Inspiration
Our inspiration for Groove Genie was to use technology to make dance accessible to anyone and promote self-expression. We wanted to build a website that could generate unique dance choreographies based on a song input and encourage individuals to express themselves through movement.
What it does
Groove Genie is a website that generates unique dance choreographies algorithmically based on a song input. Users can input any song through a YouTube link, and the website will generate a choreography that matches the energy and rhythm of the music.
How we built it
Groove Genie is powered by Google Research's AIST++ Dance Dataset and OpenAI's music LLM Jukebox, which together provide meaningful latent spaces that cluster dances appropriate for any input music, which can be combined to generate dance choreographies based on a song's energy, rhythm, and style. Given these embeddings, we are able to traverse through the AIST++ dataset and construct novel dances by stringing together dances with similar audio (as judged by Jukebox), a technique called motion matching that we leverage from the motion generation domain.
Despite its comparative simplicity and computational efficiency, our approach is capable of generating coherent, stylistically appropriate choreographies, and represents a step forward towards creative AI tools in the relatively unexplored domain of dance. By combining the strengths of neural audio embeddings and motion matching, our approach achieves high performance while necessitating minimal computational resources.
To make this system widely accessible, we've built a user-friendly interface: just type in a song from Youtube, and watch Groove Genie generate amazing dance choreographies!
Challenges we ran into
One of the biggest challenges we faced was working with large amounts of data, particularly with the AIST++ Dance Dataset, which contains over 10 million frames of 3D keypoints. In order to efficiently process this data, we sped up inference on the Jukebox model by casting to FP16 and batching.
Accomplishments that we're proud of
We're extremely proud of the user interface we've developed, which is both easy to use and visually appealing. We're also thrilled with the quality of the generated choreographies, which are often surprising and innovative.
We're also proud of the efficient algorithms we've developed to process large amounts of data and generate coherent, stylistically appropriate choreographies in real-time.
What we learned
We learned that dance is a universal language that can transcend cultural and language barriers. It can bring people together and foster a sense of community. We also learned that technology can be used to make dance accessible to more people and promote self-expression.
What's next for Groove Genie
In the future, we plan to improve our algorithm and expand the dataset of dance styles to generate even more unique and diverse choreographies. We also plan to collaborate with dancers and choreographers from different cultural backgrounds to promote and preserve cultural heritage through dance. In the future, we hope to make the algorithm more interactive such that the users can provide more detailed feedback and input for a truly collaborative process.
Log in or sign up for Devpost to join the conversation.