AI Product Thinking — How Lovart Redefines AI Image Products Through User Tasks in a Niche Market

Since Lovart went viral in May this year, the AI text-to-image product market has started to shift. In Lovart, users no longer need to study prompt engineering techniques. They only need to describe what they want in simple language, and the product can generate the image for them. At the same time, as a design agent, Lovart also has the ability to think autonomously. It can complete a series of reasoning and execution tasks based on user needs, such as proactively searching the web for information, generating a design strategy, and eventually producing the final image.

Did Lovart break out simply because it is an agent product? How should vertical products be built if they want to reach mass-market users? This article analyzes Lovart from the perspective of user tasks, looking at how it satisfies the needs of both ordinary users and professional designers. It also combines hands-on product evaluation, product architecture breakdown, and product replication to discuss what Lovart can teach AI product managers building products in vertical domains.

1. User Task Breakdown

Lovart’s users can mainly be divided into two groups: ordinary users and professional designers. Next, I will break down the user tasks from the perspective of these two types of users.

Ordinary Users

For ordinary users, the core task is to generate a specific image or edit an image according to specific requirements. Before AI, their usual workflow was to ask a designer to complete the relevant work. Under that model, apart from labor cost, the final result also depended on communication costs between both sides and whether the designer could deliver according to the requirements.

Now, this type of user task is better handled by Lovart. Ordinary users only need to express their needs inside the product, and Lovart can generate images with one click. The results are accurate and high-quality.

These ordinary users with design needs may be online store owners, small and medium-sized business founders, or operations specialists. They may need Lovart to create simple branded product images or marketing visuals.

Compared with previous AI text-to-image products such as Stable Diffusion and Midjourney, where users had to spend a lot of effort writing highly complex prompts to ensure image quality, Lovart only requires users to clearly express their needs. This makes the product immediately usable for more users and bridges the gap in a professional domain. This is also the foundation of why the product was able to go viral.

Professional Designers

Professional designers have two different types of user tasks.

The first type of professional user task is to directly create a design solution. In this case, Lovart plays the role of an executor. It completes the corresponding design task based on the designer’s requirements. Under this mode of collaboration, the product helps users save a significant amount of time while still producing high-quality results.

But unlike ordinary users, designer users have higher and more granular requirements at the execution level. When AI cannot fully complete the task according to the designer’s requirements, they are even willing to manually adjust the result themselves, because this leads to higher accuracy.

Therefore, unlike products such as Stable Diffusion and Midjourney that interact with users mainly through a chat box, Lovart provides the ChatCanvas feature, allowing professional users to freely adjust images, precisely select a specific image, or directly edit the image. This experience simulates how professional designers use tools such as Figma or Photoshop in their daily work. Since designers have already formed these work habits, the switching cost when using Lovart is extremely low.

The second type of professional user task is to first plan the design solution and then execute it. In this case, Lovart plays the role of a design expert or even a design team. This is somewhat similar to daily work scenarios where designers collaborate with each other, or where a designer works with a third-party design team.

However, Lovart’s agent capability allows it to both think and execute. At the beginning, it needs to collect information across the web based on the designer’s requirements, extract value points, think through the corresponding design strategy, and then execute the design solution based on the specific content of that strategy.

Under this collaboration model, the information collected by the product can also bring designers a lot of inspiration. An effective design strategy can directly help designers produce high-value design solutions that meet the requirements, while also saving them a large amount of execution time.

What makes this different from the first type of designer, who only asks the product to execute, is that the second type of designer does not just need a result. They also need a process of thinking and analysis. More importantly, the information collected by the product, the design strategy it defines, the final design solution it produces, and the entire process of thinking, reasoning, and execution all need to be logically connected and mutually verifiable.

Therefore, when Lovart handles this type of user task, its workflow includes: breaking down the requirement, searching the web to collect information, extracting design points from the mood board, defining the design strategy, breaking down the design system, and outputting the final design solution.

At the same time, the product shows this entire process of thinking, analysis, and execution to the user. This way, if there is a problem with the generated result later, designers can trace back the design logic. After designers receive the first draft, they can use it as a foundation and ask Lovart to continue optimizing it according to specific needs until the final design solution meets delivery standards.

This experience simulates how professional designers complete a design task from zero to one in their daily work, including execution, review, and delivery. Lovart takes over this workflow. It does not only execute; it can also assist designers in thinking and analysis.

Even when designers have multiple design ideas and directions, the product can execute efficiently and help them try different possibilities. Design work that originally took a week or even longer can be compressed into just a few hours.

User Task Analysis for Overseas Users

Lovart currently mainly serves overseas markets. Taking North American users as an example, their user tasks are different from those of domestic users. The core difference lies in different cultural backgrounds, which lead to different lifestyles and therefore different user behaviors.

For example, ordinary users in North America mainly include small business founders or photographers. Their user tasks mainly include:

• Designing a set of materials around a personal brand, such as a personal brand logo, website, banner, and so on. This also includes related marketing materials for Facebook, Instagram, or email marketing. This is completely different from how operations work on domestic platforms.
• Some users may also need printed marketing materials, such as brand brochures for offline exhibitions, stickers with brand logos, mugs, T-shirts, and so on. Users not only have requirements for the quality of the design solution, but also want the final printed product to stay consistent with the design.
• Photographers who take photos for clients in a studio can use Lovart to quickly retouch images and show clients an initial result on-site. The product not only helps them save time and quickly see results, but also helps build trust between photographers and clients.

Understanding the user tasks of overseas users, meaning the differences in product usage scenarios and purposes, can help Lovart deeply explore product value and provide more valuable services to users.

2. Product Architecture Analysis

Lovart uses an MCoT, or Multi-Chain of Thought, reasoning engine to imitate the working model of a design director. This is also the core innovation of the product.

Based on the user’s requirement, Lovart first classifies the task into complex tasks, simple tasks, and special tasks. Complex tasks involve systematic analysis, reasoning, and execution. Simple tasks involve direct execution. Special tasks involve storyboard generation. The task is then assigned to the most suitable agent to complete the design solution.

In the product evaluation, I also verified this mechanism. Different user prompts trigger different working mechanisms in the product:

• Asking Lovart to act as a design expert to complete a complex task:

• Asking Lovart to directly complete image editing:

• Asking Lovart to complete storyboard design:

Workflow Breakdown

Based on Lovart’s system prompt found publicly online, it uses a multi-agent architecture. The Coco agent is responsible for understanding user needs and assigning tasks to different sub-agents.

System Prompt:

The workflow can be summarized as follows:

The main agent works like a “router.” It is only responsible for identifying the requirement and assigning the task. The corresponding assignment mechanism is triggered by the user prompt.

• The main agent’s capability is similar to that of an experienced design director who has worked in the design field for many years. It has full professional design knowledge, can accurately understand the specific design domain involved in the task, and then assign the task.
• Each sub-agent works like a director in a specific design field:
  • These fields include brand design, graphic design, visual design, IP character design, and other vertical design areas.
  • For simple requirements, the sub-agent first understands the user intent, then calls the corresponding workflow or text-to-image model to complete the design solution.
  • For complex requirements, after understanding user intent, it proactively searches, collects, organizes, and verifies information, clearly defines the design strategy, and then starts producing the design solution.

Product Replication Approach

To analyze the product’s value in depth, I directly replicated this product using Vibe Coding. This allowed me to compare Lovart’s experience during evaluation and better understand how its product architecture works.

The key points of the replication included:

• A product architecture based on MCoT technology, allowing it to actively think, collect information, and analyze information.
• Clearly defining a design strategy and delivering high-quality design solutions based on that strategy.

Product replication demo style:

The user requirement was: “Generate a brand logo for me. It should be a cute little crocodile, for a tennis racket brand. Then based on this logo, generate a tennis course introduction poster and an event sticker.”

Its execution process included:

• Analyzing the user requirement, calling the search tool to collect information, and extracting key insights.
• Generating each required image for the user one by one.

After the design was completed, users could also drag images inside the ChatCanvas-like editing area on the left, and then use the chat box on the right to make further requests for secondary image editing. At the same time, I also added dark mode to the product so users can switch between modes.

The workflow traced by LangSmith:

• After receiving the user requirement, it called the search tool to collect information and define a design strategy.
• It handed the requirement over to the corresponding sub-agent to complete image generation.

Overall, the replicated product can play the role of a design expert or design team to complete user requirements. However, throughout the replication process, I also conducted multiple evaluations and compared it with Lovart. I found that Lovart has considered many important details very thoroughly. The following major product values are worth paying attention to:

1. System prompts:
   In Lovart, not only does the main agent have a powerful and complete system prompt, but I infer that each sub-agent responsible for a specific design field also has a clear and detailed system prompt. These prompts define the role, work tasks, tool calling rules, and other details very carefully. This is what ensures that the generated images or videos maintain high quality and strong aesthetic standards.

2. Effectiveness of the design strategy:
   When the main agent defines the design strategy, it not only needs to refer to the information found through search and the design highlights extracted from the mood board, but also needs to ensure that the sub-agent can directly generate images that meet user needs based on the specific content of the strategy. In other words, it must ensure both the effectiveness of the strategy and the completeness of the design system, including detailed definitions of brand tone, color system, typography system, visual elements, and other dimensions.

3. Frontend product experience:
   Features such as ChatCanvas, Edit Element, Edit Text, and cropping in Lovart all align with the daily work habits of designers, allowing these users to get started quickly. At the same time, the chat box on the right side of the product page lowers the learning cost for ordinary users and can effectively attract more users of this type. It is also worth mentioning that even for complex functions, the product can still generate results quickly.

The value of Lovart lies in its innovative product architecture and highly professional system prompts, which together enable high-quality output. Through continuous evaluation, the product has also accumulated a large number of real user cases and continuously iterated its product architecture and algorithms based on feedback. This is why it has been able to stand out in the vertical AI product market.

3. Thoughts on Future Product Optimization

Starting from the core idea of user tasks, the product may have three future development directions:

1. Templatized Solutions

When designer users use the product, whether as a design executor or as a design partner they can work with, if their requirements are relatively fixed, such as repeatedly generating the same type of marketing material images or the same type of IP figurine design, the product can proactively generate specific templates for them based on project context memory.

At the same time, the product can package the user’s fixed prompts into the template. During execution, the product can combine the system prompt with the user prompt to assist the user in content creation, ensuring that each generation is both fast and aligned with the user’s needs.

2. Personalized Creation

When users create a specific type of content inside the product, Lovart can proactively guide them to try multimodal content creation and present more content formats.

For example, if a user often edits images inside the product, Lovart can learn from this behavior and proactively guide the user in the chat box to try video creation. Or, if the product identifies that the user is designing products around a personal brand, it can provide multiple product design options featuring the user’s personal brand logo, giving the user more product styles to choose from.

3. Result Visualization

Lovart CEO Chen Mian mentioned in an interview: “At present, Lovart is a product with tool attributes, but in the future, the founder hopes it can directly deliver results for paying users. AI will eventually become a service that directly delivers results.”

Since this is a result-oriented product, perhaps more capabilities can be introduced into the product to help users obtain results more efficiently.

For example, for ordinary users, Lovart could use market data to tell users what kind of product images may improve conversion rates. For designer users, Lovart could use annual design trends to tell users what kinds of brand visual styles or color systems are more popular and more aesthetically appealing.

In this way, Lovart may gradually transform from a tool-oriented product into a content creation platform that helps users efficiently obtain results.

The core of vertical AI products is that they must be able to capture the deep usage scenarios and needs of professional users, fit their product usage habits, and at the same time provide a simple and direct enough experience for ordinary mass-market users to quickly get started and obtain results.

Vertical AI products will not replace experts in any field. The capability of agents will only help domain experts complete content creation more efficiently.