Rescroll

AI video editing agent

An AI agent that edits your video. And checks its own work.

Give it a brief like you'd give an editor. Rescroll plans the edit, makes every cut, zoom, caption and graphic on your footage, watches the result frame by frame, and fixes what's wrong before you see it.

Give it your first brief

The short answer

What is an agentic video editor?

An agentic video editor is an AI that carries out a whole edit from a brief, instead of waiting for you to click each tool. You say what the video needs; the agent plans the steps, makes the cuts, zooms, captions and graphics on your footage, checks the result and fixes what's wrong. Rescroll's AI video editor works this way.

The difference from AI features in an editor is the loop. A caption button or an auto-cut does one job and stops. An agent reads the brief, decides the order, does the work and then looks at what it made: Rescroll checks its own rendered frames for a graphic over a face or a caption, moves it, and checks again.

You're still the director. You watch the edit and give notes, and there's a frame-accurate timeline for anything you'd rather place by hand. What changes is that the notes are sentences.

One run, start to finish

Brief. Plan. Edit. Check. Note.

One run on a 12-second take — the same one in the film above. Five stages, the way an editor works, with the agent doing each one.

  1. 1

    Brief

    “Turn this into a tight YouTube intro: cut the pauses and the false start, punch in on the key lines, warm grade, captions, and put the 40K and the three things on screen.”

    One sentence, the way you'd brief an editor.

  2. 2

    Plan

    Plan · 6 steps

    Build 6 edits
    1. 1Cut 4 pauses and 1 false start−4.9 s
    2. 2Punch in on “forty thousand” and “three things”2 zooms
    3. 3Warm grade, skin tones keptwhole clip
    4. 4Captions, word by wordtimed to speech
    5. 5Counter 0 → 40,000 on “forty thousand”, beside himon the word
    6. 6A three-point list on “three things”, clear of his handson the word

    You read it first. Change a step in plain English, or press Build.

  3. 3

    Edit

    • Cut 4 pauses and 1 false start
    • Punch in on “forty thousand” and “three things”
    • Warm grade, skin tones kept
    • Captions, word by word
    • Counter 0 → 40,000 on “forty thousand”, beside him
    • A three-point list on “three things”, clear of his hands

    0:12.0 → 0:07.1 · pauses and the false start cut, on your footage

  4. 4

    Check

    Checked 24 rendered frames

    0:00
    0:01
    0:03
    0:04
    0:04
    0:06

    0:04 — The list covers his hand

    Moved it 128 px right · re-rendered

    All 24 frames clear

    Before you see anything. You get the fixed version.

  5. 5

    Note

    “Land the 40K a beat later and make it bigger.”

    Moved to 0:01.3, 1.2× · checked

    Done · 0:07 · 6 edits · ready to export

Not just the transcript

It watches the footage. Then decides where things go.

Most AI editors only read what you say. Rescroll's agent uses the words for timing and the frames for space, so a graphic lands on the right word and somewhere it doesn't cover you.

  • Every word, with its timing

    Where you pause, where you restart a line, and the exact word a graphic or a punch-in should land on.

  • Your face and hands

    Graphics go into the space around you, not over your eyes or the thing you're holding up.

  • Text already on screen

    Captions and titles in the footage are left readable — nothing is stacked on top of them.

  • The platform's buttons

    On vertical video, graphics stay clear of the TikTok, Reels and Shorts buttons and the caption area.

A creator talking to camera in a vertical video, with the zones Rescroll's agent detects outlined: his face, his hand, the text already on screen and the platform's buttons — and a graphic placed in the clear space beside him.
Face, hands, text on screen, the buttons — and the graphic in the space they leave, on the word “three”.

What to tell it

Brief it. Then give it notes.

The first message is the brief; everything after is a note. Each of these is one message in the chat on your video.

“Turn this into a tight YouTube intro: cut the pauses, punch in on the key lines, warm grade, captions.”

A whole edit from one brief

The agent breaks the brief into steps, shows you the plan, and makes every change on your footage.

“Put the 40K and the three points on screen, clear of my hands.”

Graphics where there's room

It sees where your face, hands and existing captions are, and puts each graphic in the space that's left — on the word you say.

“Make sure nothing covers my face.”

It checks its own work

Graphics are rendered and checked frame by frame. One that covers your face, your hands or a caption is moved and checked again before you see it.

“Land the 40K a beat later and make it bigger.”

Notes, not redos

One agent works on your video from the first brief to the last note, so a note changes the thing it's about.

“Use our real pricing page — here's the screenshot.”

Your real product, not a template

Attach a screenshot, a saved web page, a logo or your numbers, and the graphic is designed from them.

See AI motion graphics

“Show me the plan before you build the graphics.”

A plan you can change first

Before a graphic is built you read what it will be and when it lands. Change a step in plain English, then let it run.

How it works

How to edit a video with an AI agent.

From raw footage to a finished MP4, with the agent doing the steps and you doing the reviewing.

  1. 01

    Drop in your footage

    Straight from your phone or camera — vertical, horizontal or square. Rescroll transcribes it word by word and looks at the frames.

  2. 02

    Give it a brief

    Write what the video needs the way you'd brief an editor: what to cut, where to punch in, the look, what goes on screen and when. Attach a screenshot, page or logo for anything real.

  3. 03

    Read the plan, let it work, give notes

    It shows the steps, makes the edit on your footage, checks its own frames and fixes what's wrong. Ask for changes in the same chat, then export a finished MP4 with your original audio.

Three ways to edit

Your hands, a button, or an agent.

What each one does for you on a talking-head video — and where the other two are still the better pick.

Editing it yourselfAI features in an editorRescroll
Works from a written briefNoOne feature at a timeYes
Plans the edit before making itYou doNoYes
Knows where your face, hands and captions areYou doPartlyYes
Checks its own result and fixes problemsYou doNoYes
Remembers the edit for the next noteYou doNoYes
Frame-accurate manual controlYesYesYes
Works on your phoneYesMost doNo
Adds stock B-roll and musicYesMany doNo

Editing by hand gives you every decision, and the AI features in your editor save time on single jobs. Rescroll's agent is for describing the whole edit and reviewing it — on a computer, on your own footage, without stock clips or music.

Questions

AI video editing agents, answered.

What does agentic video editing mean?

It means an AI agent does the edit, not just one step of it. You give it the goal in plain English; it plans the steps, uses the editing tools itself, looks at the result and corrects it, and keeps the whole edit in mind for your next note. A caption generator or an auto-cut button is a feature; an agent strings them together toward what you asked for.

Is there a Claude Code or Cursor for video editing?

That's the idea behind an agentic editor: you describe the outcome, the agent does the steps, and you review the result. Rescroll's AI video editor works that way on your own footage — a brief, a plan, the edit, a frame check, then notes in plain English — in the browser, with no code.

Can an AI agent edit a video like a human editor?

For the repetitive part, yes: cutting pauses and bad takes, captions, punch-ins on the key lines, a grade and graphics on the right words. The taste calls are still yours — which pause makes the joke land, which moment deserves a graphic — and with an agent you make them as notes instead of clicks.

Does the AI check its own work?

Yes. Rescroll renders each graphic and checks the frames before you see them. If something covers your face, your hands or a caption, it's moved and checked again. You still have the final say: watch the edit and ask for any change.

Can the AI see my video, or only the transcript?

Both. Rescroll uses the transcript for timing — where you pause, where you restart a line, the exact word a graphic should land on — and looks at the frames for where your face, hands and existing captions are, so graphics go into clear space. On vertical video it also keeps them clear of the TikTok, Reels and Shorts buttons.

How is this different from Descript Underlord or OpusClip's AI Producer?

All three take instructions in a chat. Underlord runs Descript's tools inside a transcript-based editor, and AI Producer auto-edits a talking-head video with captions, B-roll and music. Rescroll's agent is built around custom motion graphics from your own screenshots, pages and data, with a plan before each graphic and a frame check after.

Will a note change things I didn't ask for?

It shouldn't. One agent works on your video from the first brief to the last note and knows what it already made, so “make the captions bigger” is about the captions. If something else moved, say so in the chat and it's put back.

Can I still edit by hand?

Yes. There's a frame-accurate timeline for anything you'd rather place yourself, and the agent handles everything you'd rather describe. Most people describe the edit, then nudge one or two moments by hand.

What can't the agentic video editor do?

It doesn't add stock B-roll or music, generate new footage of you, or clip a long video into shorts. It edits the footage you give it — vertical, horizontal or square — in Chrome or Edge on a computer, and exports a finished MP4 with your original audio.

How much does the agentic video editor cost?

It's part of Rescroll's Pro plan ($20 a month billed yearly, or $29 monthly) and the Ultimate and Scale plans. Edits use your plan's monthly credits.

Stop scrolling.
Start posting.

Decode every Reel. Generate every visual. Post in any style you feed it. Everything you saved is one click away.

Start Your Library