gbro-collage-broll

Generates halftone paper-collage B-roll videos from voiceover scripts using Gemini Omni Flash.

1.3k|111|Updated Jul 15, 2026
One-click install
npx skills add https://github.com/pyang5166/gbro-collage-broll --skill gbro-collage-broll-pyang5166
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gbro-collage-broll
Source: https://github.com/pyang5166/gbro-collage-broll
Command: npx skills add https://github.com/pyang5166/gbro-collage-broll --skill gbro-collage-broll-pyang5166

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires google-genai, httpx, and includes scripts (resource) components.

What problem does it solve? Turning a short voiceover line into a visually coherent B-roll clip normally requires design skills, stock footage hunting, and expensive trial-and-error with video generation APIs. This Skill converts a ~5 second script into an editorial halftone paper-collage assemble-from-empty animation while preventing wasted API spend through mandatory human approval gates. ## Core Features & Use Cases - Three-Gate Approval Workflow: Gate 1 proposes only a visual metaphor (core meaning, key objects, color field, assembly order); Gate 2 generates the final collage still plus a numbered contact sheet; Gate 3 runs video generation only after both confirmations. Batch mode supports partial approval per item. - Gemini Omni Flash Video Generation: Uses first/last-frame interpolation with gemini-omni-flash-preview via bundled Python scripts, delivering 9:16, 5-second, 720x1280, 24fps silent MP4 files with forced audio stripping. - Built-in QA Pipeline: Per-second frame extraction, empty first-frame verification, end-frame comparison against the approved still, and consolidated contact sheets for batch projects. - Use Case: A content creator pastes five voiceover lines, reviews five visual metaphors, approves three, confirms the generated collage stills, and receives three ready-to-use vertical B-roll clips to layer under narration. ## Quick Start Say to your agent: use gbro-collage-broll to turn this 5-second voiceover line into a halftone paper-collage B-roll, then confirm the metaphor, the still frame, and let it generate the video.

Frequently Asked Questions about gbro-collage-broll

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn a voiceover script into a paper-collage B-roll video?

Provide your ~5 second script line and the skill proposes a visual metaphor first. After you approve the metaphor and then the generated collage still, it calls Gemini Omni Flash to produce a 5-second 9:16 silent MP4 with QA contact sheets.

What video model does this collage B-roll workflow use?

It defaults to gemini-omni-flash-preview using first/last-frame interpolation through the bundled generate_video.py script. The model only changes if you explicitly specify a different one; Veo is retained solely as a legacy compatibility script.

What dependencies are required for Gemini Omni Flash video generation?

You need a GEMINI_API_KEY from Google AI Studio, Python 3.10 or newer, google-genai 2.10.0 or newer in a shared venv, and ffmpeg/ffprobe. The check_setup.sh script verifies all of these and prints setup guidance for anything missing.

Can I batch process multiple script lines with partial approvals?

Yes. Each script line gets its own metaphor, still, and video. When you approve only some items at a gate, only those advance to the next stage while the rest stay in revision, and failed video jobs can be rerun individually.

When should I not use this paper-collage B-roll skill?

Avoid it when you need precise layer control, editable timelines, or camera moves, which require layered animation tools like HyperFrames. It also does not fit real-person ads, talking-head footage, or requests for transparent per-layer assets.

Why does the generated video show paper pieces at the first frame edge?

Slight early exposure of paper elements at the first frame edge is an accepted limitation of the Omni Flash generation. If you need a strictly empty opening frame, the documentation recommends prepending footage with a timeline-based animation tool.