InstaCaption

Upload any image, or choose to use a live webcam version, into a caption!

Built on BLIP, an open image-captioning model from Salesforce, run locally via Hugging Face transformers. No API key required, and after the first run (which downloads the model, ~1GB) it works fully offline.

Setup

1. Python 3.9+ is recommended
2. Install dependencies: bash pip install -r requirements.txt
   • If you have an NVIDIA GPU and want faster inference, you can install a CUDA-enabled build of PyTorch first by following https://pytorch.org/get-started/locally/, then install the rest of the requirements.
3. The first time you run either script, it will download the BLIP model (a few hundred MB to ~1GB depending on variant). This only happens once.

Usage

GUI app

python instacaption.py

Opens a window with two buttons:

• Caption your Image — opens your file browser, click on any image to generate a caption
• Live Webcam — opens a webcam window with a live-updating caption overlaid on the video. Press q in that window to stop.

Note: while live mode is running, the main InstaCaption window will look unresponsive, it goes back to normal as soon as you press q.

Or run a single mode from the command line

If you only ever want one mode and prefer a CLI:

python caption_image.py path/to/photo.jpg      # caption a single image
python caption_image.py                        # ...or opens a file picker

python caption_live.py                         # live webcam captioning
python caption_live.py --camera 1              # use a different webcam
python caption_live.py --interval 1.0          # caption more frequently (default: every 2s)

Captioning runs a bit slower than real-time video, so the live mode updates the caption periodically (every --interval seconds) rather than on every single frame — this keeps the video feed smooth while still refreshing the caption regularly.

Project structure

InstaCaption/
├── instacaption.py      # GUI entry point: buttons for image mode and live mode
├── captioner.py         # Core model-loading + captioning logic (shared by both modes)
├── caption_image.py     # Image-captioning logic; also runnable standalone from the CLI
├── caption_live.py      # Live webcam captioning logic; also runnable standalone from the CLI
├── requirements.txt
└── README.md

Notes & tips

• Better quality captions: swap MODEL_NAME in captioner.py from Salesforce/blip-image-captioning-base to Salesforce/blip-image-captioning-large for noticeably better captions at the cost of a larger download and slower inference.
• Speed: on CPU, expect roughly 0.5–2 seconds per caption depending on your machine. A GPU will be much faster.
• Custom prompts: Captioner.caption() accepts an optional prompt argument (e.g. "a photo of") to help steer the caption's phrasing.

Built With

• python