louisedesadeleer/color-correct

color-correct is a skill for Claude Code that turns the AI into a video colorist. Given a piece of footage, it looks at the actual clip, diagnoses problems like color casts or bad exposure, applies a color grade, and shows a before and after comparison before rendering anything, all built on top of ffmpeg without destroying the original file. It ships with nine named color looks, with names like Golden Hour, Honey, Oat Milk, and Espresso, each one reverse engineered by studying the color choices of specific video creators the author admires. Every session starts by showing a menu, a labeled grid of all nine looks applied to your own footage, since the author believes color choices are better made by seeing them than by reading descriptions. From there, a user can pick a specific named look, ask the AI to recommend one based on what kind of footage it is, or ask it to copy the color style of a particular creator by analyzing several of their published videos and finding the consistent pattern across them rather than being fooled by one unusual scene. Each look also ships as a standard cube format 3D LUT file, a common format for color grading that works directly in editing software like Premiere, DaVinci Resolve, Final Cut, or CapCut, or even plain ffmpeg, so people who do not use an AI agent can still use the looks. The difference is that a LUT alone cannot fix problems like a bad white balance first, while the AI agent handles that correction before applying the chosen look. Installing it means cloning the repository and pointing Claude Code at its skill file. It requires ffmpeg, Python with the Pillow and numpy libraries, and optionally yt-dlp for the mode that copies a creator's style from a video link. This is aimed at video editors and content creators who want quick, tasteful color grading without learning traditional colorist tools.