AI Video Generation

**Judgement:** This skill is a strong, production-oriented orchestrator for AI video generation across modern providers. It is best suited for users who already work with cloud video models and want more reliable, scalable pipelines rather than a simple one-shot generator. **What it does well** - Integrates many current video providers (Sora 2, Veo 3, Runway Gen‑4, Luma, Seedance, etc.) with clear endpoint and model-ID handling. - Emphasizes *task-based routing* (draft vs premium, long-form, i2v, multi-shot consistency, local privacy-first) instead of brand loyalty. - Promotes good prompt engineering for motion: shot-level instructions (subject, action, camera, lens, lighting, timing, continuity). - Explicitly designs for async and failure: queued jobs, polling/backoff, retries, cancellation, and URL expiry handling. - Keeps preferences and history local in `~/video-generation/`, with no API keys stored in project files. **Key risks & limitations** - **Third-party dependency & privacy:** All generations rely on external providers; prompts and optional media are sent over the network. Not suitable for organizations requiring fully on-prem, no-cloud workflows. - **Cost exposure:** Strong support for large async pipelines means it’s easy to submit many jobs; without budgeting and caps, users can burn credits quickly. - **Complexity:** Requires reading the setup and reference docs, understanding model naming, and managing multiple providers; less ideal for beginners who want a simple GUI-like flow. - **API evolution:** The skill depends on specific model IDs and provider APIs that can change; keeping it optimal will require periodic updates. **Recommended scenarios** - Power users or teams generating a lot of AI video who need **reliable, multi-provider pipelines** and want to separate “cheap drafts” from “premium final renders.” - Technical users comfortable with async jobs, queues, and polling who want to **automate or batch** video generation workflows. - Creators who care about **shot-level control and cinematic consistency**, including image-to-video and reference-based workflows. - Users who accept cloud providers for content processing but want **local control of preferences and history**. **Less suitable for** - Users who need fully offline or strictly on-prem workflows with **no external API calls**. - Non-technical users looking for a **minimal, one-click tool** rather than a configurable, pipeline-focused skill.