The AI Video Editing Pipeline That Replaced CapCut

This video pipeline is edited entirely by AI. No CapCut, no Premiere, no Descript, no editor for hire. The motion graphics, the captions, the jump cuts, the dead-air removal: all generated from one raw phone recording by one Claude skill stack. The total human effort is hit record, hand the file to Claude, hit one command.

What You'll Learn

• How the Hyperframes pipeline works (record, trim, animate, render)
• The exact tools in the stack (Hyperframes, video-use, Whisper transcribe, GSAP)
• How to install everything in 10 minutes
• The one-command workflow you run after every recording

What this replaces: CapCut Pro ($8/mo). Descript subscription ($24/mo). Premiere Pro ($23/mo). A freelance editor at $50-200 per video. The Notion doc full of edit notes. Roughly 90 minutes per video staring at a timeline, which at daily posting is about 10 hours a week.

The Stack

Three components do the work. All open source. All free.

1. Hyperframes. Open-source Apache 2.0 framework that turns HTML + CSS + GSAP into deterministic video output via headless Chrome and FFmpeg. The motion graphics layer. Authoring is in code, rendering is one command. Repo: github.com/heygen-com/hyperframes

2. video-use. Auto-trims raw footage. Removes filler words, awkward pauses, retakes. Default backend uses ElevenLabs Scribe transcription (better cut detection than Whisper). The "remove dead air" part of the pipeline, the pre-cleaner before Hyperframes layers motion graphics.

3. Whisper transcribe. Word-level timestamps so motion graphics and captions land in sync with what you said. Hyperframes reads the transcript and uses it to time every overlay, cut, and caption frame-perfect. Local Whisper runs free.

The Workflow

Five steps. The first is the only one that requires you to do anything other than wait.

1. Record. Phone selfie camera, talking head. Don't worry about filler words, pauses, or restarts. The system strips them out. Aim for one continuous take so the audio is clean. 60-90 seconds of raw footage typically becomes a 30-45 second finished short.
2. Trim with video-use. Drop the raw recording into your video-use directory. Run one command. ElevenLabs Scribe transcribes, identifies filler/silence/retakes, and outputs a cleaned cut. The dead air is now gone.
3. Transcribe with Whisper. Hyperframes calls Whisper to get word-level timestamps on the cleaned footage. This is what enables motion graphics to appear at the exact word they reference. Auto-runs as part of the Hyperframes setup.
4. Author motion graphics in HTML. Hyperframes uses an index.html file with GSAP animations to define every overlay, caption, transition, and motion graphic. Once you've built one template, you reuse it across every video you make. Claude Code can write the HTML for you based on the transcript.
5. Render with one command. Use npx hyperframes lint and preview to check it, then npx hyperframes render to produce the final mp4. Headless Chrome plus FFmpeg do the heavy lifting. You walk away. The finished video lands in your output folder.

Install: 10 Minutes

All three components install via npm, one after the other. No paid subscriptions required for the core pipeline.

Install Hyperframes: run this inside Claude Code:

npx skills add heygen-com/hyperframes

This installs the Hyperframes framework and the related Claude skills (hyperframes, hyperframes-cli, gsap, hyperframes-registry) at user scope. After install, run this in your project folder to scaffold the index.html and config:

npx hyperframes init

Install video-use: clone the video-use repo into your developer folder, then symlink to your Claude skills directory. Set the ELEVENLABS_API_KEY env variable for Scribe transcription. Optional: switch to local Whisper if you don't want to pay per minute.

Run the pipeline: after install, the daily workflow is:

1. Record video
2. Drop the file in input/
3. npx video-use trim
4. npx hyperframes render
5. mp4 in output/

One folder, three commands, finished short.

Why This Beats CapCut

Deterministic rendering. Edit configs are code. Rerun the pipeline 100 times, get the same output 100 times. CapCut's GUI-based timeline drifts every time you reopen the project. Code-based authoring removes that entire failure class.

Reusable templates. Once you've built one Hyperframes template (your typical motion graphic style, font, captions), every future video uses the same template. CapCut requires you to redo the same overlays per project.

AI wrote the script, AI edits the video. If you use Claude to write the script and Hyperframes to edit it, the entire pipeline from idea to publish-ready short runs through one AI. No tool-switching, no context loss, no re-uploading.

The Compound Effect

Set up the pipeline once. Every video after that takes the time to record plus 5-10 minutes of render. If you currently spend 90 minutes editing each short, you save 80+ minutes per video. Posting daily, that's 10 hours a week back.

What you do with the 10 hours: more recordings, more iteration on hooks, more A/B testing of formats. The pipeline doesn't make you a better creator. It removes the editing tax so you can focus on the part of the job that compounds: the hooks, the angles, the testing.

Go Deeper

Prefer a printable version? Grab the PDF: https://drive.google.com/file/d/1fNGr-y3qRvjp92oLuGaSLHyMQNLC7gkN/view?usp=drivesdk

Want the whole system installed live, with Cooper on the call? Build Your Own AI, the two week sprint runs the first two weeks of August.

Or get a new working tool every Monday inside The AI Operator community on Whop.