Video you can ask questions to
A two-hour recording holds one answer you need, somewhere. Nova AI builds a local index of the transcript, the scenes and the text on screen, then lets you find that answer in a sentence instead of a scrub bar.
Nova AI is not a chatbot bolted onto a player
It lives in a collapsible panel on the right of the player, with tabs for Ask, Search, Chapters, Summary and Translate. Opening it never stops playback, and every result it gives you is a timestamp you can click.
- Reachable from the sidebar, the player toolbar or ⌘⇧A
- Suggested actions instead of a blank prompt box
- Command palette: “What would you like Nova to do?”
- Playback continues while analysis runs in the background
Ask This Video
Type a question the way you would ask a friend who had already watched it.
You ask
- “What did he say about Apple?”
- “Where do they discuss artificial intelligence?”
- “Explain this section.”
- “Summarise the last ten minutes.”
- “Where do they mention Ferrari?”
Nova answers
A short explanation, then the moments it came from:
- 12:14 — Ferrari history
- 28:51 — Racing technology
- 43:06 — Ferrari versus Porsche
Click a timestamp and the player is already there.
Finding things
Three different ways to locate a moment, because not everything you remember is something that was said.
Visual search
Describe what you saw. “Find the red car.” “Show scenes with mountains.” “Find the restaurant scene.” Results come back as a thumbnail, a timestamp and a short description.
OCR search
Nova reads the text inside the frames. Find the slide about neural networks, the moment a domain appears on screen, or the exact error message in a screen recording.
Scene detection
Nova identifies meaningful scene changes and lays them out as thumbnails, so you can browse a video with your eyes instead of dragging a playhead.
Understanding structure
Magic Timeline
Nova works out how a video is organised and draws those chapters onto the scrub bar. A documentary becomes Intro, History, Technology, Comparison, Conclusion. A tutorial becomes Introduction, Installation, Configuration, Example, Troubleshooting, Summary.
Chapter generator
Export those chapters as real timestamps:
- 00:00 Introduction
- 04:32 History
- 17:45 Technical explanation
- 32:11 Demonstration
- 48:20 Conclusion
Smart Timeline Preview
Hovering the timeline normally gives you a thumbnail and a time. Nova adds a line of context — “Discussion about HEVC video compression” — so you know whether that is the bit you wanted before you commit to the jump.
Video summary
A quick summary or a detailed one, plus key points, topics, chapters, speakers and the important moments. Every moment is clickable.
Language
Speech recognition first, then translation as a separate stage — which means any recogniser can pair with any translator.
Live subtitles
Nova listens to the audio and produces timed, synchronised subtitles for footage that never had any. Export them as SRT or VTT and keep them.
Live translation
Transcription feeds translation, which feeds the subtitle renderer. Show the original, the translation, or both at the same time. Source language is detected automatically.
Live Dub
Nova can translate the spoken dialogue and generate an audio track in another language, either live or as an offline pass.
Nova does not clone real people's voices. Dubbing uses generic, selectable voices.
Smart Watch
Two hours of footage, twenty-four minutes of substance. Nova assembles a virtual playback sequence from the sections that matter and skips the rest.
- Three depths: Quick, Balanced and Detailed
- Leave Smart Watch at any point and carry on normally
- Your original file is never edited, trimmed or re-encoded
Smart Audio
Processing aimed at the two complaints everyone has about modern mixes — the dialogue is too quiet and the action is too loud.
- Dialogue enhancement for speech clarity
- Noise reduction
- Voice isolation
- Smart volume, balancing quiet talk against loud scenes
- Night mode that softens sudden peaks but keeps speech intelligible
Casting assistant
“Why won't this play on my TV?”
“Your TV supports the HEVC video track but not the DTS audio track. Nova can convert the audio to AAC without re-encoding the video.”
Your keys, your choice of model
Nova does not resell AI. You supply your own API key, or skip the cloud entirely and run a model on your Mac. Either way Nova is not standing between you and the provider taking a margin.
Cloud providers
The provider layer is modular, so Nova is never welded to one vendor.
Keys are stored in the macOS Keychain and are never written to logs.
Local AI
When a model runs on your machine, Nova says so plainly: Processed entirely on your Mac.
Privacy, stated before anything happens
On Device or Cloud
Every AI feature displays its processing mode before it runs. A cloud provider must declare that it uploads user media, which triggers an explicit consent prompt.
The minimum, never the file
Nova sends only what the task needs — an audio segment, a handful of frames, the transcript, the surrounding context. A whole video is never uploaded unless it is genuinely unavoidable and clearly required.
Your data, deletable
Settings → Nova AI → AI Data lists indexed videos, transcripts, thumbnails, embeddings and the cache size, with one button to clear it. Your media is never touched.
The intelligence index
Analysis results — transcript, timestamps, scenes, OCR, chapters, topics, summaries, embeddings — are cached locally and reused. Nova does not re-analyse a video that has not changed, so you pay for the work once.
When analysis runs
- Never analyse automatically
- Analyse when Nova AI opens — the default
- Always analyse
“I didn't know a video player could do all of this.”
That is the reaction Nova is built for.