HyperFrames, but lightweight: same videos, far fewer tokens.
HTML in, MP4 out, with taste.
Ask an agent for a video and you get the same one every time: dark void, glow, everything fading up. This skill fixes that:
- Story first. Three storylines, you pick one. Then every adjective becomes a number.
- Defaults banned by name. The generated look is a list, and the list grows.
- Motion with a cause. Every cut is triggered by something on screen. Titles land and hold still.
- Real things only. Real UI, real numbers, from the source.
- Same render every time. Frame by frame, deterministic.
npx skills add CodeSquar/video-edit
Needs Node 22+ and FFmpeg. Rendering runs on HyperFrames.
Based on the HyperFrames skills by HeyGen, Inc. (Apache-2.0), condensed and modified.
/video-edit
Ask for the video. Pick a storyline. Approve the preview. Render.
One skill. Two references. No AI footage.