How I Edited a 49-Second Reel by Chatting With AI (Claude + Palmier Pro)

I just edited a complete Instagram Reel — 11 raw clips down to a tight 49-second video with text overlays — without touching a timeline. Instead, I described what I wanted in a chat, and Claude did the editing inside Palmier Pro through MCP (Model Context Protocol).

Here's exactly how the workflow went, including the parts that didn't go smoothly.

The Setup

Claude (Anthropic's AI assistant) as the "editor"

Palmier Pro, a Mac video editor with an MCP server — this is what lets Claude actually control the app: import media, cut clips, add text, export

Raw footage: 11 iPhone clips (~15 minutes total) of a 3D printing project — slicer screen recording, print process, removal, assembly, and the final product playing a song

MCP is the key piece. It's an open protocol that gives AI assistants real tools instead of just chat. Claude wasn't describing edits for me to perform — it was performing them.

Step 1: Project Setup by Conversation

I told Claude I had a project called "First Project" in Palmier Pro. It opened the project, saw it was 1920x1080, and suggested switching to 9:16 vertical for Reels before any clips landed on the timeline. One "ok set 9:16" later, the canvas was 1080x1920 at 30fps.

Small thing, but this is where an AI editor already helps: it caught a setup mistake before it became a re-editing problem.

Step 2: Import and Automatic Footage Analysis

I pointed Claude at a folder of footage. It imported everything, then — and this is the impressive part — inspected each clip visually. For every video it sampled frames or generated a storyboard overview and reported back:

Which clip was the Bambu Studio slicer screen recording

Which clips were static print-process shots (and therefore needed heavy trimming)

Which clip had the most visual variety (the best montage material)

Where the action happened inside long clips — e.g. "removal from the plate happens around seconds 4–25"

Instead of me scrubbing through 15 minutes of footage, I got a summary and a proposed story structure in a couple of minutes.

Step 3: Timeline Assembly

We agreed on a classic 3D-printing content arc: slicer → print montage → removal → assembly → payoff (the printed fidget playing Happy Birthday). I told Claude the one thing it couldn't see — where in the final clip the music actually plays — and it placed all 8 cuts in a single operation:

3s hook from the slicer recording

Quick 3-second print montage cuts from four different clips

8s of removing the print from the plate

6s of assembly

20s payoff with the melody, including a 2-second lead-in

Total: 49 seconds, assembled in one shot, with all original audio preserved (printer ASMR + the song).

Step 4: Text Overlays and Visual QA

I asked for text overlays. Claude added four phase-matched titles ("3D Printed Musical Fidget 🎵", print stats, assembly, and a golden payoff title), then rendered preview frames of the actual composited timeline to check its own work.

When I pointed out the text was hard to read, it added black background boxes. When I sent a screenshot showing the box didn't cover the second line of text, it resized the text box and verified the fix with another rendered frame. This iterate-look-fix loop felt genuinely like working with a junior editor — one that responds in seconds.

Step 5: Export

One message: export. H.264, 1080x1920, 49 seconds, rendered in the background straight to my Downloads folder, ready to upload.

What Didn't Go Smoothly (Honest Notes)

MCP server hangs. The connection froze twice — once during clip inspection, once during audio beat detection. A restart of the app fixed it both times, and no work was lost since the project auto-saves. Lesson: heavy analysis operations are the risky ones.

The AI can't hear. Claude analysed footage visually but couldn't tell me where the music played in the final clip. I had to supply the timestamps. Know what your tools can and can't perceive.

Language limits. On-device transcription didn't support Indonesian — irrelevant here since the footage had no narration, but worth knowing for voiceover content.

Is This the Future of Editing?

For this kind of content — structured, formulaic short-form video — honestly, yes. The AI handled 90% of the mechanical work: importing, analysing, trimming, arranging, titling, exporting. My job shrank to creative direction: approve the structure, supply knowledge the AI lacked, and judge the result.

The full edit took one conversation. The Reel it produced is the one embedded in my musical fidget print article — so you can judge the output yourself.

More AI workflow experiments and automation guides in the AI category. I also cover n8n automation and self-hosted tools in Automation.