Dialogue with Chang Jia, Founder of Wujie AI: Entrepreneurial Opportunities in AI Lie in the Application Layer
-
The lowering of technical barriers has spurred a surge in AI painting startups, which also means fierce competition. Wujie AI, as one of them, has firsthand experience of this.
Currently, Wujie AI has fewer than 3 million users, falling short of its annual target of 10 million. Chang Jia, the founder of Wujie AI, candidly stated that user growth has entered a challenging phase, and the company will need to continue its efforts in the second half of the year. The sluggish growth of Wujie AI is a microcosm of the industry.
As an emerging field, the AI painting industry is undergoing rapid changes. At the World Artificial Intelligence Conference a week ago, Alibaba Cloud introduced a new member to its Tongyi series of large models—Tongyi Wanxiang, positioned as an AI painting creation model. Beyond Alibaba, companies like Baidu, iFlytek, and SenseTime have also launched large models with multimodal capabilities, including text-to-image generation.
This indicates that the infrastructure in the AI painting field is becoming increasingly accessible. In fact, after the open-sourcing of the Stable Diffusion model, the barrier to entry for text-to-image businesses has significantly lowered, leading to the emergence of numerous AI painting startups in China. Wujie AI is one such company that grew during this wave, starting with a "community-as-a-service" model and iteratively optimizing through community operations.
The competition is intensifying. At the end of March this year, Midjourney, the most popular AI painting company in the industry, released its V5 version, elevating the realism of generated images to "photographic quality." This has sounded the alarm for a survival battle in the AI painting field.
The prospects of AI painting are highly attractive. According to a report by Guotai Junan Securities, by 2025, AI painting is expected to achieve a 30% penetration rate in the image content generation sector, with a market size exceeding 200 billion yuan.
Amidst both temptation and challenges, only the capable will prevail. Chang Jia, whose real name is Liu Zhipeng and who is a three-time winner of China's sci-fi "Galaxy Award," faces the question of how he will lead Wujie AI to break through the competition. His answer: "In the AI generation field, domestic startups will find more opportunities at the application layer."
The following is a dialogue between All-Weather Technology and Chang Jia, edited by All-Weather Technology:
All-Weather Technology: How many users do you currently have?
Chang Jia: We currently have close to 3 million users. We haven’t developed a mini-program; our user base is primarily from the app, supplemented by website users. Our original goal for this year was to reach 10 million users, but growth has slowed in recent months. The user growth rate isn’t as fast as when AIGC first gained popularity, so we’re still far from the 10 million target. We’ll need to keep pushing in the second half of the year.
All-Weather Technology: What percentage of users are paying customers?
Chang Jia: The C-end accounts for 20%. We offer 15 free drawings per user daily, and many users come for these free credits, which we welcome. For the C-end, we aim to build boundless industry influence. Revenue mainly comes from the B-end.
All-Weather Tech: In the first half of this year, many powerful tech giants entered the AI generation field. What pressure has this brought you?
Chang Jia: We don’t see it as pressure. Our strategies and positioning are different. Major companies' AI art tools are usually self-contained systems, not integrated with open communities like Stable Diffusion (SD). Their prompt systems and parameters are incompatible. Second, these tools often serve their own well-known products, like Meitu focusing on AI photo editing. Overall, they specialize in specific directions.
Wujie is community-based and multi-directional, with products that "flow freely like water" into commercially promising fields like architecture, interior design, fashion, comics, and gaming. In high-potential vertical industries, Wujie collaborates with top B-end clients to develop industry-specific models. These models may become the battleground for AIGC in the second half of the year, as a single general-purpose model can’t meet the needs of all verticals.
All-Weather Tech: Your B-end clients span real estate, cultural tourism, and consumer goods. What’s the logic behind selecting these industries? Is it just about profitability?
Chang Jia: Some are event-driven, like those seeking outstanding works without product-level involvement. We welcome all such clients. In the first half of the year, most B-end demand focused on marketing campaigns.
Recently, MaaS (Model as a Service) collaborations have increased, such as partnerships with 3D Home and Oppein in interior design. In the second half, Wujie will release models for architecture, interior design, fashion, patterns, automotive, and industrial sectors.
All-Weather Tech: When people think of AI art, Midjourney often comes to mind. What do you think is Midjourney’s moat?
Chang Jia: Midjourney’s moat lies in its vast community and strong product appeal. It leverages Discord for continuous iteration, and the platform’s social nature helps it dominate user mindsets. Its product has a distinct personality, forming an aesthetic school. In contrast, Stable Diffusion’s open-source ecosystem lacks such cohesion.
Though Wujie is rooted in the SD ecosystem, our team and products will increasingly resemble Midjourney. In the second half, we’ll release models for interior design, architecture, and Chinese-style designs (Guo Feng, Guo Chao, Guo Man), aligning with Chinese aesthetics. In showcasing Chinese art, I believe Wujie will outperform Midjourney.
24/7 Tech: After the open-sourcing of models like Disco Diffusion and Stable Diffusion, many AI painting startups have emerged in China. Some critics argue that their models and products are highly homogenized. What’s your take on this?
Chang Jia: This is indeed a current issue. The recent boom in the Stable Diffusion (SD) ecosystem was largely driven by the passion of the anime community, but most of these models are derivative, lacking originality—often dubbed as 'melting pots of everything.' As a result, product homogenization is quite severe. Community models tend to focus on narrow themes like beautiful women, anime, and illustrations, which are far from meeting specialized vertical demands. Stable Diffusion and similar models aren’t directly commercializable; what’s being used is their underlying technology. Real competitiveness lies in innovation at the vertical application level.
24/7 Tech: What should be the barriers for AI painting companies?
Chang Jia: For Wujie (Boundless), there are two key barriers: community and product competitiveness.
Like open communities in blockchain, AIGC is fundamentally community-driven. Users fall into two categories: model trainers (often called 'alchemists') and prompt engineers (dubbed 'wizards'). These users are highly autonomous and creative. If the model parameters they develop can’t be reused on other platforms, the switching costs become prohibitive. Beyond painting models, Wujie is also training its own language model—a fine-tuned one for painting applications—to help users craft prompts, assist alchemists in tagging images, and translate natural language into label-based language understandable by SD models. We plan to open-source this for community use.
24/7 Tech: The AI painting sector exploded in the second half of last year, but few companies have secured funding. Why do you think that is?
Chang Jia: Personally, I think investors’ focus in the first half of this year has been on large language models. In this wave of tech-driven innovation, investors prefer companies with strong technical foundations. For application-layer projects, they prioritize practical questions like: 'Who are your major clients? How do you monetize? What’s your financial data?' But most AIGC application-layer companies are still in the growth phase, so their revenue metrics aren’t impressive yet. By the second half of this year, the situation should improve, and funding prospects may brighten.
24/7 Tech: What’s your current funding progress?
Chang Jia: We’re quite relaxed about it. Funding is something that will come naturally when the time is right, so we’re not actively chasing investors. Our focus remains on product development and operations.
24/7 Tech: At what stage of product maturity or user scale do you think Wujie should proactively seek funding?
(Note: The response to the last question was cut off in the original text, so the translation ends here.)
Chang Jia: Regarding financing, we don't have phased targets. We're always open to investment if capital comes to us, but we won't force it if it doesn't. However, we do have business objectives, and the pressure is quite high this year - we aim to achieve 100 million yuan in revenue and 10 million users.
All-Weather Tech: What's your current revenue status?
Chang Jia: There's still some distance from our target. The pressure in the second half of the year remains significant, but we're monitoring the growth curve and maintain an optimistic outlook.
All-Weather Tech: AI-generated art has faced copyright controversies since its inception. How do you address this issue?
Chang Jia: We've specifically researched copyright determination. Last year, people called AI a 'stitching monster' or 'copy-paste monster' - these criticisms were valid in the early stages. But as AI develops, its artistic creativity has improved significantly. Current models are fusion models incorporating multiple art styles rather than mimicking individual artists, making direct resemblance unlikely.
Additionally, for artwork, the originality of content matters more than the tools. With technologies like ControlNet, Segment Anything, and DraGAN, users gain increasing control over AI-generated images - they can segment, drag, and warp images like in Photoshop. User input now involves carefully matched model parameters and external elements like depth maps, poses, sketches and contours beyond simple prompts, demonstrating growing user originality. Personally, I believe AI-generated works will undoubtedly qualify for copyright protection as the technology advances.