Black Friday33% off all plans See plans Enter it at checkout

AI Viral Edits

AI Viral Edits turns a talking clip into a styled viral edit: big kinetic words standing in 3D space around the person on screen, some behind them and some in front, with a camera that moves through the words the way your footage moves. You pick a type and, if you like, the words. An agent builds the edit on your timeline, and you then change it from one panel.

AI Viral Edits is a PRO feature. Without a PRO licence the licence window opens instead.

Open AI Viral Edits

  1. Select a video layer on the main timeline. The layer’s video has to be on this board; if it is not, the editor asks you to import it first. Inside a sequence, the editor asks you to leave the sequence first.
  2. In the timeline toolbar, choose Smart AI ▸ AI Viral Edits… (marked PRO). You can also type AI Viral Edits in Search Everything (⌘K).

The window is headed AI Viral Edits · words in Z space around your subject. It has four steps along the top: Type, Text & motion, Extra references and Run. Click a step to jump to it, or use Next and Back. Esc or × closes the window.

Viral Edits is also listed in AI Workflows, marked “in the Video Editor, on a layer”. It opens only from a video layer, so that entry points you here. Back to List in the Viral Edits window’s header opens the AI Workflows window.

Step 1 — Type

Each type is a card with a short looping preview. Every type puts words behind the subject and in front of them, moving with the camera.

TypeWhat the words do
Arc with ZCurve around the subject on an arc that runs behind and in front of them.
Right stackBig lines stack beside the subject; the camera pushes through them.
Circle stackA half circle beside the subject; every new line stacks toward its centre.
RingA full ring of words around the subject, in depth.
Chaotic stackWords pile up at loose angles and depths around the subject.
Brutal stackHuge accent words behind the head, smaller words in front, each at its own depth.
Object pathLetters ride the subject’s outline, over the head and shoulders, frame by frame.

The window starts on the type of your last run.

Step 2 — Text & motion

Text is optional.

  • Put stars around a word to show it in the accent colour: *you've* been asking.
  • Put a / where a new line starts.
  • Leave the box empty and the agent writes the words from what is said in the clip.

The rows under it set how the words arrive:

RowWhat it sets
EntranceHow each word comes in. The picker plays every choice as you look at it, so you see the move before you pick it. The first choice is the type’s default. The entrances include Rise, Pop, Focus Pull, Flip Up, Slam and a set of Advanced Animations presets.
Accent wordsThe entrance for the starred words: their own, or Same as words.
StartSeconds after the layer begins. Left empty, the agent times the first word to the speech.
Word gapSeconds between words. The hint shows the type’s default.
Entrance speed1 plays the entrance as designed; higher is faster.
SideCircle stack only: Left or Right, where the half circle sits.
LayoutChaotic stack only: a number for the pile; Shuffle picks another one.
LettersObject path only: Behind the subject or In front.

Anything you leave empty stays the type’s default.

Step 3 — Extra references

This step is optional. Pick pictures or clips from the board’s media for the agent to look at, such as a reference edit, a logo or brand colours. Upload adds files from your computer. Notes for the agent takes anything else it should know, for example which word to accent or which colours to use.

Step 4 — Run

The Run step sums up what will be made: the type, the text (or “The agent writes it from the clip’s speech”), the entrance and how many references you picked.

Requirements

Every type needs a mask of the subject, so the words can stand behind them. The Requirements list checks this before anything runs:

RowWhat it means
Subject maskWhere the mask comes from. A SAM cut-out of this clip already on the board, or the layer’s own SAM matte mask, is reused at no cost. Otherwise the agent makes one on this computer with SAM 2, free, and nothing is uploaded. If that fails and you have a WaveSpeed key, SAM 3 on WaveSpeed makes it instead, and the row shows what that would cost for this clip (about $0.01 per second, 3 seconds at least). Type what to cut out in Subject (the default is “person”).
Subject outlineObject path only. It is measured from the mask in every frame.
Camera movementThe app measures it from the clip. There is no tracking to set up.
Speed curve on this layerA warning when the layer has a speed curve: the words follow the clip’s own time. Remove the curve for an exact match.

SAM 2 needs a real graphics card. On a computer without one, the mask comes from SAM 3, which needs a WaveSpeed key. If there is no key either, the row says so and offers Get a WaveSpeed key ↗, Add it in Settings and Check again. See RotoScope and Magic Mask for both models.

Run with

  • GLEA runs the edit in the editor’s own agent panel. Pick the agent: Claude Code, Codex, Antigravity, OpenCode, Cursor or xAI. Click Start the edit. The project is saved first, and the new layers appear as the agent builds them. GLEA needs the desktop app or the browser terminal of GreenLight Dash Server.
  • Outside the app is for your own terminal, IDE or agent. Copy for your agent saves the brief to the board’s ViralEdits folder and copies a prompt with everything the agent needs to connect and start. Your agent needs an agent token from your account page on greenlightdash.pro. Show the prompt shows the text before you copy it.

The window keeps your choice of agent and where it runs for the next run.

What you get

The agent builds the edit in 3D above your clip, on the clip’s time range:

  • One sequence on a new channel right above the clip’s channel. Inside it, every word is its own text layer, standing still at its own depth behind or in front of a cut-out of the subject.
  • Its own camera. A camera inside the sequence moves through the words the way your footage’s camera moved, with depth of field, so each word sharpens as the camera reaches it. Your project camera stays free for your own moves.
  • Your footage stays as it was, below the sequence.

Move, trim or prerender the sequence as one block. Open it to double-click and change any single word.

Change the edit: the Essential panel

Select the edit’s sequence, or a word inside it, and the ESSENTIAL section appears at the top of the right panel. It holds what the edit was built from:

ControlWhat it changes
TypeRebuilds the edit as another type.
TextThe words, with stars and / as in the window.
Entrance / Accent wordsThe entrances, with the same playing pickers.
Words colour / Accent colourThe two colours.
Start, Word gap, Entrance speedThe timing.
Camera followHow closely the edit’s camera follows the clip’s camera. 1 follows it exactly.
Depth blurThe camera’s depth of field, focused where the words land. 0 turns it off.
Blur levelHow strongly that depth of field blurs, 30 % by default. It changes the picture at once, without a rebuild, and later rebuilds keep it.
Side / LayoutFor Circle stack and Chaotic stack, as in the window.

A change to any control except Blur level rebuilds the whole edit in place, as one undo step; the panel shows Rebuilding the edit… while it works. A rebuild replaces any changes you made by hand to the words and the camera keys. A green dot marks each control that differs from the type’s default; click it to go back.

Where to go next

Was this helpful?

0

Updated

Comments

Leave a Reply