rises quickly.

In other words:

Today, you are doing AI design;

tomorrow, you may be feeding worlds to robots.

This is why spatial intelligence is a much bigger concept than a design tool.

Its endpoint is not saving a few designer-hours.

It could become a new foundational layer for digitizing the physical world.

Zhao Lu said the company is developing a world model for interior scenes and discussing cooperation with multiple robotics companies and embodied-intelligence research institutions.

It sounds futuristic, but it is already a concrete strategic move.

When robots truly enter homes, shopping centers, hotels, offices, and eldercare spaces, what they most lack is not an individual action model. It is an understanding of environments that can be trained and generalized.

Spatial intelligence addresses precisely that gap.

06 Why Chinese Companies May Have More Opportunity Here

I increasingly feel that comparing model parameters alone is no longer particularly interesting.

What is beginning to differentiate companies is who is closest to real use cases, who can obtain data at scale, and who can integrate technology into industrial workflows.

Chinese companies have an often-underestimated advantage here:

Dense use cases and complete supply chains,

large renovation needs and large deployment scales.

In home furnishings, building materials, age-friendly adaptation, urban renewal, and chain-store renovations, China has more than a market. It has enormous, diverse, high-frequency real-world samples.

"

If China begins advancing a field at scale, its data volume and application scale will be the largest in the world. That gives us confidence in targeting urban and everyday-life scenarios.

This is not just saying there is a market.

At a deeper level:

When AI competition enters the physical world,

data is no longer only web pages, text, and images.

It increasingly becomes real-world data on spaces, workflows, construction, delivery, and operations—

the data of reality.

These are precisely the areas where China can most readily develop scale advantages.

Looking one step further, this could combine with China's supply chains to create a new form of international expansion.

Previously, taking Chinese supply chains overseas mostly meant selling goods.

In the future, it may mean:

Chinese manufacturing capabilities + AI spatial understanding + global demand for transforming spaces

You are not merely selling cabinets, building materials, or plans.

You are selling a complete capability from imagining a space to realizing it.

"

When AI can accurately understand how humans imagine spaces and make those visions concrete in 3D scenes, this revolution from images to spaces will not only change design. It will also take the strengths of China's supply chains worldwide.

This is more than a vision.

It reveals a new logic for international expansion:

Not selling Chinese goods around the world,

but packaging China's combined spatial and supply-chain capabilities

into a new generation of solutions.

07 A Final Word: AI Is Moving from a Software Brain to a World Brain

My strongest impression after reading these interviews was:

We may have underestimated how quickly AI is moving from the digital world into the physical one.

A few years ago, AI was best at processing information.

Writing copy, translating, generating content, and finding answers.

All of that essentially took place in the software world.

But when AI begins understanding spaces, generating scenes, connecting supply chains, participating in construction decisions, supporting urban renewal, and providing training environments for robots, it is no longer simply a software capability.

It gradually becomes an interface to the world.

What you give it is no longer only text,

but human intentions, spatial limits, physical constraints, and supply-chain capabilities.

What it returns is no longer merely an answer,

but a real-world plan that can be entered, modified, and executed.

So what I find worth watching about Xufeng Technology is not whether it has produced a few stunning renderings. It is that the company has identified a crucial transition:

AI is moving from understanding information

toward understanding the world.

Space may be the first layer transformed in this process and the one most readily able to develop a complete commercial loop.

If content generation was AI's hottest topic over the past two years,

what may deserve more attention in the coming years is not how many more images, videos, or pieces of copy it generates, but:

Whether it has begun generating spaces, taking over workflows,

connecting supply chains, training robots,

and genuinely entering the physical world.

Once that happens, AI will no longer be just a productivity tool inside a screen.

It will become an operating system for the real world.

And companies that move earliest from generating images to creating spaces will compete for more than a vertical market.

They will compete for the power to interpret the world in the next generation of human–machine collaboration.

Selected Q&A

Q1: What is Xufeng Technology's core value?

It is not making prettier pictures, but quickly turning people's vague spatial needs into digital-space plans that can be understood, modified, and transacted around. It solves the problem of turning abstract requirements into images and then moves further toward expressing them as spaces.

Q2: How does it differ most from ordinary AI image-generation tools?

Ordinary AI image-generation tools mostly stop at presentation. Once the image is generated, most of the value is over. Xufeng Technology emphasizes controllability, scalability, and integration into business processes, connecting design to real supply chains to complete the transaction loop.

Q3: Why is the shift from images to spaces so important?

Images help express ideas, but spaces can support actual business. Once a space is digitized, made three-dimensional, and made interactive, it can connect to design, recommendations, purchasing, construction, marketing, and even robot training.

Q4: Why did it begin with home renovation?

Home renovation is an industry characterized by high spending per customer, high communication costs, and heavy reliance on visuals. Customers may not understand technical drawings but can quickly judge renderings, making it easier for AI to directly affect sales efficiency.

Q5: Why does the article keep emphasizing the supply chain?

The biggest commercial value lies not in generation itself, but in whether AI enters the transaction chain. If products in renderings connect directly to real supply chains, AI becomes more than a tool: it becomes an engine for order conversion.

Q6: Why pursue G to B to C rather than directly building a consumer hit?

Spaces and home furnishings are inherently tied to industry. Leading enterprises, industrial clusters, and state-owned and centrally administered enterprise projects control greater resources and renovation scenarios. Entering those systems first makes scaled deployment easier, followed by reaching consumers downstream.

Q7: What does it have to do with robots?

The 3D scene data, structural understanding of spaces, and world-model capabilities accumulated through spatial intelligence can provide robots with high-quality simulation training environments. Put simply, spatial modeling today may be laying the groundwork for embodied intelligence tomorrow.

Q8: Why might Chinese companies have an advantage in this direction?

China has denser demand for transforming spaces, more complete building-material and home-furnishing supply chains, larger urban and everyday-life application scenarios, and an easier path to accumulating real-world data at scale. Once industrialization begins, that advantage can become particularly apparent.

This article was compiled from an in-depth interview with Xufeng Technology.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260321-03html
On-site reading page: https://ffcap.cn/en/research/src-20260321-03html