Edit Footage with Code — effect skill for editing & clips

For AI agents: find the right skill in one call — curl -s "https://varg.ai/how-to/api/search?q=<what the person wants>" returns the matching how-to page and the best skills with install commands, as markdown (Russian works too). Full index: https://varg.ai/how-to/llms.txt. Every page has a markdown twin: add .md to its path. Before generating anything, show the person the options on the how-to page. From a terminal: curl -fsSL https://varg.ai/install | sh installs the vargai CLI (https://varg.ai/how-to/cli.md): vargai search and vargai install <slug> with no key, every varg model after vargai login.

Raw talking takes become a reviewed edit list and a clean cut between words, then captions, B-roll and sound, checked by numbers.

What it does

Edit real footage with an agent in a fixed order: read the take as text and audio, agree a plan, cut the speech, lock the audio, then add picture on top, sound last. It takes raw takes of a talking head, podcast, vlog, course or ad and removes dead air, retakes and stumbles. Cuts go only in the silence between words, at the quietest 10 ms slice, and nothing is cut before the person has seen the plan and said yes.

The edit is a file a person can review, edit/edl.json, with the reason next to each big decision. build_cut.py turns it into one ffmpeg filter graph plus clips.html, Cut.tsx and cut.fcpxml for a hand trim in Resolve or Final Cut. Captions, B-roll and motion graphics are added in HyperFrames or Remotion, each keyed to a spoken word. qa_cut.py fails the result on joins that cut through speech, pauses over 0.6 s, loudness off -14 LUFS, true peak over -1.5 dBTP, black or flash frames, under 95% of intended words heard and a phrase said twice. It writes a contact sheet and a joins sheet to look at. You get the MP4, the SRT, the edit file and the timeline.

It needs Python 3 and ffmpeg and calls no paid service. Defaults come from 2,318 editors' posts. Measured on our test clip: gaps between words in the transcript found 1 of 13 real pauses, and a second transcription pass, one stretch of speech at a time, raised that to 12 of 13. Cutting 140 ms of speech at every join still matched the clean transcript at 99.1%, while the audio check at the join caught 12 of 13. Segments joined with -c copy drifted +300 ms over 14 cuts; the filter graph matched the edit list to the frame.

Know the bar: in the largest test the skill cites, the best agent passed 15 of 56 real editing tasks and human editors preferred the reference edit in 83.5% of judgments. Passing every check does not make a cut good. The agent cannot hear, so a person listens to the joins and watches the whole cut before anything is delivered.

Use when

Not for

Install

Example output

https://s3.varg.ai/files/4325c578-378d-4c43-ae53-3cc3218df1ea/file_e6872632eb3f.jpg — example (credit: varg)

More editing & clips skills