Meitu's Amazing Large Model - An AI-powered visual large model launched by Meitu Inc.

What is Meitu's Fantasy Large Model?

Meitu MiracleVision is an AI visual model launched by Meitu Inc., focusing on aesthetic creation, including Eastern aesthetics, portraits, and commercial design. The model has undergone a comprehensive upgrade in video generation capabilities, capable of generating 1-minute, 24-frame-per-second, 1080P high-quality videos, significantly improving video quality, smoothness, and realism. The upgraded model is used in Meitu's products such as MeituPic, BeautyCam, and Wink, and will be gradually rolled out to other products, such as Kaipai, Meitu Design Studio, WHEE, and MOKI. Meitu's further development in the field of AI enhances the user's visual experience, improves workflow efficiency, and has application potential in e-commerce, advertising, gaming, animation, and film industries.

The main functions of Meitu's Qixiang large model

• Image and video generationGenerates images and videos of various styles and types, such as animal photography, graphic design, digital rendering, cartoon, digital 3D, animation and games, advertising photography, e-commerce advertising, concept art, industrial design, concept design, film and games, portrait beautification, installation art, portrait photography, and handicrafts.
• Textual and graphic representationsInput text or images to intelligently generate creative images, offering a variety of styles and rich adjustable parameters for precise image control.
• Video generation capabilitiesIt supports generating high-quality videos up to 1 minute long, at 24 frames per second, with a 1080P resolution, improving the video's picture quality, smoothness, and realism.
• AI screen expansionTo make the artwork larger and more detailed.
• Partial modifications: To make precise modifications and adjustments to certain parts of the image.
• resolution improvementSupports generating high-definition images, making detail rendering, color display, and object recognition more accurate and vivid.

The technical principles of Meitu's imaginative large-scale model

• Deep learning and visual TransformerIt generates images and videos based on deep learning technology, adopts a visual Transformer architecture, and combines self-attention mechanism and positional encoding to capture global and local information in images.
• Self-attention mechanismThrough the self-attention mechanism, the model can focus on important parts of the input data and compute representations of the same sequence, thereby expanding the receptive field of the image and obtaining more contextual information.
• Location codingPositional encoding is used to add positional information for each element in the sequence, enabling the model to understand the positional relationships between elements in the image.
• Aesthetic evaluation systemIntegrating an aesthetic evaluation system, this approach enhances the model's aesthetic expressiveness and creative capabilities through machine learning.
• Dataset optimizationCollaborate with external designers to build high-quality datasets and optimize the model's aesthetic performance.
• Video generation capabilitiesIt adopts a technical approach that combines Diffusion and Transformer models, and uses the DiT video generation architecture to solve the core problems of video generation capabilities such as subject consistency, motion coherence, and physical logic rationality.

Meitu's Fantasy Large Model Project Address

• Project official websitemiraclevision.com

Application scenarios of Meitu's large-scale models

• Image enhancementIn apps like Meitu and BeautyCam, it offers creative features such as AI portraits, AI removal, AI painting, AI image enlargement, and AI image editing to enhance users' image beautification experience.
• Video editing and generationVideo editing software such as Wink uses AI technology to provide functions such as animation and image quality restoration, improving the quality and efficiency of video creation.
• e-commerce designAI productivity tools such as Meitu Design Studio use AI product images, AI models, and AI photo editing functions to solve the material pain points of the e-commerce industry in shooting, design, and video creation.
• Voiceover video productionApplications like Kaipai utilize AI scripts and AI removal features to help self-media creators easily and efficiently produce spoken videos.
• Advertising and MarketingIn the advertising industry, it covers the entire workflow from creative brainstorming to offline previews, improving the efficiency of advertising material production.
• gaming industryIt provides functions such as scene design, character design, and prop design, expanding the imagination of designers and helping the game industry reduce costs and increase efficiency.