vox-explainer

Set the ATLASCLOUD_API_KEY environment variable, install ffmpeg with libass, and ask for an explainer on your topic. The skill writes a beat-structured script, generates voiceover first to drive timing, creates consistent keyframes, animates them, adds music, and assembles a subtitled MP4 with ffmpeg.

Roughly $3-8 per 60-second film at July 2026 Atlas Cloud pricing, with animation dominating the cost at about $0.112 per second of video. The skill quotes a budget before the first expensive stage and offers a cheaper video tier for draft passes.

Use the style anchor pattern: generate one approved anchor frame, then produce remaining frames with Seedream 5.0 Pro's edit endpoint using the anchor as a reference image. This locks palette, texture, and framing across every keyframe in the film.

You need an ATLASCLOUD_API_KEY, ffmpeg and ffprobe built with libass for subtitle burning, and Python 3.10+ with the requests library. All model calls go through the Atlas Cloud unified API, while assembly runs locally with ffmpeg.

Yes. All state lives in project files like script.json, vo_durations.json, and per-beat assets, so any stage can be re-run in isolation. You can tweak one beat's narration, keyframe, or clip and regenerate only that part.

Animation models default to footage-like motion such as 3D camera moves and morphing, which breaks the collage look. The skill constrains every prompt to subtle motion graphics vocabulary like slow 2D push-ins and parallax drift, with explicit negative constraints against camera shake and style drift.