An open-source "skill" for AI coding environments, video-spec-builder, has been released to turn vague video ideas into shootable, shot-by-shot specifications through dialogue, drawing more than 600 GitHub stars shortly after launch.
Early July 2026 · feicaiclub
Video Spec Builder: the AI "director" that renders nothing
Instead of generating video itself, this open-source skill interrogates a vague idea and distils it into a shootable, editable spec — a shift from "prompt-to-video" toward spec-driven production.
600+
GitHub stars gathered shortly after release
1
Command to install: npx skills add
$0
Free & open source (MIT license)
The workflow: idea → detailed spec → HTML rendering
Input
Vague idea
"premium", "high-impact" — refused as too vague
→
Video Spec Builder
video-spec.md
Per-shot timing, camera, transitions, BGM, narration, captions
→
HyperFrames
MP4 output
HTML/CSS/JS → headless Chrome + FFmpeg, deterministic
Key specifications
License
Builder: MIT · HyperFrames: Apache 2.0
Compatibility
Agent-agnostic: Claude Code, Cursor, Codex
Scope
Runs locally · works across multiple agents
Output
video-spec.md with second-by-second storyboard
Why developers like it
The "short-tempered director" forces specific shots and motion
Flags missing captions, pacing and shot order
Easy iteration — fix by pointing out one sentence you dislike
Fits promo videos, AI ads, launch clips (Codex + Seedance)
The catch
Renders nothing by design — HyperFrames handles output
Requires setting up FFmpeg and Node.js separately
No major criticism so far; seen as reaffirming the value of prompt refinement
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…