The lyricCanvas dataset contains approximately 10M lines of lyrics with corresponding visual elaborations (visualizable prompts).
It could be used to train large language models to translate highly abstract concepts and metaphorical
phrases to visualizable prompts for image generation, see ViPE.
Due to copyright policies, we are not allowed to publish the lyrics, however, we release the visual elaborations and the scraper through which
you can collect the lyrics… See the full description on the dataset page:
https://huggingface.co/datasets/fittar/lyric_canvas.
