video-cog

**Judgement** `video-cog` is a powerful, high-level orchestration layer for long-form AI video creation built on CellCog agent teams. It is best suited for users who already have (or are willing to set up) the `cellcog` skill and want end-to-end video generation (script → visuals → audio → edit) from natural language prompts. **Strengths** - Covers the full multi-step pipeline: scripting, scene planning, image/video generation, voiceover, lipsync, music, and final editing. - Well-scoped to marketing, explainer, educational, documentary, UGC, and spokesperson formats with concrete prompt patterns and tips. - Asynchronous, notify-based workflow is appropriate for long-running, heavy jobs and avoids tight polling loops. **Key Risks / Limitations** - **Latency & performance:** Long-form multi-model video generation will be slow and resource-intensive; results are delivered asynchronously and may vary in quality. Not suitable for low-latency or high-frequency use. - **Reliability:** Complex multi-agent pipelines (6–7 foundation models) tend to have many failure points (script quality, scene coherence, lip sync, etc.). Expect occasional artifacts, inconsistencies, or task failures under load. - **Dependency on `cellcog`:** Requires prior setup of the `cellcog` skill and its SDK; if CellCog or its backing models change, behavior and quality can shift. - **Safety / misuse potential:** Can generate realistic spokesperson / lipsync content, which raises deepfake, impersonation, and misinformation risks. The documentation does not clearly specify guardrails, moderation, identity-consent checks, or content filters; any production use should wrap this skill with strict policy, filtering, and auditing. **Recommended Scenarios** Use `video-cog` when you: - Need **prototype or medium-stakes** marketing, explainer, or educational videos and can manually review outputs before publishing. - Want to experiment with **multi-agent orchestration for video** (script, visuals, audio) rather than implementing your own pipeline. - Are comfortable with asynchronous, batch-style workflows and moderate-to-high per-job cost and latency. Avoid relying on it for: - Real-time or near-real-time video generation. - Sensitive spokesperson content involving real individuals, brands, or regulated topics without additional safety, consent, and review layers.