Meitu has launched the 3.0 version of its in-house vision model, seamlessly integrated into Meitu's suite of image and design products


On October 9th, Meitu held its 15th anniversary party and released version 3.0 of its self-developed AI vision model, MiracleVision One hundred days following its launch, Meitu's MiracleVision 3.0, an advanced AI vision model, will be seamlessly integrated into the company's image and design products. This cutting-edge technology is set to support five key sectors, including e-commerce, advertising, game, animation, and film and television.

The 100-day evolution of Meitu's vision model: Innovative Cognition and Smart Creation

At Meitu's 15th-anniversary celebration, Wu Xinhong, the founder, chairman, and CEO, unveiled version 3.0 of their self-developed AI vision model, MiracleVision.

Wu Xinhong, Founder, Chairman and CEO of Meitu

Wu Xinhong announced that MiracleVision, which had a basic artistic skills just three months ago, has evolved to version 3.0, capable of crafting artwork with realistic and intricate details. Users can explore this transformation by visiting the official WHEE website.

Wu closely compared his own 15-year-old painting to artwork generated by the 100-day-old AI model. He found that AI training in drawing shares traits with human learning. By extensively studying various styles and techniques, AI can create distinct original pieces of art.

Meitu approaches the core abilities of MiracleVision as two halves: Innovative cognition and smart creation.

Innovative cognition: MiracleVision's keyword association feature simplifies the user experience by automatically applying elements like lighting, texture, style, and image quality based on entered keywords, making it more accessible for all users. In addition, the ability to recognize keywords enables MiracleVision to satisfy professional design requirements, such as medium close-ups, long shots, front-lighting and back-lighting.

Smart creation: MiracleVision enhances the details and expressiveness of visuals with the Detailed Development feature. AI Image Expansion expands images with natural and rich details. Localized Adjustment allows for precise and refined modifications. Quality Enhancer brings out the vibrancy and clarity of image details, colors, and objects.

This cutting-edge technology is set to support five key sectors, including e-commerce, advertising, game, animation, and film and television

According to Wu, the impact of large-scale AI vision models goes beyond improved visual representation, encompassing the optimization of workflow. Nevertheless, since the latter is still in the developmental stage, it is essential to give priority to meeting the needs of diverse application scenarios.

According to Wu, MiracleVision will cater to five major industries, namely e-commerce, advertising, game, animation, and film & television. Its aim is to enhance the work efficiency of these industries.

For e-commerce, MiracleVision assists throughout the entire process, including sketch drawing, coloring, generating product photos, try-on effects, and final output for sale.

For advertising, MiracleVision's services encompass providing inspiration, enhancing creative ideas, assisting in composition, dynamically adjusting size, and streamlining the production process.

For the games industry, MiracleVision can handle various processes such as scene design, character design, prop design, UI design, promotional materials, etc., expanding the creative space for designers while helping to reduce costs.

For the animation industry, MiracleVision integrates the processes of concept design, storyboard generation, sketch coloring, frame interpolation, and video-to-animation conversion, supporting the fast implementation of ideas into finished products.

For the film and television industry, MiracleVision can fully meet the requirements of scene design, storyboard design, character modeling, prop design, and publicity materials.

Wu stated that MiracleVision will leverage its capabilities to the fullest extent for the five aforementioned industries and continue supporting additional design scenarios.

he widespread application of large-scale vision models will enter a period of maturity between 2026 and 2030

At the event, Wu also shared his thoughts on large-scale AI vision models. "The AI vision model is essentially an endless inspiration library of visual elements. Parties in need can extract necessary content from this library based on user requirements, making it usable for specific scenarios. The AI vision model and applications work in tandem, with the model providing technical support to applications while applications contribute to the model's growth."

Wu explained that there are three existing issues with AI vision models in the production process: ultimate effect pursuit, workflow integration, and monetization capability. As the model and production processes become increasingly integrated, these three challenges will gradually be overcome.

According to Wu, the popularization of vision models will pass through three stages: the exploration period before 2024, a fast development period between 2024 and 2025, and a mature period from 2026 to 2030. As production applications become more sophisticated, vision models will enable millions of design scenarios, driving aesthetic advancement and socio-economic growth.

During the event, Meitu also presented the latest data on its image productivity tools that were launched on June 19th. The visual creation tool "WHEE" has generated over 5.5 million AI effects, and the designer ecosystem is beginning to yield results. The talking video tool "Action" has achieved a monthly active user base exceeding 350,000. The desktop AI video editing tool "WinkStudio" has served nearly 100,000 video creators, while "Meitu X-Design," an AI commercial design product, is aiding almost one million small and medium-sized e-commerce businesses in reducing costs and increasing efficiency. The Meitu AI assistant "RoboNeo" assists close to 10,000 users with automatic photo retouching every day, generating a 35% increase in efficiency compared to traditional methods. Lastly, the virtual human generation tool "DreamAvatar" was launched at the end of August and is dedicated to providing "AI actor" services.