Experience

• Technical Artist at Tencent

  Shenzhen, Guangdong, CN

  July 2024 - Present

  Presenting our team's open-source project,
  
  Tencent HY 3D Large Model
  
  2.1:
  
  https://github.com/Tencent-Hunyuan/Hunyuan3D-2.1
  
  2.0:
  
  https://github.com/tencent/Hunyuan3D-2
  
  1.0:
  
  https://github.com/Tencent/Hunyuan3D-1

  Tencent HY World
  
  1.1:
  
  https://github.com/Tencent-Hunyuan/HunyuanWorld-Mirror
  
  1.0:
  
  https://github.com/Tencent-Hunyuan/HunyuanWorld-1.0

  Welcome to follow the project and give it a star on the GitHub repository 🌟

• Digital Human Expert (TA) at ByteDance

  Shenzhen, Guangdong, CN

  September 2023 - July 2024

  Creating the "Cartoon Face" special effect that over 9 million people on TikTok are using:
  
  https://mp.weixin.qq.com/s/TN7D9_im-4G9-zASAjlv-Q

  • 3D GAN - 3D Graphics + GAN/AIGC, a complete and reusable automated technical pipeline
  
  - Participate in building an automated 3D face reconstruction workflow. Leverage the high-quality advantages of the offline engine through the AI rendering pipeline to achieve film-level special effects on mobile devices.
  
  - Achieve real-time facial driving from real-life people to 3D cartoon-stylized models, reaching the top-level in the industry in terms of real-life people's lighting, skin color texture, facial expression flexibility, personalized appearance, versatility, and stability, surpassing competitors like Snapchat.
  
  - Participate in building a designer-centered procedural rendering pipeline, including art asset inspection and construction, rigging, expression blending, curve mapping etc. Automate 80% of the process, liberating R&D manpower while maximizing the freedom of art design.

  • Establish a comprehensive Blendshape production specification to provide a unified standard for the project, strike a balance between the anatomical rationality and stylistic aesthetics, and improve the controllability and consistency of expressions.

  • Build an ARKit annotation tool for a 200,000-dataset to visualize data-driven standard head models. Design an intuitive and easy-to-use interface to enable designers to participate in data annotation and help optimize the expression coefficient estimation algorithm.

• Technical Artist at Microsoft (Xiaoice - ICEGamer)

  Beijing

  February 2023 - July 2023

  • Collaborated with the AI team to establish and optimize the UDE facial animation workflow, minimizing dependence on motion capture and manual artistic intervention. Addressed discrepancies between FLAME 3DMM and game model specifications. Implemented automatic cross-model blend shape transfer to integrate driving signals and automated processes for data optimization and lip morphology adjustment. Supported character driving across any model.

  • Spearheaded the sound-driven animation system in coordination with development, game design, art, and AI teams. Leveraged driving rule configuration and action decomposition and reconstruction mechanisms to generate infinite actions from a finite set, seamlessly harmonizing with all AI-generated dialogue content and ensuring natural and sophisticated motion synthesis.

  • Developed the format conversion pipeline to automate and standardize the transformation of animation assets, ensuring uniform data dimensions and minimizing errors. Seamlessly integrated it with AI training processes to facilitate AI-driven character animations.

  • Defined comprehensive guidelines for creating facial expressions and visemes. Integrated the Faceit plugin into the artistic production pipeline to automate the generation of 52 blend shapes conforming to ARKit specifications. Designed conversion rules between 3ds and Blender formats based on game model requirements.

  • Worked closely with programmers to formulate rules governing eye movements, utilizing blueprint configurations to achieve realistic gaze, blinking, and eye jitter effects. Enhancing the naturalness and randomness in eye expressions.

• Technical Artist at Surreal AI

  Shenzhen, China

  July 2022 - November 2022

  Unreal Engine/ Python/ Blender/ Metahuman/ iClone/ Omniverse

  • Designed and built a simple and lightweight rendering service enabling the AI team to quickly generate character animations from action tags, and wrote complete technical documents

  • Built a batch retargeting system suitable for different characters, integrated multiple animations into single character in single scene through animation layers, and unified the skeleton hierarchy and the remapping preset of source animations. Ensured the stability of the entire pipeline >99%

  • Cooperated with AI team to develop an "Audio2Blendshape" algorithm and summarized an automatic method to optimize facial ctrlRigs. Improved CG avatars' lipsync accuracy (better than NVIDIA's Audio2Face)

  • Fully upgraded the naturalness of 20+ CG avatars on Movio (https://www.movio.la/). Improved mouth animations, expressions, micro-expressions, eyes, and body movements to an extremely real-like level
  
  - By analyzing Metahuman SDK, summarized a function that enables CG avatars to switch expressions through linear transformation, and implemented to product
  
  - Analyzed the correlation between the movements of various body bones to increase randomness to the preset animation while ensuring reasonable movements
  
  - Designed an animation smoothing workflow with controllable smoothness, which also prevents the loss of details (micro-movements) while smoothing the animation

• Technical Artist at TG3D Studio

  Taipei, Taiwan

  August 2020 - May 2022

  Python/ 3ds Max/ Blender/ ZBrush/ Ornatrix

  • Avatar Creator
  
  - Designed and built solid pipelines for creating 3D customized avatars by user-defined genders, styles, and body types, improving modeling efficiency by 10+ times
  
  - Collaborated with engineers to implement an AI-based size prediction algorithm, and developed an iterative mechanism of partial deformation to generate a body model with 99+% accuracy
  
  - Resolved auto-measuring errors and overlapping deformation issues by unifying the topology and delivering an optimized normalized weight map
  
  - Customized NRE models for fashion companies (SST&C, STAMP, SPORTSKITE, etc.) contributing 10% to TG3D 2021 revenue

  • Created a virtual model with photorealistic shader (by Substance; rendered by Arnold), 3d hair (by Ornatrix), and facial rigging system (by Rigify)

  • Managed TG3D collection (200+ mannequin models of different sizes, styles, and regions) and optimize the workflow integrating 3ds Max and Zbrush for making standardized models with consistent topology

  • Developed extensible plugins and tools supporting body modularizing for TG3D’s Metaverse (low2high poly converting, auto shoe-wearing, topo symmetry, auto-rig, IKFK advanced rigging system integrated with Mixamo and Rigify)

  • Developed a simplified version of VTO (virtual try-on), increasing cloth simulation speed by 10+ times

• Intern at TG3D Studio

  Taipei, Taiwan

  July 2018 - January 2019