Visual Skills
Agent skills that direct AI video and image generation like a film director.
About
Visual Skills is an open-source collection of agent skills (CC-BY-4.0) that turns coding agents into AI film directors. The video sub-skill defines scene desire, obstacles, camera geometry, and edit rhythm before generating prompts in exact Seedance, Kling, and Veo syntax; the image sub-skill handles storyboarding and keyframes routed to Nano Banana or GPT Image.
It fills a real gap: raw video models produce better results when prompted with cinematic structure rather than bare descriptions, and these skills encode that dramaturgy as reusable agent workflows. Honest caveats: the skills only generate prompts, not finished videos — actual generation depends on having access to the underlying model APIs — and some model syntax docs are in English.
What it can do
- Cinematic video prompt generation for Seedance/Kling/Veo
- Image keyframe and storyboard routing to Nano Banana/GPT Image
- Scene dramaturgy (desire, obstacle, camera geometry, rhythm)
- Reusable SKILL.md workflows for agents
Setup
Clone https://github.com/smixs/visual-skills and install via skills.sh or copy the skill folders into your agent's skills directory.
Prerequisites
A Claude Code-compatible agent; access to Seedance, Kling, Veo, or image model APIs for actual generation.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 59; Trending score 62/100 from researcher assessment.
★ 292 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.