Tutorial

Getting started with MediaFind

From a clean install to searching, asking and exporting your library — in about ten minutes. Everything below runs on your own Mac; nothing you index is ever uploaded.

MediaFind turns a folder of audio and video into something you can actually search — by meaning, by what's on screen, by who's speaking, or by just asking a question. This guide walks through it end to end. You don't need an account, an API key, or an internet connection beyond the first-run model download.

Prefer to poke around first? There's a hands-on interactive demo built right into this guide — sample data, running in your browser.
↓ Jump to the demo
1 · Install pick your models 2 · Add folder audio & video 3 · Index on your Mac Search Ask Export
The whole product is three setup steps and then everything you can do with the result. Steps 1–3 happen once per library.

Try it now — interactive demo

Before installing anything, here's a hands-on sample you can click around right now. It's a small, freely-licensed library that's already indexed, so you can search by meaning, ask a question, jump to the exact moment, and browse people — without adding any media of your own.

This is a guided demo with sample data — not the real app. It runs entirely in this page to show you the workflow. The actual MediaFind app runs on your own Mac and indexes your media; nothing in the demo is uploaded, and nothing you do here touches your files.

Open the demo full screen ↗

Got the feel for it? Here's how to set it up on your own library.

1 · Install & first run

Download the app for your platform from the download page. macOS is signed and notarized; Windows, Linux, and the Android companion are beta builds and may show an OS install warning you can safely dismiss.

The first time you open MediaFind, a short setup wizard walks you through a few one-time choices:

  1. Welcome & ffmpeg check. The macOS app bundles its own ffmpeg, so there's nothing to install — it just confirms decoding works.
  2. Pick a transcription model. Choose a Whisper-class model sized to your machine: tiny/base rip through a backlog, small/medium are the everyday sweet spot, and large is for the hardest audio. You can change this later per library.
  3. Pick an AI assistant model (optional). The on-device language model that powers Ask and summaries downloads once. Skip it if you only want search — everything keyless still works.
  4. Add your first folder. Covered in the next section.
One-time downloads, then offline. The only network activity during setup is fetching the model weights you chose. After that, transcription, search, and Ask all run with the internet unplugged.
Welcome ffmpeg check ✓ Transcription tiny → large AI assistant optional · powers Ask skip & stay keyless First folder audio & video
The one-time first-run wizard. Only the model you pick is downloaded; the AI assistant step is optional and can be added later from Settings.

2 · Add a folder — or try the samples first

Point MediaFind at any folder of audio or video. It handles the common container and codec formats, indexes what it finds, and keeps watching the folder so anything you drop in later gets picked up automatically.

Not ready to point it at your own files? On the welcome screen, “Try MediaFind on real videos” loads a handful of bundled, freely-licensed sample clips so you can explore every feature before adding a single byte of your own media.

If you want to…Do this
Test-drive the appClick Try MediaFind on real videos to load the sample clips
Index your own libraryClick Add folder and choose a directory of audio/video
Pull something off the web PROUse the Media downloader to fetch a public URL

3 · Let it index — locally

Once a folder is added, MediaFind gets to work on your machine. For each file it transcribes the audio with word-level timestamps, embeds the transcript for semantic search, extracts keyframes, runs OCR on on-screen text, detects scenes and objects, labels speakers, and groups everything into chapters.

You don't have to wait for the whole library — files become searchable as they finish, and a progress indicator shows what's left. Indexing is parallel across files, so it scales with your CPU cores.

Get the transcript right once and every other feature — search, Ask, chapters, speaker labels — inherits the timing for free.

One file audio / video Transcript · word timestamps Semantic embeddings Keyframes · OCR text Scenes · objects · speakers Faces · chapters · summary Searchable on your Mac file-by-file, in parallel
Every file fans out into many signals during indexing — all computed locally, all sharing the transcript's timing. Files become searchable as they finish.
MediaFind — your library
The MediaFind home screen showing an Asset Library: a search box, file and people counts, category chips, a People panel, and an AI Insights rail with a library overview.
Once a folder is indexed, the home screen becomes a browsable Asset Library — counts, auto-categories, recognized people, and an AI overview of what's inside.

This is the heart of the app. Type a query in plain language and MediaFind ranks matching moments with exact timestamps. It searches by meaning, not keywords, so “the part where she explains the pricing change” finds the right clip even if nobody said those exact words.

The same search box reaches across many channels — pick a modality or let MediaFind blend them:

  • Natural-language / transcript — semantic search over what was said.
  • Visual & scene — CLIP search over what's on screen (“a rocket on a launchpad”).
  • On-screen text (OCR) — find a slide, sign, or caption.
  • Speakers — click a voice to see every clip where they talk.
  • Color & reverse-image — find clips by a dominant color or a sample frame.
  • Brand logos PRO — by name (“Nike”) or an uploaded logo image.
MediaFind — Find
The Find workspace: a query 'a rocket launching at dawn' returns ranked moments grouped into transcript, visual and file hits, each with an exact timestamp and a play link.
A single query — “a rocket launching at dawn” — ranks matching moments across transcript, visual and file channels. Each hit carries an exact timestamp and links straight to that second.
Tip: results show a relevance ranking, not raw scores. The top match is the one to click first — it links straight to the second it happens.

5 · Ask your library

Beyond finding clips, you can ask questions of everything you've indexed. The Ask assistant retrieves the few segments that actually matter, grounds a local language model on them, and answers — with citations that link to the exact timestamp. It's conversational and multi-turn, so you can follow up.

“What did we decide about the launch date?” → a grounded answer, with sources and timestamps you can click to verify.

MediaFind — Ask
The Ask workspace answering a question about a video. The answer quotes the library and lists numbered sources, each with a clip name and a clickable timestamp, plus an 'Export cited clips' button.
Ask retrieves the segments that matter and answers with numbered, clickable citations — every claim links to the exact moment it came from. Shown here in keyless mode; a downloaded model makes answers fuller.

Like everything else, no prompt, transcript, or media leaves your machine. If you skipped the AI model during setup, Ask still works in a keyless mode and gets smarter once you download the assistant model from Settings.

6 · Jump to the exact moment

Click any result to open the player. The transcript scrolls in sync, on-screen text appears in its own panel, chapters let you skim a long recording in seconds, and the word-level timestamps mean playback lands on the exact moment — not “somewhere around there.”

MediaFind — Player
The player view: a video paused on a frame, a 'Moments' segment navigator shown as a waveform you can click to jump, and actions to export a clip, open the video editor, share, or build a highlight reel.
The player opens on the moment you clicked. The Moments navigator lets you click or arrow-key to scrub, and one-tap actions export the clip, send it to the editor, or start a highlight reel.

7 · Export to your editor NEW

Found the moments you need? Select any set of results and export them as a timeline that opens directly in your NLE:

  1. Select the results you want from a search (use the bulk selection bar).
  2. Choose Export to editor and pick a format — FCPXML (Premiere Pro / Final Cut Pro) or EDL (DaVinci Resolve and others).
  3. Open the file in your editor. In/out points are frame-accurate, snapped to each clip's own frame rate, and each clip carries the text that matched.
Selected moments from any search FCPXML EDL Premiere Pro Final Cut Pro DaVinci Resolve frame- accurate
Selected results export as an industry-standard timeline. In/out points are snapped to each clip's own frame rate, so the cut lands exactly where the match did.

DaVinci Resolve users can also send moments straight onto the current timeline via the bundled integration — no file shuffling.

Make it — MediaCreate NEW

Find locates the moments; Create assembles them. Open MediaCreate from the Find ⇄ Create toggle in the top bar, or go straight to /create. The first run shows a short coach-mark tour, and you can replay it later from Take a tour on the Projects home.

  1. Start a project. Choose Blank project, New from a search, or From a template. Recent projects sit below; click the title to rename, or use the gear for sequence fps and resolution.
  2. Use the Bin. The left rail searches your indexed library live, so you can pull footage into a cut without leaving Create. Click or drag a Bin clip onto the timeline.
  3. Move through rooms. Create is organized by job: compose, edit, enhance, deliver, version, and queue work all stay inside the same local project.
RoomWhat it does
Storyboard PRO
Auto-Reel PRO
Script PRO
The AI assembly trio: turn an intent into cited beats, rank highlights from a long source, or match narration lines to B-roll. Gaps are flagged; Create does not invent footage.
TimelineFrame-based multi-track editing: add, split, trim, move, duplicate, ripple-delete, cross-dissolve, and undo. The Roll / Slip / Slide pro trims live in the trim panel. Editing is free.
Transcript
Captions
Audio
Edit by text, remove silences or fillers, keep one speaker, style captions, burn in or export SRT/VTT, translate on-device, set gain/loudnorm/ducking/fades, and add ducked music beds.
Reframe
Text & Graphics
Color
Brand
Templates
Convert to 9:16, 1:1, or 16:9 with face-aware Auto or Center crop; add titles and lower-thirds; apply bounded looks; reuse brand kits and project templates.
DeliverRender MP4 1080p, MP4 720p, or MOV ProRes; create clip packs; repurpose variants for TikTok, Reels, Shorts, Square, YouTube, and X; export FCPXML/EDL or import FCPXML back. Render and packs are free (free degrades to 720p with a watermark); NLE interchange is PRO.
Versions
Queue
Name snapshots, diff them against the current cut, restore with a pre-restore backup, and watch cross-project render, clip-pack, repurpose, and proxy jobs. Cancel or retry jobs from the Queue.

MediaCreate stays local-first: projects reference files by path, can relink moved media by filename, and share the same on-device index as Find. No account or API key is needed for core work. Free mode degrades where it can instead of hard-walling: clip packs and repurpose jobs can render at 720p with a watermark and count cap. Pro-only actions show an in-app upgrade panel; Pro is a subscription with a 7-day free trial, and the license check works locally/offline.

8 · Browse people, brands & chapters

MediaFind auto-organizes your library so you can browse, not just search:

  • People & faces PRO — an on-device face library, checked by default for Pro/trial installs and off for free installs. Click a face to jump to every appearance; well-known public figures are auto-named with a ⭐. You can turn it off before indexing and wipe all face data with one click.
  • Speakers — a roster of every voice; click one to see all their clips.
  • Categories & knowledge map — zero-shot tags group your library by topic, with a knowledge overview of what's inside.
  • Chapters & summaries — per-file summaries and auto-segmented chapters make long recordings skimmable.
MediaFind — People & Voices
The People & Voices view: a grid of clustered faces, each with a face count and the number of files it appears in, plus buttons to refresh faces, forget all faces, and identify notable people.
The opt-in, on-device People & Voices library clusters every face it finds. Click anyone to jump to their appearances — and wipe all face data with one button whenever you want.

9 · Built-in video tools FREE

The Tools section includes a keyless ffmpeg-powered editor — cut, convert, resize, compress, extract audio, grab a frame, or make a GIF — without leaving the app or installing anything. Outputs land in your exports folder.

11 · Meetings mode PRO

Drop in any Zoom, Teams, or screen recording and MediaFind transcribes it on-device, extracts decisions and action items with speaker attribution, and makes every word searchable across your entire meeting archive.

MediaFind — standup_2024-06-14.mp4
Action Items
@alice — share Q3 roadmap draft with team by Friday · 02:14
@bob — follow up with design on the onboarding flow · 05:38
@carol — send updated pricing deck · 08:01
@alice — schedule retro for next sprint · 11:22
Key points
Team agreed to defer the API v2 launch to next quarter. Design review set for Thursday. Alice owns roadmap, Bob covers design loop.
Speakers detected
Alice (38%)
Bob (31%)
Carol (31%)
Watch the demo
2 min · on-device · no audio required
Meetings mode extracts named action items, decisions, and a summary — with exact timestamps and speaker labels — entirely on your Mac.
  • Auto action items. Every "you'll handle that" or "can you send…" surfaces as a named, timestamped task — no separate notes app needed.
  • Search across all meetings. Ask "when did we last talk about pricing?" and get the exact moment across every meeting you've ever recorded.
  • Stays on your machine. Your call recordings never leave your Mac. Transcription and extraction run fully on-device — no cloud, no upload.

10 · Privacy & your data

Everything above runs on your Mac. Your library lives in a local data folder — a SQLite index, thumbnails, and preferences — that you can open, back up, or delete yourself. The only network activity is opt-in and predictable: the one-time model downloads, license/update checks, and any public URL you explicitly ask MediaFind to fetch.

If you want to verify rather than trust, MediaFind ships a built-in privacy check that confirms the core path opens zero external sockets:

$ mediafind audit
✓ core path opened 0 external sockets.

Free to start, try Pro free for 7 days

MediaFind is free to try with no account required — enough to feel the whole workflow on a real project. Pro unlocks people & faces, brand logos, meetings mode, the media downloader, and unlimited library size, and starts with a 7-day free trial — then $24.99/month, or $107.88/year ($8.99/month equivalent). Cancel anytime; the license check works locally/offline on your Mac.

Ready to make your library searchable?

Free trial. No account, no API keys, nothing uploaded.

Download for macOS