A system that turns existing video media into 'tastable video'
Since the invention of video technology, humanity has recorded various footage. Recently, with the development of gustatory media, it has become possible to record and reproduce taste along with video, but taste cannot be added to previously recorded footage. This paper proposes a system that estimates and adds taste to videos without recorded taste, similar to colorizing black-and-white footage. Taste is estimated for all scenes using GPT-4 Vision. The prompt instructs it to first infer the name, ingredients, and amount of food/drink, and then estimate the taste, enabling high-accuracy estimation through step-by-step reasoning. Viewers can specify food in the video and output the corresponding taste. This system makes existing video media 'tastable.'
The GitHub link above hosts a read/write library for FlavMP4, which can also record taste data.