Video Comprehension Plugin

KiraAI video comprehension plugin: automatically detects QQ video messages, compresses them, extracts frames at three levels, and analyzes them as a collage; speech transcription (Bilibili official subtitles take priority, and it can be skipped when the model has built-in audio and video understanding); supports Bilibili link download, sending, search, AI summary, and html5 direct-link passthrough; 4 independently configured model groups (native video / frame extraction / built-in audio and video understanding); supports local videos and time-range analysis.

Core Capabilities

1. Automatic video message detection: recursively scans the message chain (including quotes Reply.chain and merged forwards Forward.chains, with multi-level nesting supported) → raw_message → proactively calls get_msg by message ID, as a three-level fallback
2. Automatic caching of received videos: downloads a video locally as soon as it is detected, and tells the bot the path in the next conversation turn (auto_cache_video, enabled by default; the caching scope is controlled by cache_scope, which defaults to caching only videos that were…