Matrix Pixel secures 7 million yuan in seed round financing, with Star List Interactive Entertainment as the investor.

36Kr learned that Beijing Matrix Pixel Technology Co., Ltd. recently completed a seed - round financing of 7 million RMB. The investor is Xingbang Interactive Entertainment. This round of financing will be mainly used to continuously deepen the R & D of multi - modal Agent technology, accelerate its large - scale implementation in scenarios such as AI comic dramas, AI e - commerce, and information flow advertising. At the same time, it will promote the R & D and market promotion of new products - the full Agent proxy workspace and OpenPaper.

Beijing Matrix Pixel Technology Co., Ltd. was founded by CEO Zhang Anyang. The core team members come from companies such as Huawei, Baidu, Novartis, Handy Mobile, and BlueFocus, with experience in algorithms, products, commercialization, and industry scenarios. Different from many startup teams starting from model capabilities, Matrix Pixel emphasizes a system view for real - world business: it does not regard large models as a superposition of single - point capabilities, but as a production system that needs to be constrained, orchestrated, corrected, supervised, and continuously evolved. The company currently focuses on the B - end market, mainly providing Agent products and service capabilities around content production, process collaboration, and business automation for e - commerce, comic drama, and information flow advertising companies.

In the eyes of the outside world, the AI content track is already quite crowded: there are more and more models with stronger and stronger generation capabilities, and canvases, workflows, plugins, and knowledge bases have almost become standard. However, when it comes to actual enterprise applications, the problem is not "whether it can generate", but "whether it can produce deliverable, reusable, controllable, and reviewable results in a long - term and stable manner". Many tools perform amazingly in the demo stage, but once they enter real - world business, they will expose problems such as context drift, style out - of - control, task interruption, knowledge forgetting, unstable aesthetics, and non - accountable results. Matrix Pixel believes that the core competitiveness of AI startup companies in the next stage will no longer be the ability to call a single model, but the ability to understand complex task systems.

Based on this judgment, Matrix Pixel defines its technical route as follows: based on multi - modal large models, using Agents as execution units, an engineering orchestration system as the framework, and industry knowledge and aesthetic norms as constraints, to ultimately build a digital employee system for industrialized content production. The company's core product, NextCut AI, and the newly released full Agent proxy workspace are not tools for single - prompt generation, but attempt to connect "understanding requirements - disassembling tasks - generating content - automatic modification - multi - round feedback - version convergence - result precipitation" into a business closed - loop that can be repeatedly called.

According to the company, its technical advantage is first reflected in the underlying Harness Engineering ability. Compared with general - purpose workflow products that focus on node connection and interface assembly, Matrix Pixel pays more attention to the fine control of the Agent behavior process itself, including task disassembly strategies, context inheritance paths, long - and short - term memory scheduling, knowledge retrieval priorities, action call constraints, abnormal result rollback, version iteration management, and multi - Agent collaboration rules. This means that the system does not simply "string together" multiple models, but attempts to establish a control layer similar to a production operating system, so that the output of each Agent is under structured supervision.

Furthermore, Matrix Pixel does not understand "intelligence" as the natural spill - over of model parameter scale, but emphasizes the usable intelligence jointly shaped by algorithms and business constraints. In high - frequency content industries such as e - commerce, comic dramas, and advertising, what is truly valuable is not a one - time burst of inspiration, but the ability to continuously produce content that meets platform logic, placement logic, narrative logic, and aesthetic logic. Therefore, Matrix Pixel has emphasized several types of capabilities in product design:

Firstly, it is the multi - stage task planning ability, which enables Agents not only to answer questions, but also to actively disassemble steps, organize materials, and schedule different capability modules around business goals;

Secondly, it is the knowledge and habit learning ability. By accessing the enterprise - specific knowledge base, style specifications, and historical operation records, the system can gradually form an understanding of the team's preferences and business rules;

Thirdly, it is the result verification and reflection mechanism. After generation, it automatically conducts structure checks, style consistency checks, logical conflict checks, and executability reviews to reduce pseudo - completed results that "seem right but actually cannot be delivered".

In the field of multi - modal content generation, this method is particularly important. Because the difficulty of content such as images, videos, copywriting, and audio is never just the generation itself, but cross - modal consistency. Whether the character settings are stable, whether the shot language is unified, whether the emotional expression is continuous, whether the picture style is consistent, and whether the copywriting and pictures complement rather than conflict with each other. These problems cannot be solved by a single prompt. Matrix Pixel hopes to abstract the functions of different roles such as screenwriters, visual designers, auditory designers, and post - production compositors into callable proxy capabilities through a multi - Agent collaboration mechanism, so that content production can move from "single - point generation" to "multi - role collaboration". This is also one of the key investment directions in the AI comic drama and AI advertising scenarios.

In addition to the construction at the technical and algorithm levels, Matrix Pixel also shows a strong sense of security in its product philosophy. The company believes that the real requirements of B - end customers for AI are not "omnipotence", but "clear boundaries, controllable results, and traceable exceptions". Therefore, in system design, Matrix Pixel does not pursue completely unrestricted automation, but emphasizes "authorized automation": the parts that should be completed autonomously by Agents should be completed as autonomously as possible, but in key nodes such as brand style, placement strategy, business - sensitive information, customer - specific materials, and result release, review, confirmation, tracking, and rollback mechanisms must be retained. In other words, Matrix Pixel tries to make Agents act like high - performing employees, rather than unpredictable black boxes.

This sense of security is also reflected in the way of using enterprise knowledge and generating content. According to the company, the product already supports automatic modification and production of canvases through instructions on office platforms such as Feishu, and integrates multi - business - specific knowledge bases. For enterprises, this means that AI is no longer just a "public model interface", but gradually has internal organizational context: it knows what can be used and what cannot be used, what expressions conform to the brand tone, and which processes must be manually reviewed. For enterprises that need to produce a large amount of product content, advertising materials, plot storyboards, character settings, and marketing copywriting, security is not a limitation on creativity, but a prerequisite for releasing creativity. Only when the system is reliable enough will enterprises dare to truly hand over core processes to AI.

In terms of aesthetics, Matrix Pixel also tries to get out of the "average generation" dilemma commonly existing in the current industry. Although many current AI content tools can significantly improve production speed, they still tend to have problems such as templatization, plastic feeling, homogenization, and insufficient emotional density in aesthetics: the pictures seem delicate but lack style judgment; the copywriting seems smooth but has no real rhythm and tone; the videos seem complete but are difficult to form memorable points for dissemination. Matrix Pixel believes that the value of the next - generation content Agents is not to help users produce "standard answers" faster, but to help teams produce works with recognition, completeness, and commercial conversion value more stably.

For this reason, Matrix Pixel is not satisfied with using AI as a cost - reduction tool, but regards "aesthetic control" as one of the core barriers. In the company's internal definition, high - quality content is not only about high - definition, coherence, and generability at the technical level, but also includes narrative rhythm, visual unity, restraint in expression, brand consistency, and platform dissemination adaptability. This means that aesthetics is no longer just a "subjective feeling", but will be disassembled into a set of production standards that can be learned, tested, and iteratively optimized. The company hopes that through the long - term learning ability of Agents, the system can gradually understand the style habits of different business teams, so as to precipitate "aesthetic experience" into "executable organizational assets".

At the commercialization level, Matrix Pixel has shown a certain verification basis. According to the data provided by the company, before the product was fully launched to the public, nearly 200 users and 9 enterprises had joined during the internal testing phase; the new product achieved a cumulative recharge income of 16,000 yuan within two days after its release. Although in terms of absolute scale, these are still the starting data of an early - stage team, in the context where many current AI tools are still in the experience and trial - use stage, being able to obtain the first batch of paying users without fully expanding the channels indicates that there is a real market demand for its product direction. More importantly, these users are not simply paying for "trying something new", but have continuous demands for efficiency, quality, and process collaboration in actual work.

The industrial resources behind the company also provide a realistic foundation for this verification. As a company in the B - round, Xingbang Interactive Entertainment, the controlling shareholder of Matrix Pixel, has a relatively mature industrial foundation and business resources, and can provide support for Matrix Pixel in customer reception, scenario verification, content production, and business collaboration. For early - stage AI startup teams, the most difficult thing is not to create a fully - functional product, but to polish the system to be stable, useful, and worthy of payment under continuous business pressure. The introduction of industrial resources means that Matrix Pixel can train its products in real - world business links, rather than staying in laboratory - style ability demonstrations.

From a competitive perspective, there are already several leading players in the current market, and the industry concentration is rapidly increasing. Leading products usually have a strong community ecosystem, multi - model access capabilities, or a mature workflow system, occupying the main attention and user minds in the market. However, Matrix Pixel does not define itself as "just another AI canvas tool" or "just another workflow platform", but tries to start from a deeper - level task system: it hopes to solve the fundamental problems in enterprises' complex content production - how to make AI truly understand business goals, how to make the results continuously approach the team's aesthetics, how to build automation on the basis of safety and responsibility, and how to accumulate organizational experience with the use of the system, rather than reinventing it in each generation.

CEO Zhang Anyang said: "Our goal is not to create a tool that can generate content, but to build a digital employee ecosystem that can understand business, execute complex tasks, and self - iterate and evolve. A truly valuable Agent is not one that answers like a human, but one that works more and more like a mature team." In his view, what enterprises really need is not a more lively AI interface, but a more stable, controllable, and aesthetically - judgmental production system, which can free the team from mechanical execution and allow them to focus on higher - level creative judgment, business strategies, and resource allocation.

In the current situation where generative AI is moving from "model worship" to "system competition", the technical view, product view, and security view emphasized by Matrix Pixel represent the direction of a new type of startup company to some extent: they no longer regard large - model capabilities as the end - goal, but use them as a foundation to build an Agent system that can truly enter the daily work of enterprises. Whoever can combine algorithmic capabilities, engineering control, safety boundaries, and aesthetic quality will have the opportunity to cross the demo stage and become the production infrastructure that enterprises truly rely on.