MiniGPT-4 is a powerful tool that improves the comprehension of vision and language by integrating a fixed visual encoder with a fixed large language model (LLM) through a single projection layer. This tool has the ability to generate elaborate descriptions of images- transform handwritten drafts into websites- compose stories and poems inspired by specific images- offer solutions to problems depicted in images- and instruct users on cooking techniques based on food photos. Moreover- MiniGPT-4 is highly efficient in terms of computational resources- as it only requires training the linear layer to align the visual features with the Vicuna using approximately 5 million paired image-text examples.