Create your Image Dataset with this Caption Tool

Through the use of Ollama or downloadable Huggingface models, you can easily caption all images in a directory using the VLM Caption Server project. It runs as both a command line script and a web service in case you want other programs to interact with it (or just run the web UI to select the options). With this tool, you don't have to select images to upload since it runs on your local file system.

Currently, the models supported are:

• Qwen3-VLM-8B with Ollama

• MiniCPM-V-2.6 8b with Ollama

• Microsoft Florence-2-base-ft through Huggingface transformers (fork)

3 prompts are available for captions:

• detailed: Describe the image in detail in one paragraph.

• short: Provide a brief description of the image within 40 words or less.

• tags: Provide a comma delimited list of tags that describe the image in order of relevance.

Suggestions and feedback are welcome. Comment below or write an issue on the Github project.