Media Judge — guide skill for tools & renderers

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.

A finished video, image or track becomes a list of timed, measured defects with evidence frames, and takes ranked by defects.

What it does

Judge a finished AI-made video, image or track by measuring it. One command, judge.py, reads every frame and every sample and writes facts with timestamps and evidence: broken frames, banding after a simulated X or YouTube upload, numbers that jiggle, spelling, too much on screen, pinned graphics, sliding labels, a face, outfit or light that changed, sound after the cut, loudness, voice against the script, and distance to references against the template. There is no score out of 10.

It works in three steps. Code finds the moment. A fresh agent that did not make the media opens only the flagged evidence and answers narrow questions. The verdict is a table of counted defects, each with a time, a number, a file and a fix. It writes report.md, facts.json, contact sheets and evidence frames, and judge.py rank orders takes by blockers, then major flags, then minor flags, then the reference margin. It can also compare against a reference or a template, and turn a finished piece into a measured shot list. It runs on a laptop CPU with ffmpeg (built with libvmaf) and uv.

Tested on planted defects in footage it never saw while being tuned: 18 of 22 caught and 6 of 8 clean controls clear. In a blind test on 19 anonymous clips, an agent with this skill found 12 of 14 planted defects with no false claims on clean clips, using 116k tokens. An agent given only the failure list found 11 of 14, using 197k tokens. Both missed a face swap they were not told to look for, so pass --one-character when the brief says one person.

Expect to overturn flags: on a real 47 s film, all 4 spelling candidates opened were OCR misreads. A clear report does not mean a good video. Code proves something is broken or correct, not that it is good.

Use when

Not for

Install

Example output

https://s3.varg.ai/files/4325c578-378d-4c43-ae53-3cc3218df1ea/file_cee9681521af.jpg — example (credit: varg)

More tools & renderers skills