What is image-to-prompt?

Image to Prompt is a local web app that takes an image you upload and converts it into a structured JSON prompt formatted for Ideogram 4, an AI image generation service. The idea is to reverse-engineer an image into a description that you could feed back to an image generator to recreate or remix it. The analysis is done by a Microsoft AI model called Florence-2 that runs on your own machine, with no data sent to external servers. Florence-2 reads the image and returns a scene caption, object locations, region descriptions, and any text it finds in the image (using optical character recognition). The app turns all of that into editable boxes on a canvas so you can adjust what gets included in the final prompt before exporting. You can drop multiple images at once and they process one at a time in a queue. A row of thumbnails at the top shows which images are done, still queued, or failed. You can switch between images and keep editing while the queue runs. Each image keeps its own set of boxes and labels independently. When you are finished, you can export a single JSON file or download all finished prompts as a zip. Installation can be done with one click through a tool called Pinokio, or manually by cloning the repository and running a Python setup. The app requires PyTorch, which you install separately with a command that matches your hardware (CPU, Apple Silicon, or NVIDIA GPU). On first launch it downloads the Florence-2 model from Hugging Face, which may take a few minutes. The app also exposes a simple API endpoint that accepts an image file and returns the analysis as JSON. This makes it straightforward to call from JavaScript, Python, or the command line if you want to integrate image analysis into your own tools.