I’m excited to share my project, MiKaPo (MikuMikuDance Pose Picker) , a web-based tool that automatically poses MMD models from video input in real-time. The tool currently supports pose and face detection, with both video upload and camera input options.
The tech stack includes BabylonJS for 3D rendering, babylon-mmd for MMD model support (big credits to @noname0310), Mediapipe for 3D keypoints detection, and Vite+React for the web framework. The key challenges revolve around efficiently controlling MMD models with 3D keypoints detected by Mediapipe, and optimizing performance for real-time updates.
My current plan is to develop an Electron version for desktop with more advanced functions like user model selection, animation export, and Ollama integration. I’m looking forward to hearing your perspectives on this, as well as any feedback, new function proposals, or suggestions on future deployment platforms as we explore the potential of this project together!