"Captions With Attitude" is a Go application that uses a Vision Language Model (VLM) to show live captions from your webcam in your browser all running entirely on your local machine!

It uses yzma to perform local inference using llama.cpp and GoCV for the video processing.

You must install yzma and llama.cpp to run this program.

See https://github.com/hybridgroup/yzma/blob/main/INSTALL.md

You must also install OpenCV and GoCV, which unlike yzma requires CGo.

See https://gocv.io/getting-started/

Although yzma does not use CGo, yzma can co-exist in Go applications that use CGo.

You will need a Vision Language Model (VLM). Download the model and projector files from Hugging Face in .gguf format.

For example, you can use the Qwen3-VL-2B-Instruct model.

https://huggingface.co/ggml-org/Qwen3-VL-2B-Instruct-GGUF

go build .

$ ./captions-with-attitude 

Usage:
captions-with-attitudes
  -device string
        camera device ID (default "0")
  -host string
        web server host:port (default "localhost:8080")
  -model string
        model file to use
  -p string
        prompt
  -projector string
        projector file to use
  -v    verbose logging

./captions-with-attitude -model ~/models/Qwen3-VL-2B-Instruct-Q8_0.gguf -projector ~/models/mmproj-Qwen3-VL-2B-Instruct-Q8_0.gguf

Now open your web browser pointed to http://localhost:8080/