Captions, silence, and shorts

The workflows below are the everyday reason to reach for the AI. Each one is an "ask the timeline" flow: you describe intent, the agent proposes a typed patch, and you review a before/after diff before applying. Everything here is reversible — apply, preview, and undo in one click each. For the modes behind these flows, see the AI agent.

Auto captions

FramePilot transcribes your audio and generates captions with word-level timestamps, so captions land in sync with speech rather than in loose blocks.

• Generate a caption track from the transcript by asking the timeline to add captions.
• Style captions with templates, and highlight keywords for emphasis.
• Edit any caption manually — the text and timing are yours to adjust.
• Burn in captions on export, or keep them as an editable track until you are ready.

Because the caption track is added as a patch, you can undo the whole thing or tweak individual lines without redoing the transcription.

Remove silence

Dead air is the fastest thing to cut. Ask the timeline to remove silences and the agent analyzes the audio for silent gaps, then proposes a patch that deletes them.

Remove the silences.

You review exactly which ranges will be removed as a diff before applying, and undo restores the original timing if a cut is too aggressive. This is a reversible timeline operation, not a destructive re-encode of your source media.

Fix pacing

Slow sections lose viewers. Select a range and ask for tighter pacing, and the AI proposes targeted operations — trimming slack, adding a speed ramp, or a punch-in zoom on a key moment.

Make this section faster and more engaging.

The result comes back as a small patch you can apply, edit, preview, or reject, so you can dial in the feel without committing blind.

Find the hook

The opening decides whether anyone keeps watching. The agent can analyze your transcript to detect the strongest hook moment and propose restructuring the opening around it — for example, leading with 00:08–00:13 and moving the intro later or cutting it.

As with every flow, the change is a proposed patch. You see what moves before it happens and can undo it if the original order worked better.

Turn a recording into a 9:16 short

The flagship flow ties the others together. Hand the agent a raw recording and a goal:

Create a 45-second product demo for Instagram Reels.

In agent mode it will typically:

1. Analyze the transcript.
2. Detect the hook and the strongest segments.
3. Detect slow or repeated sections to cut.
4. Draft an edit plan and ask for your approval.
5. Apply the timeline operations (cuts, captions, zooms, reframing to vertical).
6. Render a preview and self-check the output.
7. Suggest improvements before you export.

Every step is a reviewable, reversible operation, and the final render goes through the deterministic, auto-validated export path.

Next steps

• The AI agent — chat, plan, edit, and agent modes in depth.
• Render and export — export presets and how output is validated, including 9:16 Reels.