Blur Sensitive Info with the AI Assistant
Quick answer: Open the Assistant tab and type what you want hidden, for example “blur the access token” or “blur every email”. Zella reads the text on your video’s frames on your device, finds the secret (API keys, access and bearer tokens, passwords, emails, card numbers, or any phrase you name) and lays a blur (or a pixelate mosaic) over it for its whole time on screen, following it if the page scrolls. Everything it adds is one undo step. It works the same on Mac and iPhone, it is free, and your video is never uploaded. If it misses something, place a blur yourself with a blur callout.
On this page: what it finds · how to use it · example prompts · how the blur behaves · what stays private · limits · blur by hand · on iPhone · FAQ
For blur versus pixelate versus a solid box, preventing leaks before you record and checking every frame before you export, read how to blur sensitive info in a screen recording.
What does it find?
Screen recordings leak things: an API key in a terminal, a token in a settings page, an email in an inbox. The AI Assistant can now read the screen and hide them for you. Out of the box it looks for:
- API keys - the shapes services print, such as keys starting
sk-, GitHub tokens startingghp_, AWS key ids startingAKIA, and Slack tokens. - Access and bearer tokens, including JWTs (the long
eyJ...strings web apps use). - Passwords and other labelled secrets - any value written after a label like
token:,password:,secret:orapi key:. - Long key-like strings - long runs of hex or random-looking characters.
- Email addresses.
- Card numbers.
- Anything you name - a customer’s name, a project code, a phone number. Say the words and it looks for them too.
How to blur a secret by typing
- Open your project and go to the Assistant tab. (First time? Add a free AI key, see setting it up.)
- Type what you want hidden: “blur the access token”.
- Zella scans the video and adds the blurs. On the Mac it also jumps the playhead to the first one so you can check it.
- Read the reply. It lists each blur it added, what kind of secret it was (masked, never the secret itself), and the time range it covers.
- Not happy? Press ⌘Z on the Mac (or ask “undo that”) and every blur from that request goes at once.
By default it scans the whole video, sampling about every half second. Long videos are sampled more coarsely, and if a very long scan runs out of time the reply says exactly where it stopped, so you can ask it to carry on from there.
Example prompts
| You type | What happens |
|---|---|
| ”Blur the access token” | Finds the token and blurs it everywhere it appears |
| ”Blur every email” | Blurs each email address on screen |
| ”Pixelate the API key” | Same search, with a blocky mosaic instead of a soft blur |
| ”Blur the token and the label too” | Also covers the label next to it, such as Access token: |
| ”Blur the API key from 10 s to 40 s” | Only scans and blurs inside that range |
| ”Hide the customer name Jane Doe” | Looks for your exact words instead of a secret shape |
| ”What text is on screen at 12 seconds?” | Reads the frame and tells you, with secrets masked |
You do not need these exact words. “Censor the password”, “redact my email” and “hide the sensitive info” all work.
How the blur behaves
- One blur per secret, for its whole time on screen. A key that sits in view for 20 seconds gets one blur that starts when it appears and ends when it leaves, not a flicker of short ones.
- It follows scrolling text. If the page scrolls while the secret is visible, the blur moves with it.
- It never animates in or out. Callouts can normally pop or fade, but a fade would let the text show through for a few frames, so redaction blurs appear and disappear instantly.
- They are ordinary blur callouts. Each one lands in the editor like a blur you placed yourself, so you can move, resize, retime or delete it afterwards (see Callouts).
- One undo removes the whole batch. Everything one request added is a single step.
- Cut parts are skipped. On the Mac, stretches you have already deleted from the timeline are not scanned.
- It is baked into the export. What you see blurred in the preview is blurred in the exported file.
What stays on your device
The text reading is on-device OCR (optical character recognition: the app reads letters off the picture). Your video and its frames never leave your Mac or iPhone.
What goes to the AI provider you chose is the same as for any Assistant request: your typed message and a short text summary of what happened. The report the Assistant reads back is masked: it says “API key” or “card number ending 42”, never the secret itself. The Assistant is free on every tier, with your own key.
What it can miss
OCR reads crisp, steady text best. The Assistant may miss text that is:
- very small (tiny print in a full-screen 4K capture),
- blurry or out of focus,
- moving fast (a quick scroll or a swipe),
- not in a Latin script (for example Arabic, Bangla or Chinese text).
Always scrub through the result before you publish. When it misses something, the reply says so, and you can place a blur yourself.
How to blur something by hand
Two ways, both free:
Ask the Assistant with a place and a time. Callouts can now be placed precisely by chat: “add a blur box in the top right corner from 3 s to 8 s” adds a blur at that spot for exactly that window. The Assistant can also read the screen first (“where is the API key at 12 seconds?”) and then place the box over it.
Add a Blur or Pixelate callout yourself. On the Mac, open the Arrows tab, add a Blur or Pixelate callout, set its start and duration, and drag it over the text. On iPhone, the same callouts are in Studio → Arrows & Labels. Full steps: Callouts.
On iPhone
It works the same way on the phone: open the Assistant tab in the editor, type “blur the access token”, and the blurs land on your timeline in one undo step. The detection, the half-second sampling, the scroll-following blur, the pixelate option and the masked report are all shared with the Mac, and the frames are read on the phone itself. For a hand-placed blur, use the blur or pixelate-censor callout in Studio → Arrows & Labels (see captions, titles & callouts on iPhone).
Blur FAQ
How do I blur an API key in a screen recording? Open the Assistant tab and type “blur the API key”. Zella finds it on screen and blurs it for as long as it is visible.
Does it upload my video to find the secrets? No. The frames are read on your device. Only your typed message and a masked text summary go to your AI provider.
Can I pixelate instead of blur? Yes. Say “pixelate the token” and it uses a blocky mosaic instead of a soft blur.
It blurred the value but not the label. Can it cover both?
Yes. Ask it to “blur the label too” and it includes text like Password: next to the value.
It missed one. What now? Add a Blur callout over it by hand, or tell the Assistant where and when: “blur a box at the bottom left from 5 s to 9 s”.
Is this a Pro feature? No. The Assistant, the callouts and this tool are free on every tier, on Mac and iPhone.
Pro tips & gotchas
- Name what you see. If the screen says “Access token”, say “blur the access token”. Your own words help it match the right line.
- Scrub before you export. Small, blurry, fast-moving and non-Latin text can slip through.
- Narrow the range on long videos. “From 2:00 to 3:30” scans that stretch more densely and finishes faster.
- One undo clears the batch, so it is safe to try a broad request like “blur the sensitive info” first.
Related: AI Assistant → · Callouts → · Effects (blur, masks) → · Workflow recipes → · Captions, titles & callouts on iPhone →