Alibaba Luban: AI-based Graphic Design Tool
By Alibaba Cloud Enterprise Application Team
This year, in addition to providing services for Tmall and Taobao, Luban exported AI design capabilities to customers in various industries through Alibaba Cloud. During the 2018 Alibaba Double 11 Shopping Festival, Luban has cumulatively designed about 6 million banners for 200,000 enterprises/merchants. Most banners and promotional pictures that you see at Tmall are designed by Luban.
A Taobao merchant once said, “We have more than 400 products, and we can’t design pictures for so many products on our own. Therefore, we basically gave up the pictures in the past, even for important events like Double 11. However, Luban helped us complete the design of all pictures for Double 11 this year.”
It is said that Luban has already achieved the level of an advanced designer. It is able to provide dozens of solutions and design 8,000 different banners every second. All these abilities are attributable to the artificial intelligence technology.
Luban’s Design Knowledge Profile
Luban’s knowledge profile is the data foundation of picture effects. Luban add tags to and analyzes finished pictures/designs in the dimensions of content, visual style, and structure. This helps it build the cognition of picture composition, color blending, matching, and copywriting in the design element level. After comprehensively analyzing millions of marketing design picture sets, Luban established the largest knowledge profile database of marketing pictures in China.
During the construction process of the underlying layer of the knowledge profile, Luban uses undirected graphs for modeling, and used relational databases and NoSQL databases to build the storage infrastructure. Through data synchronization, Luban synchronizes post-processed data to the ES index that the algorithm depends on.
Luban’s Intelligent Design Deep Learning Algorithm
Based on the intelligent design deep learning algorithm, Luban changed the traditional picture design process. Users can simply input the desired design style and picture size, and leave the time-consuming design jobs, such as material analysis, picture retouching, color blending, and layout to Luban.
Luban also has a pipeline for the execution of intelligent algorithms. This pipeline abstracts complex and diversified design requirements into processes, such as feature extraction, planning, action, quantification, and synthetic visualization. This process has been currently applied in transparent banner design and full-frame banner design. In addition to improving the business iteration speed, it reduces the coupling between algorithm modules, as well as the O&M costs of algorithm services.
Image Protocol and Image Rendering Capabilities
GPU-accelerated image generation guarantees the efficiency of Luban. Luban established a set of DSL language for describing pictures at the bottom layer of image engineering. The DSL language allows Luban to describe the main composition and visual details of a picture by separating it into layers. This builds up the protocol basis for manipulating images between programs.
In combination with the GPU-based image rendering technology, Luban is able to design and generate tens of thousands of images per second. This allows users to create a large number of banners, posters, and venue pictures in a short time.
In addition, a GPU solution is more efficient than a CPU solution when describing the increasingly complex image layers. This efficiency enhancement is very important for image synthesis that requires specific complex design effects.
It is said that Luban can provide enterprises, merchants, and designers with one-click intelligent design image generation, size expansion, color expansion, and other design services.
Currently, Luban provides one-click intelligent design image generation, intelligent layout, size expansion, color expansion, and other design services. Users can generate multiple sets of design solutions that meet the requirements in real time by simply inputting the desired style and size.
