Five seconds of voiceover, three approval gates, one paper collage

What it does You feed it a roughly five-second voiceover line and it compresses the idea into a flat-color, halftone paper-collage B-roll clip. The output is a 9:16 silent MP4 where black-and-white photo cutouts and colored cardstock pieces slide into an empty frame and snap into place, stop-motion style, at 720×1280 and 24 fps.

The interesting bit The skill treats bureaucratic friction as a feature, not a bug. A three-gate approval pipeline forces you to sign off on the text-only metaphor and the static collage frame before gemini-omni-flash-preview ever wakes up. The README is refreshingly candid that this exists to stop you from wasting actual API money on bad prompts.

Key highlights

• Batch mode supports partial approvals, so one rejected metaphor does not stall the entire queue
• Gate 3 includes built-in QA: per-second frame extraction, empty-first-frame verification, and tail-frame comparison
• Ships with evals for gate behavior and a setup checker that validates Python, ffmpeg, and API keys
• Defaults strictly to Gemini Omni Flash; other models require explicit override

Caveats

• Requires a Codex environment with the built-in image_gen tool for static-frame generation
• The FAQ warns that slight paper-edge artifacts can appear in the first frame; strict empty-frame needs may require external editing
• Video generation is billed per use through the Gemini API

Verdict Editorial producers and short-form creators who treat API credits as a line item will like the guardrails. Those seeking a fully hands-off, fire-and-forget video generator should look elsewhere.