Search Your Timeline by Words and Pictures, Not by Scrubbing
Type what was said or describe what was on screen. VoxStudio for Mac blends exact-word, meaning and visual search to put the playhead on the right moment in a long edit.
By David, Product Specialist at VoxStudio · · 7 min read
Key takeaways
- Find a moment by the words said, the meaning of what was said, or what appears on screen.
- Exact words answer instantly; smart search adds meaning and visuals from a local model.
- Click a result and the playhead lands in the middle of that moment, ready to cut.
The problem: you remember the moment, not the timecode
Somewhere in a 40-minute lecture, the speaker says the line that should open your clip. Or there is a slide with a chart you need as B-roll. You know it exists. You do not know where. So you drag the playhead, listen for three seconds, drag again, and lose ten minutes.
Long timelines break three habits at once. Scrubbing is slow because the answer could be anywhere. Thumbnails are too small to read. And plain text search fails the moment your memory and the transcript disagree: you remember "cost", the speaker said "pricing"; you remember a whiteboard sketch, but nobody described it out loud.
What timeline search removes
- Scrubbing and re-listening — Type a few words and jump straight to the candidates instead of sampling the timeline in 3-second bites.
- Wording mismatches — Describe the idea ("the part about shipping delays") and find lines that say it differently.
- Silent visuals — Find a slide, chart or whiteboard moment that nobody mentions in the audio.
- Tool hopping — The search runs inside the editor, on the clips already on your timeline, so there is no export or second app.
The 30-second workflow
Open the magnifier in the timeline toolbar. Start typing a word, a phrase or a description such as "drawing a chart on the whiteboard". Results appear as a list, each with its time range and the subtitle text around it.
A small badge on each row tells you why it matched: a speech bubble means the words appear in the subtitles, a film icon means it matched what is on screen, and sparkles mean both. Click a row and the playhead jumps to the middle of that range. From there you can trim, cut or add a marker without ever scrubbing.
In the example, searching "newton" lists the line "Newton wasn't wrong" first with the sparkle badge, because the phrase is both spoken and shown on a slide, followed by lines that only mention Newton in speech. That agreement between ears and eyes is a strong signal, so those moments rise to the top.
Habits that make it faster
- Search short, then describe — Start with one distinctive word. If you only remember the idea, write it as a short sentence instead of a keyword.
- Transcribe first — Speech search works on the subtitle clips on your timeline. Bring in a reviewed session or generate captions, then search.
- Let the first index finish — The first time a video is analyzed, on-screen matches can be incomplete. The popover says so while it works. Exact subtitle matches are available immediately.
- Search the cut you are editing — Results come from clips on the timeline, so trimmed-away footage and hidden tracks do not clutter the list.
How it finds a moment: three searches, one list
Behind one search box there are three different ways of matching, because each fails where another succeeds. Here is the plain-language version of what VoxStudio does.
Three signals, three blind spots
| Exact words | Meaning (semantic) | On-screen (visual) | |
|---|---|---|---|
| Matches | The phrase or its terms, ignoring case and accents | Subtitle lines with a similar idea | Video frames that look like your description |
| Great for | Names, numbers, quotes | Paraphrase: "cost" vs "pricing" | Slides, charts, whiteboards, scenes |
| Misses | Anything worded differently | Rare names the model never saw | Text too small to read in a frame |
| Needs | Subtitles on the timeline | Smart search model on your Mac | Smart search plus a one-time video index |
1. Exact words: instant and always on
Exact matching runs first and needs no model. The whole query as a phrase counts as a full match. If it is not found as a phrase, the editor scores each subtitle by the share of your words it contains. Case, accents and character width are ignored, so "Newton", "newton" and "NEWTON" all agree. This is the layer that keeps proper nouns and numbers reliable.
2. Meaning: text turned into numbers
Smart search uses an embedding model that runs on your Mac. An embedding is a list of numbers (256 of them here) that places a piece of text in a "meaning space": sentences about similar ideas land close together, even when they share no words. Your query and every subtitle line become points in that space, and the editor ranks lines by how close they are.
To keep noise down, a line must clear a minimum similarity and sit near the best score. A small bonus for exact wording means a literal match outranks a loose paraphrase when similarity is otherwise equal. Subtitle embeddings are cached while you work, so repeated searches are fast.
3. On-screen: frames in the same space
The same model understands images as well as text, so a frame and a sentence can be compared directly. When a video is imported, VoxStudio samples frames in the background, roughly every couple of seconds, and detects scene changes by comparing a coarse brightness grid between samples. A long static shot is still sampled periodically, so it stays searchable. Each kept frame becomes one embedding saved on your Mac, once per file.
At search time your text is embedded and compared with those frame vectors. Only the best frame of each shot counts, so one long scene cannot flood the list. Source-video times are then mapped onto your timeline, accounting for trims and speed changes, so a result lands where the shot is in your edit rather than in the original file.
Combining the lists: reciprocal rank fusion
Scores from different methods are not comparable: a text similarity of 0.6 and a frame similarity of 0.6 mean different things. So the editor ignores raw scores and merges by rank using reciprocal rank fusion. A result earns 1 / (60 + rank) from each list it appears in, and the totals decide the order. Something ranked first in one list can still lose to something that is good in both. If a matching shot overlaps a matching subtitle, they are merged into one result with the sparkle badge instead of being listed twice. The popover shows up to 30 moments.
What stays on your Mac
Smart search relies on a local model that you download once (about 2 GB) and enable from the popover. The model, the frame index and the subtitle vectors all live on your Mac. Until the model is ready, the popover shows exact subtitle matches and says so, with a button to download and enable smart search. Frame analysis pauses while an export is running, so it does not compete with your render.
What to expect, honestly
Semantic and visual matching are probabilistic. Expect near-misses, especially for abstract queries, and expect to click two or three rows before you land on the right one. That is still far fewer steps than scrubbing. Exact-word results remain the most dependable for names and numbers, and the first analysis of a long video takes time on a busy Mac.
FAQ
How do I search for a moment in a video timeline in VoxStudio?
Open the magnifier in the timeline toolbar and type a word, a phrase or a description. Click a result to move the playhead to the middle of that moment.
Can I find a scene by what appears on screen, not what is said?
Yes, with smart search enabled. Describe the scene, such as a chart or a whiteboard, and moments that look like it are listed with a film badge once the video has been analyzed.
What is hybrid search in a video editor?
It means combining several ways of matching: exact words in the subtitles, similarity of meaning, and similarity of on-screen content. Their ranked lists are merged so moments supported by more than one signal rise to the top.
Does timeline search upload my video?
No. Smart search uses a model downloaded to your Mac, and the frame index and subtitle vectors are stored locally.
What if smart search is not installed?
You still get exact subtitle matches. The popover tells you that only exact matches are shown and offers to download and enable smart search.
Why do I need subtitles on the timeline?
Searching what was said works on the subtitle clips in your timeline. Reuse a reviewed session or generate captions in VoxStudio first. On-screen matching does not need subtitles.
Explore the Mac workflows
- Offline Transcription for Mac
Transcribe audio and video on your Mac with local speech models, speaker labels and timed captions. Built for Apple silicon and macOS 15 or later.
Spend your editing time editing.
Open a project in VoxStudio for Mac, press the magnifier and jump straight to the moment you remember. Download VoxStudio for Mac
David — David works on the VoxStudio product team and spends most days inside the video editor, turning real customer workflows into faster ones.