This uses OpenAI’s CLIP model which is open source: https://github.com/openai/clip