Claude Code edits a gym-fridge nature documentary
Makelky reports a published 89.6-second nature-documentary parody about his gym fridge, edited by Claude Code with approved HeyGen narration, corrected crops and a looping final render.
aaronmakelky.com
Original prompt
Shared by the source author. See source notes for context.
i need to approve the voiceover script before you start generating
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Aaron Makelky
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Makelky reports a published 89.6-second nature-documentary parody about his gym fridge, edited by Claude Code with approved HeyGen narration, corrected crops and a looping final render.
- Human involvement
- Makelky approves the script, reviews rendered corrections and authorises publication. The account discloses a product partnership and affiliate links.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Copy and verify the twenty-three clips without modifying the camera card.
- Match footage to the shot list and prepare the rough cut.
- Approve the narration script before generating the designed voice.
Aaron Makelky films twenty-three clips of stocking his home-gym fridge and directs Claude Code to make a nature-documentary parody. The agent copies footage from the camera card, verifies file sizes, compares shots with a Notion beat list and builds the edit using FFmpeg and Python.
Makelky approves the narration script before paid voice generation. HeyGen supplies a voice he designed and word timestamps, which the agent uses for captions. Because his FFmpeg build cannot draw text, Python creates transparent caption images for overlay instead.
Review exposes practical errors: vertical crops lose the fridge and bottle, a music cut sounds broken, and captions exceed his stricter saved margins. The agent produces a horizontal master and vertical alternatives, smooths the music transitions and narrows caption placement. It also removes a black opening frame and joins the ending to the beginning for a loop.
The final vertical video runs 89.6 seconds and is published on YouTube Shorts. Makelky discloses a Liquid Death partnership and affiliate camera links. His explicit instruction prohibits publishing without approval. The useful outcome is a completed comic edit with visible corrections, not proof that the first instruction was sufficient: the video skill was read late, and several requirements had to be supplied during review.
Prompt
Original prompt
Source: Blog · Aaron Makelky
Frequently asked questions
When was voice generation authorised?
After Makelky explicitly approved the narration script.
Why generate caption images?
His FFmpeg build could not draw text, so Python rendered transparent caption states using HeyGen’s word timestamps.
What did review catch?
Incorrect vertical crops, abrupt music silence and captions outside the creator’s stricter saved safe margins.
More details
Remaining steps
- Review crops, music transitions and caption margins on rendered versions.
- Create a looping final export and publish only on explicit approval.
About this account
Claude Code ingests footage, edits through FFmpeg and Python, calls approved HeyGen narration, constructs timed captions and corrects reviewed exports.
Firsthand account by Aaron Makelky; original publication date verified against Primary creator field-note October9,2026. Outcomes are the author’s reports, not independently reproduced. Original execution-time instruction from the author; the full editing prompt and narration text are not reproduced.
Credit: Aaron Makelky
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · FFmpeg · Python · HeyGen CLI · Notion shot list · Linear ticket
Gym-fridge parody · Approved synthetic narration · Looping vertical edit
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.