Alisa0808/vox-director

What it solves

Vox Director automates the creation of editorial-style "paper-collage" explainer and advertisement videos. It eliminates the manual effort of scriptwriting, asset creation, animation, and audio synchronization by turning a single topic into a finished MP4 file.

How it works

The project operates as an agent skill that integrates with coding agents (like Claude Code or Codex) and uses the Atlas Cloud API and local ffmpeg. The process follows a structured pipeline:

1. Beat Mapping: A narrative arc is chosen and a beats.json file is created for user approval.
2. Style Bake-off: Multiple visual themes are rendered for the user to select the desired look.
3. Keyframe Generation: High-quality collage posters are generated for each beat using text-to-image models.
4. Motion: These posters are animated via image-to-video models or a local keyframe engine for piece-by-piece assembly.
5. Audio: Narrations and background music are generated using TTS and music models.
6. Assembly: ffmpeg is used to concatenate clips, duck music under the voice-over, and burn in captions.

It supports three input modalities: B-roll (topic-to-video), A-roll (re-styling existing talking-head video), and C-roll (anchoring a single still photo into the collage).

Who it’s for

Content creators, marketers, and developers who want to produce high-quality, stylized explainer videos quickly using AI agents without needing professional motion graphics software.

Highlights

• Automated Collage Aesthetic: Specifically targets the "Vox-style" look with torn paper, halftone dots, and newspaper clippings.
• Multi-modal Inputs: Can generate videos from a text topic, an existing video, or a single photograph.
• Agent-Integrated: Designed to be discovered and executed by coding agents as a skill.
• Human-in-the-loop: Includes decision gates for approving the narrative structure and visual style.