Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
A/B video pipeline that generates multi-scene clips with Veo 3, serves a browser preview for iterative feedback, and stitches approved clips into final MP4 combinations with ffmpeg. Built around 15-second videos split into hook, core, and CTA segments, so any segment can be swapped to produce multiple video variants from a single storyboard. Token spend is limited to writing the storyboard and interpreting feedback; preview, generation, and assembly run outside the AI loop.
Because preview and assembly run outside the AI loop, you can review and iterate on Veo 3 outputs without paying token costs on each pass.
When to use it
- Creating A/B test ads with swappable hooks and CTAs
- Generating multi-scene product demo videos
- Previewing and iterating on clips before final assembly
- Re-generating only revised scenes after human review
- Building vertical and square format variants from a shared storyboard
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: create a 15-second A/B test ad with two hook variants and two CTA variants
- 1Write storyboard.json defining 5 scenes (hook_a, hook_b, core, cta_a, cta_b) and _ab_combinations for all 4 video variants
- 2Run batch_generate.py to call Veo 3 and download all 5 clips
- 3Run generate_preview.py to open preview.html in the browser for clip review
- 4Receive feedback JSON and run apply_feedback.py to re-generate only scenes marked revise
- 5Run ffmpeg_assembler.py to stitch approved clips into final_AA.mp4, final_BA.mp4, final_AB.mp4, final_BB.mp4
4 final MP4 files covering all A/B hook and CTA combinations, ready for upload and analytics
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.