CaptionChameleon is a containerized batch processing system for automatically generating video captions in multiple distinct styles. This project leverages zero-shot learning with advanced vision-language models (such as Qwen3-VL) to caption videos without requiring any model fine-tuning or few-shot examples. The system processes videos by: Sampling frames: Intelligently extracting 8 frames from each video for efficient processing Batch style generation: Generating all 4 caption styles in a single API call per video, reducing latency Parallel processing: Using a thread pool executor to process multiple videos concurrently Multi-platform deployment: Support for both local execution and containerized Docker deployment The four caption styles are carefully tuned for different audiences and contexts: Formal: Professional, objective descriptions suitable for technical documentation Sarcastic: Dry, witty commentary with ironic undertones Humorous Tech: Programming and DevOps humor for technical audiences Humorous Non-Tech: Everyday humor accessible to general audiences This architecture enables fast, cost-effective video captioning at scale without the overhead of model training or fine-tuning, making it ideal for rapidly processing large video libraries.
Category tags: