The Video Compositor makes short AI video clips. Most models turn a still image into motion: you give a clip a start frame, pick a model and a camera move, and INTERSTITIAL animates it. One model, Kling O1 Edit, works the other way round — it takes a video you already have and rewrites it from a prompt. Chain several clips together and assemble them into one continuous video.

When to use

You have a rendered view and want it to move — a slow dolly through a space, a reveal, or a walkthrough.
You want a short sequence built from several images that flow into each other.
You already have footage and want to change something in it — the season, the sky, a material, an object. That is Kling O1 Edit.

You can drop a video file straight onto the canvas (MP4, MOV, WebM or M4V, up to 200 MB). It becomes a layer like any image: drag it, resize it, group it. It plays when you select it, it has a dock point on its right edge so you can wire it into a compositor, and its toolbar can extract the frame you are looking at as a new image layer.

Add a Video Compositor

On the canvas, open the Add Element (+) menu.
Choose Video Compositor (Multi-clip video assembly).
Click the canvas to place the widget. It opens with one empty clip ready to fill.

Make a clip

Each clip is a column inside the widget. Fill it out top to bottom:

Set the start frame. Click the Start slot, then Upload Image or Pick from Canvas to use an image already on your board. Frame slots accept images only.
Pick a model. Use the model picker in the widget header — Kling, Seedance, Wan, Runway, Veo or Gemini. Each model has its own durations, resolutions, and aspect ratios.

Edit a video you already have (Kling O1 Edit)

Pick Kling O1 Edit in the model picker and the clip changes shape: instead of Start and End frame slots it shows a single source clip slot. The ⓘ button beside the picker explains the whole flow in the app.

Attach a clip. Upload a video, or drag a connection from a video already on your canvas into the source slot.
Trim it. Drag the handles under the tile. The edit is always exactly as long as the clip you send, so this is what decides both what gets edited and what it costs. The window can be 3 to 10 seconds; a longer clip opens with the first 10 seconds selected, and a clip shorter than 3 seconds cannot be used at all.
Attach up to 4 reference images (optional) and point at them in your prompt with @. Type @ in the prompt and pick from the list — the first slot is @image1, the second @image2, and the source clip is @video1. Writing “the reference image” instead of a token does nothing: the token is the only way the model knows which picture you mean, and an unmentioned reference is silently ignored. The clip warns you when that happens, and auto-prompt writes the tokens for you.
Describe the change, not the clip. “Change the season to winter”, not “a building in a park”.

What it does well: adding, removing or swapping objects; changing a background, season, weather or time of day; relighting; restyling; changing materials. It struggles with rearranging things in space, fine facial detail, and anything physics-dependent.

It will not move the camera — it preserves the source’s motion, camera path and timing exactly. That is the point of it, and it is why this model offers no camera presets, no duration slider and no aspect-ratio picker: all three come from your footage.

Extract a frame from a video

Select a video layer on the canvas, scrub to the moment you want, and click Extract this frame as a layer in its toolbar. The frame lands beside the clip as an ordinary image layer — ready to use as a start frame, a reference, or anything else an image can be. It is free. 3. Write a prompt, or press Auto to have the AI draft a prompt and camera move from your frame. The refresh button regenerates it. 4. Choose a camera move from the Camera dropdown (presets are listed below). 5. Set the details in the controls row — duration, quality (720p / 1080p on Kling), an audio toggle, and a seed on the models that support one. 6. Press the round generate (arrow) button. The status dot turns amber while it runs and emerald when the clip is done.

An End frame slot appears once a start frame is set. It is optional — add one to steer where the motion finishes.

Camera moves

The Camera dropdown groups the shared presets into four categories:

Essentials — Static / Tripod, Dolly Forward, Dolly Backward, Walkthrough POV.
Pan & Tilt — Pan Left, Pan Right, Tilt Up, Tilt Down.
Specialty — Crane Up / Reveal, Orbital / Turntable, Aerial Drone, Tracking Shot, Threshold Crossing.
Zoom — Zoom In (Detail), Zoom Out (Establish).

Pick None to leave the camera to the model and your prompt.

Chain clips into a sequence

When a clip finishes, its last frame auto-links to the next clip’s start frame so motion carries across the cut. A banner marks a linked frame, and you can still replace it.
Use the floating + button to add a clip from the last frame and keep the chain going.
Once two or more clips are done, an Assemble button joins them into one video. Use Preview all to play the clips back to back first.
The aspect ratio is auto-detected from your first start frame; you can override it per widget.

Tips

Start with a clean, high-quality frame — the clip inherits its look from the start image.
Short prompts about motion work best; describe the camera and what moves, not the whole scene.
Costs vary by model, duration, and quality — check the per-clip figure shown next to the generate button and the in-app pricing screen.
On the Blueprint tier, video is limited to Veo; other video models prompt an upgrade. See Plans & Credits.

Generation & Models — how the models and credits work across image and video.
Editor & Canvas Objects — placing, selecting, and arranging widgets on the canvas.
Plans & Credits — tiers, costs, and Blueprint limits.