Inspiration
Anime, idol culture and AI tinkering leading to wanting to make real anime girl (and boy) companions
What it does
Provides a way to use LLMs through a talking AI companion with realistic voice and have it become an immersive real experience.
How we built it
Combining Elevenlabs, Mistral, Perplexity and FastAPI along with three.js into a full application where we animate Vroid (vtuber) models with speech. Also includes an AI memory system.
Challenges we ran into
Issues with engineering saving of memories, issues with web services running differently to local processes, some CORS issues and slow voice processing making it difficult to get the whole process running. Immense difficulty with timing of mouth movements.
Accomplishments that we're proud of
Getting the models running live and talking as a full integration at https://ai-waifu-platform.onrender.com/
What we learned
Learned about how to create realistic mouth movements and overall engineer a feeling of realness through the models. Also how to construct AI memory.
What's next for Waifu Maker
Add a subscription model for voice usage and a cheaper voice model as a part of the live app. Make it available as a beta for people to use and play with.
Built With
- elevenlabs
- fastapi
- javascript
- mistral
- perplexity
- python
Log in or sign up for Devpost to join the conversation.