Multimodal content chunking in AWS

For multimodal content (audio, video, images), chunking behavior differs from text documents:

Nova multimodal embeddings: Chunking occurs at the embedding model level. You can configure audio and video chunk duration from 1-30 seconds (default: 5 seconds). For video files, only the video chunk duration applies, even if the video contains audio. Audio chunk duration only applies to standalone audio files.

Bedrock Data Automation (BDA) parser: Content is first converted to text (transcripts and scene summaries), then standard text chunking strategies are applied to the converted text.