qwen35-timeframe-captioner
Applies a Qwen3.5 vision-language model to video frames selected by input TimeFrame annotations for prompt-driven captioning / scene description. Per-TimeFrame captioning is composite: every frame sampled from a TF is fed to the model in a single prompt and yields one caption per TF. A model runtime parameter selects the Qwen3.5 variant (default Qwen/Qwen3.5-2B); larger variants need substantially more VRAM (see metadata.py for the GPU/VRAM sketch). Qwen3.5 is a reasoning model – this app runs it for captions and strips any <think> reasoning block from the output.