Check a render against its reference by Refs — guide skill for launch videos

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.

Score your render against the reference film: cut timing, brightness, motion, loudness, format, frame-exact fixes.

Type: Guide · Launch videos
Slug: refs-check-render
Status: live
Publisher: Refs (vendor)
License: Refs: proprietary hosted service (free search, Pro for the rest)
Source: https://refs.video/connect
Works in: Claude.ai, Claude Code, Codex, Cursor, Any MCP client
Requires: Refs MCP signed in with Pro (2 credits a check), Node 20+ and ffmpeg for scripts/measure.mjs from the Refs starter kit
Cost: search_videos is free with no key or account (about 30 searches an hour without one). Everything else needs Refs Pro, $16 a month or $120 a year, with 3,000 credits a month: get_blueprint 5, get_frames 5, get_formula 5, plan_video 10, check_render 2, the rest 0. Library ($6 a month) opens every film on the site. varg credits are only spent on what you generate on varg.

What it does

check_render (2 credits on the Refs MCP) closes the loop after a build. Measure your MP4 with the starter kit's scripts/measure.mjs (node scripts/measure.mjs out/video.mp4 > render.json), then ask your agent to call check_render with the reference you built from, that JSON as render, and target set to where you'll post it (x, youtube, tiktok or web). It scores cut timing, brightness, motion, loudness and format against the reference's measured timeline and returns frame-exact fixes: which cut is late, which stretch parks, how many LU you are off the platform's target. Fix, re-render, measure, repeat. Works for any render, from HyperFrames, Remotion, the starter kit or a cut made on varg.

Use when

After the first render of a film built from a reference
An agent loop that keeps re-rendering until timing and loudness match
Checking a launch video meets X, YouTube, TikTok or web loudness before posting

Not for

Judging taste: it measures timing, light, motion and sound, not whether the film is good
Films with no reference to compare to

Install

Claude Code: claude mcp add --transport http refs https://refs.video/mcp
Claude.ai: Customize → Connectors → Add custom connector → https://refs.video/mcp
Codex: codex mcp add refs --url https://refs.video/mcp
Any MCP client: https://refs.video/mcp
Any agent, with the vargai CLI: vargai install refs-check-render

Example output

https://s3.varg.ai/files/4325c578-378d-4c43-ae53-3cc3218df1ea/file_2bfd9f0dbec0.mp4 — 7 s cut from Refs' film: a film page's measurements, the cut map and the pacing plate, the numbers check_render compares against. (credit: Refs, the Refs film (refs.video/film.mp4), https://refs.video)