crossover

crossover

A Claude Code skill that builds a video where something physically crosses the Instagram feed: a car drifts over the black caption bar, paint pours over it, a curtain drops, something falls past it. The bar itself is a real screenshot placed in the start frame, and the video model animates the crossing motion on top of it. Nothing gets cut out or composited afterward, so there are no jagged edges and no chopped subjects.

What it does

Every failed version of this format fails the same way: someone generates the subject, then tries to cut it out and paste it over the bar in an editor. This skill skips that step entirely. It takes a screenshot of the Instagram caption bar, places it flat into the first frame of the video as a real graphic, and lets an image-to-video model animate the subject (the car, the pour, the falling object) moving in front of it, the same way it would animate anything else in the frame.

Install

Clone this repo into your Claude Code skills folder:

git clone https://github.com/coopersimson96/crossover.git ~/.claude/skills/crossover

Claude Code picks it up automatically. Restart your session if it was already running.

Quick usage

Ask for it directly:

Make a video where a BMW drifts across the Instagram feed and covers the caption bar.

Or reference the format by name:

Build a crossover video, pour of paint over the bar, landing in the post below.

The skill walks through building the start frame, writing the crossing prompt, and generating with your video model of choice. See SKILL.md for the full workflow and references/ for prompt templates, bar-asset handling, and per-engine notes.

Engine-agnostic

This works with any image-to-video model that accepts a start frame, not just one provider. It has been used with Seedance 2.5, Gemini Omni Flash, Seedance 2.0, and Kling 3.0. kie.ai is one option for running these models, not a requirement, swap in whatever image-to-video API you already have access to.

License

MIT, see LICENSE.