What problem does it solve? Turning a single character reference image into a coherent multi-scene video usually fails because each scene drifts in character appearance and long API generations block the conversation. This Skill orchestrates a four-phase, user-reviewed pipeline that keeps the character consistent across scenes and offloads heavy generation to sub-agents. ## Core Features & Use Cases - Four-phase approval workflow: Script drafting, user confirmation, image generation, then video generation, with review gates between each stage. - Character consistency controls: Uses the original reference image directly as img2img input, enforces character-preservation negative prompts, and applies a three-layer prompt structure (preservation, motion, scene). - Delegated generation and stitching: Runs Agnes image and video API calls via delegate_task sub-agents, polls video jobs with 600-second timeouts, retries 503 errors with exponential backoff, and concatenates scenes with ffmpeg including optional crossfade. - Use Case: A user uploads an anime idol character image and asks for a concert video; the Skill produces a four-scene 9:16 vertical video (stage entrance, singing, dancing, bowing) with consistent character appearance. ## Quick Start Here is my character image; create a four-scene anime idol concert video in 9:16 vertical format and show me the script for approval first.
