Analysis updated 2026-05-18
Generate a matching B-roll clip for a short voiceover line in a video project.
Batch-produce paper-collage style B-roll for multiple voiceover lines at once.
Add a distinct editorial visual style to short-form video content.
Review and approve a visual concept and still frame before spending on video generation.
| pyang5166/gbro-collage-broll | fluxions-ai/vui | karpathy/hn-time-capsule | |
|---|---|---|---|
| Stars | 654 | 659 | 649 |
| Language | Python | Python | Python |
| Last pushed | — | — | 2025-12-10 |
| Maintenance | — | — | Quiet |
| Setup difficulty | moderate | hard | moderate |
| Complexity | — | 4/5 | 2/5 |
| Audience | vibe coder | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Needs a Gemini API key (billed by usage), Python 3.10+, ffmpeg, and a Codex-style agent environment with image generation.
gbro-collage-broll is an agent skill, meant to be dropped into an AI coding agent's skills folder, that turns a short voiceover line, around five seconds long, into a polished B-roll video clip. The visual style is a halftone paper collage look: flat colored paper backgrounds, black and white halftone photo cutouts, and bits of colored cardstock, with pieces sliding in and locking into place one at a time like stop motion, rather than a simple fade in or slow zoom. The delivered clip is a silent, vertical 9 by 16 video, five seconds long, at 720 by 1280 resolution and 24 frames per second, meant to sit directly under a voiceover. The core of the skill is not just a prompt template but a required three stage approval process, so that a person's attention goes toward judging the creative idea rather than burning money on generation costs. In the first gate, the skill proposes a visual metaphor, meaning the core idea, the key objects, the background color, and the order pieces assemble in, without generating any image or video yet. Once that is approved, the second gate generates a colored still collage frame plus a contact sheet for review. Only after the still frame is approved does the third gate generate the actual video, using Google's Gemini Omni Flash model to build an animation between a first and last frame, along with quality checks like frame by frame sampling and verifying the first frame starts empty. When processing many voiceover lines at once, only the lines that pass each gate move on to the next stage. Getting set up requires a Codex style agent environment with a built in image generation tool for the still frame stage, a Gemini API key from Google AI Studio since video generation is billed by usage, Python 3.10 or newer, the google-genai Python package, and ffmpeg for handling video frames and building the contact sheet. A setup check script runs automatically the first time the skill is triggered and points out anything missing. The README does not mention a license for the code.
An AI agent skill that turns a short voiceover line into a stop-motion style paper-collage B-roll video using Gemini's video model.
Mainly Python. The stack also includes Python, Gemini Omni Flash, google-genai.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.