CLIP Interrogator answers a question every image-generator user asks: what prompt would make a picture like this? It analyzes an existing image and reverse-engineers a text prompt for it, combining OpenAI’s CLIP and Salesforce’s BLIP models to describe content, medium, style, and artist influences.
The tool is open source on GitHub, where it has earned nearly 3,000 stars under the MIT license. It works in two stages: BLIP generates a caption describing the image, and CLIP then ranks and selects the most fitting artistic terms, mediums, movements, and trending styles from curated lists, assembling them into a prompt optimized for text-to-image models like Stable Diffusion. It is built on OpenCLIP, so it supports many pretrained CLIP models; the project recommends ViT-L-14 for prompts targeting Stable Diffusion 1.x, with version 2 improving quality and a newer line adding BLIP2 support. You can run it as a Python library with a few lines of code, in Colab notebooks, on Hugging Face or Replicate, or inside Stable Diffusion Web UI through its dedicated extension, which made it a fixture of the early Stable Diffusion workflow: generate or find an image you like, interrogate it, then remix the resulting prompt. Development has slowed since 2024 as generation models gained native image understanding, but CLIP Interrogator remains the reference open implementation of image-to-prompt and is still widely embedded in pipelines.
CLIP Interrogator is free and open source under the MIT license. Hosted demos exist on third-party platforms, which may have their own usage terms, but the code itself costs nothing.
It is for Stable Diffusion artists, prompt engineers, and creative tool builders who want to deconstruct reference images into reusable prompts, and for developers adding image-to-prompt features to their own applications.