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

AI Fill

AI Fill generates a bridging video clip between two moments of your timeline with an AI video model. Use it to smooth a hard cut, connect two scenes, or invent a transition no preset covers. Every time you press Generate you get a new take, so you can try several models or lengths and keep the one you like.

Insert a fill zone

Select the video channel the bridge belongs on, park the playhead where it should go, then open Smart AI in the timeline toolbar and choose AI Fill (“Insert an AI Fill zone at the playhead”). You can also type AI Fill in Search Everything (⌘K). A fill zone — an empty AI Fill gap, 3 seconds long by default — opens at the playhead:

  • A clip under the playhead is split in two, and the later clips on that channel move right to make room.
  • Only the selected channel changes. Clips on other channels stay where they are, so check that layers above or below still line up with the new gap.
  • On a channel whose clips play back to back, a playhead past the channel’s last clip puts the zone at the channel’s end.

AI Fill needs a video channel: on a voice, audio or avatar channel the menu item is disabled with the hint “AI Fill needs a video layer — select one first”.

You can open several zones, on the same channel or on different ones, and each keeps its own frames, model, prompt and takes. Choosing AI Fill while the playhead sits inside or on the edge of an existing zone selects that zone instead of opening a second one. The Smart AI button stays highlighted while any zone is open. Select a zone on the timeline to open its AI Fill panel on the right.

The panel’s header shows the zone’s time range and length, and how many takes are generating. Below it are three sections — Frames, New take and Takes — and at the bottom the Apply button and Remove zone.

Set the start and end frames

In Frames, click Extract from timeline: the editor snapshots the composited frames just before and just after the gap — everything visible on screen, not only the selected channel — as the Start and End frames. The pictures are saved in the board’s extracted media folder. You can also pick either frame yourself: click the Start or End slot to choose an image from your board media or upload one.

Save on board (the icon beside Extract) puts both pictures on the canvas, in a frame named AI Fill from Video Editor, each with a “Start frame” or “End frame” note.

Generate a take

In New take:

  • Choose a Model. The list shows the video models that take a start and an end frame first, then the Omni and References to Video models. It shows “No first/last-frame, Omni or reference models available” until you add a key for a provider that has one.
  • Pick a Length for the take. The buttons (or, for a model with many lengths, a menu) list the lengths the selected model supports. For a model that sets its own length, the panel says “This model makes its own length; the gap fits the take it returns.”
  • With an Omni or References to Video model, the Start frame is @image1 and the End frame @image2. Reference images adds more pictures, tagged @image3 onward, up to what the model takes; mention them in the prompt by their tag. Pictures past the model’s limit are dimmed and skipped.
  • Adjust the Prompt if you like. It starts as “Generate a smooth, natural transition between the two frames.” With an Omni or References to Video model it starts as an instruction to begin on @image1 and end on @image2. A prompt you have written yourself stays when you switch models.
  • Click Generate take N.

Generate never waits. Every press starts a new take with the model, length, references and prompt set at that moment, running alongside any take that is already generating. You can keep editing, or generate in other zones, while they run. The request asks for your project’s frame size. A notice tells you when each take finishes or fails, even while you work on another zone.

Pick a take

Takes lists every take, newest first, with its number, length, model and state: Generating (with the time so far), Ready, On the timeline, or Failed with the reason. A take that finished while another one was in use carries a small new mark. On the timeline, the zone shows how many takes it has and which one is in use, with a spinner while any take is generating.

  • The first take to finish goes straight into the gap. Later ones wait in the list.
  • Use (or a click on a ready take) puts that take in the gap.
  • The gap follows the take in use. Using a 2 s take makes the gap 2 s; switching to a 5 s take makes it 5 s, and everything after the gap on that channel moves with it. The gap takes the length of the video the model actually returned. With a take in use, Length sets only the next take, and the panel says so.
  • A failed take has a Try again button, which runs the same model, length and prompt again.
  • × discards a take. On a running take it stops waiting for it; the provider still finishes and bills it. Discarding the take in use switches to the newest other finished take.

Play the gap to check the take in use. Then click Apply take N at the bottom of the panel (or ✓ Apply on the zone in the timeline) to keep it. An applied take becomes an ordinary video clip named AI Fill on that channel: trim, move, add effects or delete it like any other clip.

Remove a zone

Remove zone takes the gap out again and moves the channel’s later clips back. A clip that was split to open the zone stays in two pieces, back to back. Deleting the zone on the timeline does the same.

Saving

Takes are saved with the project. When you reopen it, each zone is still there with its frames, model, prompt and finished takes; the takes come back listed but not in use, and Use puts one back in the gap. If the project reloads while it is open — an agent edits it, or you load another version of it — takes that are still generating keep going and land when they finish, and the take in use goes back into the gap.

Where to go next

Was this helpful?

0

Updated

Comments

Leave a Reply