AI agent skills that turn Claude (or any Markdown-skill-capable agent — Claude Code, Cursor, Codex, Copilot) into a video director for Capturable: plan multi-scene videos with consistent characters, generate images and product shots, and render everything through one API.
Works on the free plan out of the box (flagship Capturable Gen 2 models, self-hosted GPUs). Paid plans unlock aggregated providers — Kling, Google Veo, and more — through the same calls.
- Real API, no Playwright. Every skill calls
https://api.capturable.aidirectly with your API key. No browser windows, no flaky selectors, no "prompt bar didn't clear" bugs. - Plan before you spend. Planning is a costed review step: Claude drafts the full scene plan and shows you the credit estimate before any GPU renders.
- Scene continuity. Characters, products, and branding stay consistent across every scene of a multi-shot video — the thing single-clip generators can't do.
- One pipeline, end to end. Idea → scene plan → keyframes → motion → final video, polled and downloaded by your agent.
/plugin marketplace add CapstraLabs/capturable-skills
/plugin install capturable@capturable
npx skills add CapstraLabs/capturable-skills
Clone this repo and copy the skill folders into your agent's skills directory (e.g. .claude/skills/ or .cursor/skills/):
git clone --depth 1 https://github.com/CapstraLabs/capturable-skills.git
Agent-driven install: paste INSTALL_FOR_AGENTS.md into your agent.
- Create a free account at capturable.ai.
- Create an API key at capturable.ai/settings/api-keys (
aicl_…). - Export it:
CAPTURABLE_API_KEY=aicl_...(env var, or your agent's secret store).
| Skill | Invoke | Description |
|---|---|---|
| capturable-video-director | /capturable:video-director |
Flagship. Claude becomes the director: prompt → costed multi-scene plan → review → continuity render → final video. Characters, branding, and places stay consistent across scenes. |
| capturable-generate | /capturable:generate |
Fast single operations: text-to-image, image-to-video animation, one-call quick videos, job polling, model listing, quotes, credit balance. |
| capturable-product-shot | /capturable:product-shot |
Product placement photography: your real product image composited into generated scenes (studio, lifestyle, hero, seasonal). 1–4 variants per call. |
| capturable-character | /capturable:character |
Register reusable characters and brand assets (logo / product / packaging) from photos. Returns IDs consumable by the director and UGC skills. |
| capturable-ugc-ads | /capturable:ugc-ads |
UGC-style ad pipeline: presenter character → hook-first 9:16 script → continuity render. Includes hook frameworks, beat timelines, and platform presets (TikTok / Reels / Shorts). |
The skills chain: register a character or brand asset with capturable-character → reference it in capturable-video-director or capturable-ugc-ads → render. capturable-generate and capturable-product-shot are self-contained.
| What you want | Skill | Note |
|---|---|---|
| Full multi-scene video with consistent characters | capturable-video-director | Plan is reviewed (with cost) before rendering |
| One-shot video from a prompt, no review step | capturable-generate | POST /videos/quick — plan + auto-approve + render |
| Image from a prompt | capturable-generate | Capturable Gen 2 Image on free plan |
| Animate a still image into a clip | capturable-generate | first-frame animation |
| Product photo with my real product in it | capturable-product-shot | Image-edit recipe keeps the product pixel-faithful |
| Reusable face/character across videos | capturable-character → director | Consent-tracked character references |
| Brand logo/packaging in every scene | capturable-character (branding) | brandingType: logo | product | packaging |
| TikTok/Reels UGC ad | capturable-ugc-ads | 9:16, hook-first beat structure |
| Cost before committing | any | POST /quote is free and authoritative |
| Use Kling / Veo instead of Gen 2 | any (Pro plan) | Same calls, set videoModel; GET /models lists what your plan allows |
- Free / Connect — API keys are pinned to the self-hosted Capturable Gen 2 stack (
capturable-gen2-image,capturable-gen2-video): real renders, zero aggregator spend, weekly allowance. - Pro+ — full-mode keys choose any plan-eligible model (Kling, Google Veo, Seedance, GPT Image, …) via the same endpoints.
Skills always call GET /v1/integrations/models first, so they only ever offer models your plan can actually render with.
- INSTALL.md — all install paths
- COOKBOOK.md — end-to-end recipes (brand ad, UGC batch, product catalog refresh)
- OpenAPI:
https://api.capturable.ai/v1/integrations/openapi.json - Prefer MCP?
npx -y @capturable/mcp-serverwraps the same API as tools.
MIT — see LICENSE.