This is an interactive dashboard application that was built as part of a human-computer interaction study at Dartmouth. It can interpret user's drawings and translate them into an endless stream of music. This interactive interface allows students to express emotions through drawings, offering a unique fusion of art creation and music generation leveraging advanced AI technologies.
For more details, see our research paper and demo video.
The frontend can be installed with bun. Simply:
cd frontend
bun install
Then, to start the frontend:
bun run dev
This will start an instance of the frontend on port 5173. This won't actually work for generating music without the corresponding backend, but can be used for testing/demoing the canvas and other strictly frontend functionality.
There are two parts to the backend: the image-to-text server and text-to-music model. They need to be run separately, and operate on different ports.
Text-to-music: First, the text-to-music model. Install the requirements:
pip install transformers flask flask_cors
pip install nltk
pip install torch torchaudio
Then, the text-to-music model can be started by simply running:
cd backend
python musicgen.py
Note that this will download a rather large model the first time the script is run. By default, the script will use the musicgen-stereo-large model, open-sourced by Meta and available on the HuggingFace Hub here.
Image-to-text:
There are two ways to run the image-to-text model. You can either use a local model that's compatible with llama.cpp, or use the OpenAI GPT-4V API if you have an account with access (and a corresponding API key). A good local model to use is BakLLaVA, quantized to Q5_KM. You can follow the instructions in the llava-cpp-server README to download and get set up.
To run the model locally (assuming it's already been downloaded):
cd backend/llava-cpp-server
git submodule init && git submodule update
make
bin/llava-server -m ggml-model-q5_k.gguf --mmproj mmproj-model-f16.gguf
Alternatively, you can use the OpenAI GPT-4V API. To start that server, add your API key to openai_server.py. Then, simply:
python backend/openai_server.py