How Zush Analyzes Audio and Video
Zush combines the evidence available in a media file to suggest a name. It does not need to understand every moment of a recording, and a filename is not a full transcript.
Audio files
Section titled “Audio files”Zush can use embedded track metadata, an available transcript, music recognition, or speech analysis. The route depends on the file and your selected AI mode.
For cloud speech analysis, Zush uses an excerpt from the first two minutes of an audio file. A long recording may cover topics outside that excerpt, so review the suggested name before applying it.
Music recognition can identify a track when there is a reliable catalog match. It is not guaranteed for every recording, remix, live performance, or noisy clip. Zush can use metadata or other available context when no match is found.
Video files
Section titled “Video files”Video naming can combine:
- sampled frames for visible subjects and scenes
- embedded metadata
- readable subtitles, including supported subtitle tracks in MKV files
- a speech excerpt from the first five minutes, when the cloud route supports it and speech is detected
- music recognition when available
A matching song identifies the audio source; it does not establish who is visible or where the video was recorded. Review those details in the proposed name.
Cloud, BYOK, and local differences
Section titled “Cloud, BYOK, and local differences”| Mode | What to expect |
|---|---|
| Zush Cloud AI | Managed analysis can use bounded audio excerpts for transcription, alongside visual or text evidence. |
| BYOK | Media handling depends on the provider. Gemini can process supported audio directly; supported OpenAI, OpenRouter, and Anthropic routes use a transcription step. |
| Ollama or LM Studio | The local model uses available text, metadata, and image previews. Zush does not send raw audio to these local models or switch to cloud transcription automatically. On-device speech support varies by platform and available components. |
Local modes also skip online music catalog recognition. If a local result lacks the spoken subject, try adding a transcript or use the available metadata in your Naming Blocks.
Build a useful media workflow
Section titled “Build a useful media workflow”- For recordings and clips, start with Title and add a date or project only when useful.
- For a library of films and episodes, use the Movies & TV Template guide.
- For repeatable names, save the pattern as a Template.
- Test representative files before assigning a media folder to Monitor.
See Supported file types for containers and the codec caveat, and Privacy and security for what cloud processing sends.