A 3-tool Claude Code stack — Hyperframes, video-use, and Whisper — that turns a raw phone recording into a finished short in under 10 minutes with one command.
This guide is reviewed for clarity, service accuracy, and AI-search readability. The next quarterly content review is tracked internally before unsupported metrics or client proof are added.
What this pipeline replaces
CapCut Pro ($8/month). Descript ($24/month). Premiere Pro ($23/month). A freelance editor at $50–200 per video. The 90-minute manual edit for each short.
The pipeline below replaces all of it. Three open-source tools wire together under Claude Code. You record a talking-head video, hand the file to Claude, run one command, and walk away. The finished short lands in your output folder.
- Removes filler words, awkward pauses, and retakes automatically
- Generates word-level transcription so captions and motion graphics sync to exactly what you said
- Authors motion graphics in HTML and GSAP — reusable templates across every future video
- Renders deterministically — same config, same output, every time
The 3-tool stack
- Hyperframes — open-source Apache 2.0 framework that turns HTML + CSS + GSAP into video output via headless Chrome and FFmpeg. Repo: github.com/heygen-com/hyperframes
- video-use — auto-trims raw footage, removes filler words, awkward pauses, and retakes. Uses ElevenLabs Scribe transcription for cut detection.
- Whisper Transcribe — word-level timestamps so every caption and overlay lands in sync with what was said. Runs locally, free.
Install — 10 minutes
bash# Step 1 — Install Hyperframes npx skills add heygen-com/hyperframes # Step 2 — Initialize in your project folder npx hyperframes init # Step 3 — Install video-use and set ElevenLabs key export ELEVENLABS_API_KEY=your_key_here # Step 4 — Verify npx hyperframes lint # preview npx hyperframes render # test render
The daily workflow — 5 steps
- Step 1 — Record. Phone selfie camera, talking head. Don't worry about filler words — the pipeline strips them. 60–90 seconds of raw footage typically becomes a 30–45 second finished short.
- Step 2 — Trim with video-use. Drop the raw file in your input directory. Run one command. Dead air and retakes are removed.
- Step 3 — Transcribe. Hyperframes calls Whisper automatically for word-level timestamps. Auto-runs as part of the Hyperframes setup.
- Step 4 — Author motion graphics. Hyperframes uses your index.html with GSAP animations. Claude Code writes or updates this HTML for you based on the transcript.
- Step 5 — Render. Run
npx hyperframes render. The finished mp4 lands in your output folder.
The one-command daily workflow
bash# Drop your raw recording in input/, then: npx video-use trim # remove dead air npx hyperframes render # produce finished mp4 # Your video is in output/
Why this beats CapCut
- Deterministic rendering — edit configs are code; same config, same output, every time
- Reusable templates — build one Hyperframes template, every future video uses it automatically
- Full AI pipeline — script to finished short runs through one AI with no tool-switching
- Time math — 90-minute edit → under 10 minutes. Posting daily: 10 hours a week back
Frequently Asked Questions
Is this free to run? The core pipeline (Hyperframes + local Whisper) is free. video-use uses ElevenLabs Scribe for better cut detection, which has a per-minute cost. Switch to local Whisper for trimming if you want a fully free setup.
Do I need to know HTML or GSAP? No. Claude Code writes the index.html for you based on your style description. Describe what you want and Claude generates the animation code.
What video formats does this output? Hyperframes renders to mp4 via FFmpeg. Specify aspect ratio in your config — 9:16 for Reels/TikTok, 1:1 for feed posts.