share a markdown folder with your talent. they read it on their own phone, at their own pace, and the take comes back to you. no accounts, no upload step on either side.
every brand brief has the same tension. you need the talent to hit the message. the talent needs it to sound like them, not a read script. push too hard on the words and the delivery goes flat. loosen up and the disclaimer goes missing.
the usual fix is a worse version of the problem. you email a Google Doc, the talent copies it into some teleprompter app, the app wants an account, the doc gets pasted wrong, and the take comes back as a file named IMG_4471 with no idea which script it goes to.
there's a cleaner loop. it runs on a folder.
the setup is a shared folder#
open prompter reads markdown files straight from a folder. iCloud Drive, Dropbox, an Obsidian vault, any folder iOS can reach. no paste, no import.
so the whole collaboration fits in one shared folder.
you write the scripts as .md files. one file per video. you drop them in a folder you share with the talent. that's the agency side done.
the talent points open prompter at that folder once. every script you write now shows up in their library. the app flags the last edit, so they can tell what's new.
no account on your side. no account on theirs. the app makes no network calls of its own. the folder is shared by whatever service you already use, and that service handles delivery. open prompter just reads what lands there.
delivery that sounds like them, not a read script#
here's the part that fixes the staleness.
most teleprompters scroll at a fixed speed. the talent has to chase the line. speed it up and they rush. slow it down and they drag and start sounding robotic. either way the words win and the delivery loses.
open prompter has voice tracking. on-device speech recognition listens to the delivery and advances the page when the talent actually says the words. they talk at their own pace. the script keeps up with them, not the other way around.
so the talent can pause for a beat, lean on a word, ad-lib a half sentence, and the prompter waits. it scrolls because they spoke, not because a timer fired. that's the difference between hitting the message and reading the message.
it's opt-in and labeled BETA, and it runs fully on-device. the audio never leaves the phone. for a guide to the two handles that tune the feel, see how to use the READ and FEATHER lines.
revisions are a file save#
legal flagged a line. the brand wants the CTA reworded. the hook isn't landing.
you don't re-export anything. you don't send a new link. you open the .md, fix the line, and save.
the file syncs to the talent's folder, and the next time they open the script it's the new version. the app shows the last edit so they know it changed.
be honest about timing. sync latency is whatever your service delivers, usually ten to sixty seconds on iCloud. open prompter doesn't fake instant. but by the time the talent has set up the shot, the edit is there.
no version-control spreadsheet. no final_v3_REALLY_final.docx. one file, edited in place, current on every device that points at the folder.
the take comes back to you#
this is the loop closing.
open prompter has a setting called "save next to the script." flip it on, and after the talent records, the finished take is copied into the same folder as the script they read.
which is your shared folder. which means the video syncs straight back to you.
the talent reads the script you wrote, records, and hits stop. the take lands in the shared folder, named after the script. it shows up on your machine automatically. no upload, no WeTransfer link, no email with a file attached.
the take is named {script-stem}__{timestamp}.mov. so a script called spring-launch.md gives you something like spring-launch__2026-06-17-1432.mov, right next to the script that produced it. shoot three takes and you get three timestamped files in a row, all named after the script. picking the keeper is reading a list, not scrubbing a camera roll.
for the full mechanics of how this works, see recordings that save back to your script folder.
what the day looks like#
picture a small team and one creator.
monday, you write four scripts in markdown and drop them in the shared folder. you invite the creator. setup over.
the creator opens open prompter, sees the four new scripts in their library, and shoots them over the week. each take saves back into the shared folder, named after its script.
friday, your editor opens the same folder. the four scripts are there. so are the takes, each one matched to its script by name. the editor pulls the keepers into a timeline and starts cutting.
nobody exported anything. nobody renamed a clip so the editor could tell them apart. the brief went out through the folder and the footage came back through the folder.
write so it reads spoken, not written#
the workflow gets the talent on-script. the writing decides whether they sound human doing it.
scripts written to be read silently fall apart out loud. long sentences run out of breath. formal phrasing sounds stiff. a clause that scans fine on the page trips the tongue.
a few habits fix most of it.
- short lines. one idea per line. if the talent has to hold a thought across three commas, they'll lose it. break it.
- contractions, always. write "you'll" and "it's" and "we're." nobody says "you will love this" out loud. they say "you'll love this."
- the talent's own phrasing. if the creator says "wild" and "honestly" and never says "furthermore," write it that way. give them words they'd use anyway. let them swap a word or two without it breaking the message.
- read it out loud before you send it. if you stumble reading your own draft, the talent will too. the spot where you trip is the line to rewrite.
markdown helps here. headings break the script into beats. bold marks the line that has to land. bullets hold a list of points without forcing a run-on. open prompter renders all of it and strips the scaffolding at display time, so your [B-roll: storefront] cue never shows up on the talent's screen.
write it to be said, not read. the talent will sound like themselves, and the message will still be in there.
try the loop#
make a folder. write one script as a .md file. share the folder with your talent.
have them point open prompter at it, read with voice tracking on, and flip on "save next to the script." the take lands back in your folder a minute later, named after the script, with nothing left to send.