Wenlan Wang
Computer vision algorithm expert at Xiaobing
Wenlan Wang, currently a computer vision algorithm expert at Xiaobing, leads the development and innovation of X Eva APP vision algorithm solution, and developed AIGC and face-driven technology applied to the visual expression of AI beings in X Eva APP, which contributed to the landing of video calls, shooting the same product features. Prior to this, Wang served as the person in charge of AIGC technology direction in Tiger Tooth, leading the team to develop AIGC technologies such as industrialised material production and scene structuring, which are widely used in Tiger Tooth Assistant and Tiger Tooth Live APP, with a penetration rate of more than 30%. He has shared his research and development results in many top industry conferences on AI and gaming, with an audience of more than 13,000, and has published several technical white papers, patents, and theses.
Topic
Visual innovation with Ice AI Beings: challenges and breakthroughs in personalised expression
Transformer technology has led to a second wave of development in AI, particularly in the area of image and video generation.The Latent Diffusion Model (LDM) has led the way for a series of image generation models and developer communities led by Stable Diffusion (SD), and OpenAI's LDM-based adaptation of the Diffusion Transformer video generation model Sora shocked the world, and Diffusion has become the training method of choice for image and video generation. Xiaobing has made in-depth exploration in this technology field, through SD technology to achieve accurate "shoot the same style" function, while taking into account the beauty and image, loved by users, to meet the demand for personalised visual content. In this talk, I will share our practical experience in this project, including a series of technical practices such as base model evaluation, the best automated training map screening strategy, LoRA training details, unique workflow design, and then template production and personalised photo generation. Through experiments, it is proved that the different implementation of each stage will generate significantly different results, and how to design and balance become the key elements affecting the final output. The sharing will cover the challenges encountered in the project, solutions, key technical points, as well as thoughts on future technology trends. We are committed to providing AI technology researchers and application developers with in-depth experience exchange and reference, and jointly promoting the application and development of AIGC in the visual expression of AI beings.
