Produce a clothing design draft in 30 seconds, Semir uses function computing + AIGC to create a "new life"!

The open source China community team made its first live broadcast, telling the story behind the open source China community in the name of sharing."

How innovative projects can empower our business is very important at Semir. Alibaba Cloud Function Computing helps us to eliminate many problems such as GPU computing resource reserves, procurement costs, and technical thresholds in implementing AI in actual business scenarios, so that we can make decisions quickly, stand on the right starting point, and experience new things. Technology has brought changes to the entire fashion design and marketing links.

—— Lin Jianxia, Head of Innovation Business R&D of Semir Digital Center

2023 is called the first year of AIGC by many. Many large AI models have shown people their amazing creativity, and the imagination of thousands of industries for the implementation of AI technology has grown infinitely. As an indispensable computing power participant in the AIGC era, Alibaba Cloud escorts enterprises to the ground, helping them to use cloud computing to effectively obtain the "dividends" brought by the AI wave at lower costs and with higher efficiency.

At the beginning of 2023, Semir hopes to quickly deeply integrate AI technology into the company's core business processes and realize the intelligent upgrade of the entire chain in the apparel industry, from product design, manufacturing, marketing and promotion to after-sales service. On a sunny afternoon, we walked into Semir and talked about how this company, which has accompanied the growth of generations of Chinese young people, quickly realizes the "collision" between the traditional clothing industry and AI based on Alibaba Cloud Function Computing, and uses AIGC in multiple scenarios. Improve production efficiency.

Semir resonates with the times

Founded in 1996, Semir has been committed to providing young people in the new era of China with matching products suitable for various life scenarios, conveying the brand concept of intimacy, fashion and comfort. After 27 years of development, Semir has resonated with the times and grown into one of the representative brands in China's fashion industry. In addition to its efforts in the field of adult clothing, Semir also owns many children's clothing brands such as Balabala, and has become a leader in the field of children's clothing in China.

By 2022, Semir's e-commerce GMV (gross merchandise transactions) has exceeded 12 billion yuan, and its offline channel outlets have increased to more than 8,000, covering more than 400 cities across the country, and has more than 2,000 franchisees across the country. In the process of continuous iteration and scale expansion of its business, Semir has also achieved a transition from informatization to digitalization in the field of dataization.

Three major challenges faced by enterprises under the AIGC wave

The rapid development of AIGC technology has enabled Vincentian graph models to continuously achieve better generation effects. At the same time, open source projects represented by Stable Diffusion (hereinafter referred to as SD) are unprecedentedly active, and Vincentian graphs have become one of the hottest application areas for large models. Semir hopes to quickly deeply integrate AI technology into the company's core business processes to realize the intelligent upgrade of the entire chain in the apparel industry, from product design, manufacturing, marketing and promotion to after-sales service. Like the problems faced by many companies, Semir also encountered many difficulties in the early stages of the project.

The initial decision-making cycle of enterprise-level AI projects is long

Tasks such as Vincentian graphs that highly rely on complex computing models have extremely high requirements on hardware facilities, especially GPU computing power. In addition, in conjunction with model inference services, corresponding AI applications need to be built, which require CPU computing, storage, high-speed networks, etc., each of which determines whether the project can operate efficiently.

For the project leader, for an innovative project that is still in the exploratory stage, it is more expected to verify the value output at the minimum cost. However, faced with the expensive GPU computing power on the market and the long procurement cycle of IDC data centers, it is necessary to Accurately estimating and making decisions on huge hardware investments is undoubtedly a difficult task. Not only is the financial pressure huge, but also the uncertainty in the procurement, deployment, maintenance and future upgrades of hardware equipment, as well as the recruitment of professional operation and maintenance personnel, may cause unnecessary cost waste and time delays, and become a stumbling block to project advancement.

Self-built GPU clusters are costly and difficult to operate and maintain

Faced with the overwhelming popularity of AI, the first thing many companies think of is to build their own GPU clusters and deploy large models locally. However, in addition to purchasing high-end GPU servers, you also need to consider the costs of storage, network and other infrastructure. The initial resource investment is very large, and the later operation and maintenance problems are even more prohibitive for many companies. Power consumption, cooling systems, maintenance personnel Ongoing operating costs and other costs make it a difficult decision for companies to build their own GPU clusters, and the same is true for Semir in the apparel industry.

Open source models are difficult to deploy and users are difficult to isolate

Stable Diffusion is an open source AI image generation model. For enterprise users who have stricter requirements on stability and efficiency, directly using this open source model faces the challenge of difficult user isolation and difficult resource allocation. The most obvious example is that when multiple people use the same Stable Diffusion WebUI with high concurrency, there will be long queues, "stuck" situations, and model changes that interfere with each other, resulting in poor user experience for designers, reduced production efficiency, and interruption of creativity. And other issues.

Semir launched, Alibaba Cloud Function Computing’s AIGC solution

Function Compute creates the simplest AI inference application development and deployment platform

Function Compute is an event-driven, fully managed computing service. Using Function Compute, there is no need to purchase and manage infrastructure such as servers. You only need to write and upload code or images. Function Compute will prepare computing resources to run business requests or tasks flexibly and reliably, and provide functions such as log query, performance monitoring, and alarms. Function Compute is a typical serverless computing service that allows enterprises and developers to focus on business and save resource maintenance and other work. Function Compute provides elastic scaling based on volume and payment based on actual resource usage.

Serverless GPU + classic AIGC scenario

As early as 2 years ago, Function Compute was deploying GPU computing power to provide users with flexible, volume-based and highly virtualized GPU computing power clusters. It currently supports mainstream inference card types such as T4 and A10, and has been virtualized. Segmentation supports a minimum video memory configuration of 1GB for a single instance, and new card specifications will continue to be introduced in the future.

After AIGC exploded last year, the demand for AIGC-driven innovation from enterprises and developers also exploded. However, during the implementation process, they generally faced pain points such as the high cost of GPU computing power and the complexity of building AI applications. Therefore, the Function Computing team keeps abreast of industry trends and is committed to creating the simplest AI inference application development and deployment platform. Not only are many classic open source models hosted in the Serverless Application Center, such as SD, ComfyUI, etc.

Taking the SD model as an example, it supports one-click deployment and can be used in production in about a few minutes. It has been optimized for open source SD models, supporting SD model management interface, SD WEB UI front-end and back-end separation, and highly encapsulated flexible API interfaces, allowing users to use SD model services in various forms. At the same time, Function Compute has also been deeply integrated with the Magic Community. Many models in the Magic Community can also be deployed to Function Compute with one click for quick experience and use. If you are interested, you can refer to this article "Magic Big Model 1" Key deployment to Alibaba Cloud Function Compute, the GPU idle accounting function can significantly reduce overhead .

In addition to model hosting, Function Compute also plans AI Runtime, which allows users to quickly convert their models into API services (Model To API). Function Compute can help users create images and complete model deployment online. Finally, it must be mentioned that no matter which method of use is used, it is inseparable from the elastic, operation-free, pay-as-you-go and pay-as-you-go GPU computing power provided by Function Compute, which can carry AI more flexibly and at a lower cost. Reasoning applications. The combination of Serverless Application Center and Function Computing allows developers to truly deploy and use AIGC applications with zero technical threshold, allowing everyone to have an exclusive AIGC environment. If you are interested, you can refer to this article "Is the AI Painting Platform Difficult to Develop and Monetize?" Try the Stable Diffusion API Serverless solution .

Function computing helps Semir surf AIGC

Alibaba Cloud Function Compute FC exactly solves the usage problems faced by Semir. Based on the one-click deployment of Stable Diffusion capabilities provided by Function Compute, Semir quickly launched multiple SD usage environments. In order to allow more employees to conveniently experience the effects of using SD, Semir’s internal DingTalk workbench has integrated an online environment to provide most employees with a specific effect SD usage page, where models and plug-ins can be fixed and used as needed. May simplify the complexity of using native SD. At the same time, for professional designers, a native SD usage environment will be provided, allowing designers to freely combine DIY models and plug-ins.

Semir AI workbench

After more than half a year of exploration, Semir has expanded from the initial use of SD mapping by function calculation in two business units to now having 6 BU units and 13 groups connected one after another, using AIGC capabilities to improve service design efficiency and simplify work processes. During this process, Semir also received a lot of real experience feedback from front-line designers. For example, the native SD page is still too complicated, and it hopes to introduce AI workflow to improve the efficiency of drawing, etc.

To this end, Semir and the Function Computing team have launched more in-depth cooperation to discuss upgrade usage plans:

Introducing ComfyUI to provide AI workflow drawing capabilities.
Introduce front-end engineers to develop a UI interface exclusive to Semir, and integrate the SD Serverless API provided by Function Compute to further enhance the user experience for a wide range of designers and shield the complexity of the native SD interface.

The cooperation process with Semir has also enabled the technical teams of both parties to jointly improve their technology accumulation in the field of AIGC applications.

Improve efficiency! From 3 days to 30 seconds

Fashion design scene, only 30 seconds from inspiration to final draft

In the design inspiration stage, fashion designers originally need to spend 2 to 3 days to complete a clothing design draft, from early inspiration collection, line drawing conception, coloring, to finally presenting the pattern and submitting it for review. But now, with the assistance of AI technology, the same process may only take 30 seconds. Designers can quickly interact with AI, combine their own descriptions, and quickly transform ideas into visual works, opening up a new hot design path with the help of AI technology.

Promotional marketing scenario, one person completes complex poster design

In the visual filming process, behind a poster usually lies the hard work of a professional team. Imagine that any beautiful promotional poster requires the professional shooting of a photographer, the appropriate styling of the model, the skillful embellishment of the makeup artist, and the matching artist to select clothing and accessories for the model. The selection of the scene is also crucial, and the entire process requires the cooperation of multiple people to complete.

Thanks to the assistance of AI technology, visual photography has also undergone changes. From relying on an entire team for shooting and production, it can now be completed by just one person + AI. Through model training, the real upper body effect of amateur models or mannequins can be accurately transformed, and the background of product posters can also be generated as desired. Not only does it greatly simplify the design and marketing workflow, it also speeds up the conversion process from idea to final product.

Explore new paths for deep integration of AI technology and service industries on the cloud

Ms. Lin Jianxia, head of the innovation project of Semir Digital Center, also shared with us the potential application prospects of AI technology in Semir’s business:

Analyze massive user data and fashion trend information through AI algorithms, accurately predict and quickly respond to changes in the clothing preferences of young people in the new era, and help designers create clothing styles that are more in line with market demand;
Use AI to optimize supply chain management, improve production efficiency, reduce inventory risks, and ensure that products can reach consumers faster and at better prices; thirdly, use AI for personalized recommendations and intelligent customer service to enhance the shopping experience and enhance User stickiness;
Through AI-assisted data analysis, real-time insights into market feedback can be provided to provide scientific basis for brand strategic decisions.

In the future, the Semir Digital Center Innovation Project Team will actively explore the deep integration path of AI technology and the apparel industry with an open mind, strive to achieve effective breakthroughs in AI technology in the business in the shortest possible time, and lead the new trend of digital and intelligent transformation in the industry. .

This year, Semir Clothing plans to deploy this cutting-edge AI drawing technology in terminal stores, allowing customers in the store to make DIY T-shirts on-site. As a technical supporter of this project, the Function Computing team applies Serverless technology to GPUs and proposes a new vision of "everyone can have an exclusive AIGC environment", significantly reducing users' trial and error costs and usage thresholds in the AI field. , allowing users to focus more on their own business development without worrying too much about the operation and maintenance of underlying resources.