Vox Director
Agent skill that turns one topic into a Vox-style paper-collage explainer with Atlas Cloud and ffmpeg.
A Claude Code or Codex skill: the agent drafts a beat map, runs a style bake-off, generates collage posters and motion, adds narration and music, and assembles final.mp4. Every model call goes through the Atlas Cloud API; the look lives in editable prompt guides.
- I want a 60-second explainer in the Vox collage look for a topic, produced while I approve two checkpoints
- I have a product photo or a talking-head clip and want it restyled into a collage ad
- Single provider: everything runs through Atlas Cloud and costs credits per stage
- Model IDs drift; the skill fetches the live model list before each run
- Real people and brands are routed to Kling, which is pricier and has content filters
Last activity Aug 11, 2026. Fit: adaptable 3, operable 2.
Adaptable: what is yours to change
Make it yours
Config only8 things to edit
Export ATLASCLOUD_API_KEY, then change the look by editing references/prompt-guide.md (theme presets and prompt structures), the narrative arcs in references/beat-layer.md, and the voice roster in references/voices.md. Per-project beats.json is the working file; examples/ has starters.
- Env vars
ATLASCLOUD_API_KEY- Config files
beats.jsonexamples/- Prompt files
SKILL.mdreferences/prompt-guide.mdreferences/beat-layer.mdreferences/voices.mdreferences/models-and-gotchas.md- Accounts you need
- Atlas Cloud
Source
- github.com/Alisa0808/vox-director
- MIT license
- Last commit Aug 11, 2026
- 1,700 stars
Operable: what an agent can drive
Interface
NativeCLIpublished by the builderlocal
How an agent can drive an Operable Listing, in five levels: native (built for the agent as the primary interface), official (the Builder ships an agent interface), wrapped (a third party wraps the software's API), config (an agent can only drive it by editing files), gui (screen control only). An interface is official only when it is published by the same organization that ships the software. GUI-level Listings are allowed and sorted last.
- Docs
- github.com/Alisa0808/vox-director/blob/main/SKILL.md
- Tools
scripts/ one per pipeline stage
Reach
Complete coverage. 9 of 9 jobs reachable through the stated interface.
| Job | Through the interface | Evidence |
|---|---|---|
| Write a beat map from a topic | Reachable | README pipeline stage 1; beats.json per project; Gate 1 approval |
| Style bake-off across themes | Reachable | README stage 2; 9 theme presets in references/prompt-guide.md |
| Generate collage keyframe posters | Reachable | README stage 3 with nano-banana-2 |
| Animate posters into motion | Reachable | README stage 4; gemini-omni-flash or Kling for real people |
| Narration and background music | Reachable | README stage 5; xai/tts and minimax/music |
| Assemble with captions, ducking, watermark | Reachable | README stage 6; ffmpeg concat and Pillow overlays |
| Re-style an existing talking-head video (A-roll) | Reachable | README: ASR-segmented, gemini-omni-flash/video-edit with seedance fallback |
| Anchor a still photo in the collage (C-roll) | Reachable | README: nano-banana-2/edit; optional voice cloning |
| Element-level piece-by-piece motion | Reachable | README: optional local keyframe engine in references/local-engine.md |
Code stack
- Languages
- Python
- Frameworks
- ffmpegPillow
- Package manager
- pip
- Database
- none
- Hosting target
- local
- Built
- Unknown (recorded, never ranked)
Model stack
- Providers
- Atlas Cloud
- Models
- google/nano-banana-2/text-to-imagegoogle/gemini-omni-flash/image-to-videokwaivgi/kling-video-o3-pro/image-to-videoxai/tts-v1bytedance/seed-audio-1.0minimax/music-2.6
- Keys you need
ATLASCLOUD_API_KEY- Prompts live in
SKILL.mdreferences/prompt-guide.mdreferences/beat-layer.mdreferences/voices.md
Skills
Repo is a single voice-direction skill: SKILL.md plus reference prompt files and helper scripts, no app of its own.
- Kind
- pure
- Paths
SKILL.mdreferences/vox-director.skill
Evidence and confidence
| Field | Where it was seen | Confidence |
|---|---|---|
| formFactor | README: agent skill; Claude Code auto-discovers it, AGENTS.md for other agents | 95% |
| executionModel | README: one script per stage in fixed order; models generate images, motion, voice, music | 90% |
| operability | README: usable by any coding agent; the skill is the only interface | 90% |
| effort | README install: export ATLASCLOUD_API_KEY; 'What's in the box' lists the prompt files | 90% |
| editions | README license: MIT | 95% |
| active | GitHub pushedAt 2026-08-11 | – |
| reach | 85% |
Open questions
- Operable fit held at 2 because the software is prompts plus scripts around one third-party API; the Curator may judge it a thin wrapper or a real product on the strength of 1.7k stars
- Provenance: single-author agent-skill experiment; no statement on how the scripts were written
Sources
- github.com/Alisa0808/vox-director sourcing:video-streaming:curated