arch-dog/video-prompt-engineer
Data science & MLThis page was translated automatically. Technical details may differ from the original. View the original
AI short drama storyboard Agent Skill for Seedance: Converts reviewed scripts and assets into executable video prompt sheets.
Overview
Convert finalized scripts or script segments into 10–15 second text storyboards and video prompt phrases that can be directly accessed in the Seedance generation process. A portable Agent Skill for turning finalized scripts into executable, Seedance-oriented text storyboards and video prompts. Video Prompt Engineer is a skill that follows the Agent Skills directory structure. It is designed for Chinese vertical animation short dramas and Seedance workflows, responsible for final script reading, audio verification, fixed asset extraction, 10–15-second storyboard planning, internal shot design, emotion externalization, action causality checking, and final video prompt delivery. It generates **. Storyboards, character assets, video generation, finished video editing, and reworking of generated results are still handled by the corresponding tools or production stages. - People who use Codex, Claude Code, or other agents to produce AI short plays. - Directors, screenwriters, and prompt engineers who need to convert literary scripts into executable shot descriptions. - Production teams that need to control dialogue integrity, shot length, character consistency, and scene continuity. - AI content teams aiming to distill storyboard production workflows into reusable skills. 1. : Identify scenes, people, actions, dialogue, narration, inner OS, and ambiguity. 2. : Count all spoken content and estimate duration, checking for omissions and duplicates. 3. : Pre-fix repeated, over 15-second or non-continuous characters, scenes, and standalone props. 4. : Control a storyboard to generate for 10–15 seconds each, containing several shots internally. 5. : Check subject position, movement direction, physical connections, visible results, and end states. 6. : Identify boundaries based on subject changes, reactions, object states, emotional transitions, and spoken relationships. 7. : Convert abstract narratives into physical phenomena that the camera can record. 8. : Design visible or audible connections for adjacent storyboards that must be disassembled. 9. : Output item configuration, fixed assets, connection plans, and complete storyboards. 10. : Check audio, duration, continuity, causality, asset discipline, and independence of prompts. - Line, narration, and internal OS verifiability. - Controllability of fixed asset quantity and reference image costs. - Visual focus clarity for each prompt. - Continuity when generating across segments in the same scene. - Spatial direction and causal integrity of complex actions. - Executability of abstract emotions in the visuals. - Consistency in asset naming and prompt structure. These results are process control objectives and do not guarantee the video model will succeed in generating it on the first try.
Overview (machine translated)
Convert finalized scripts or script snippets into ones that can directly enter the Seedance generation process 10–15 second text storyboards and video prompt phrases.
A portable Agent Skill for turning finalized scripts into executable, Seedance-oriented text storyboards and video prompts.
Project Positioning
Video Prompt Engineer is a skill that follows the Agent Skills directory structure. It is designed for Chinese Vertical animation short dramas and the Seedance workflow, responsible for reading finalized scripts, audio verification, fixed asset extraction, 10–15 second storyboard planning, internal shot design, emotional externalization, action causality checks, and final video prompt delivery.
It generates text storyboards and video prompt sheets. Storyboards, character assets, video generation, final clips, and result rework are still handled by the corresponding tools or production stages.
Who is it for?
- People who create AI short dramas using Codex, Claude Code, or other agents.
- Directors, screenwriters, and prompt engineers who need to convert literary scripts into executable camera descriptions.
- Production teams that need to control dialogue integrity, screen time, character consistency, and scene continuity.
- Aiming to distill the storyboard production process into an AI content team capable of reusing skills.
What problems does it solve?
| Frequently Asked Questions | How to handle Skill |
|---|---|
| Lines, narration, or the inner OS are missing when disarming the camera | First, create an audio ledger, then allocate shots and duration |
| Descriptions like “sadness, oppression, hesitation” cannot be directly generated | Converted into eye contact, movements, body distance, and object state |
| Duplicate characters, scenes, and props are not fixed in advance | Complete a fixed asset extraction before storyboarding |
| Ordinary barrels, ropes, and furniture are excessively dismantled into assets | Fixed scenes absorb ordinary environmental elements, while short-term elements are directly written into prompts |
| The same scene is cut into many short clips, making it easy for characters and spaces to drift | A storyboard is prioritized to be kept at 10–15 seconds to generate units |
| Adjacent generated clips lack connections | Plan actions, line of sight, sound, objects, light color, or occlusion anchors |
| Cramming multiple visual focal points into a time period | First identify the beat boundary, then assign time periods |
| Prefix prompts mixed with weak constraints like “game ad” or “beautiful visuals” | Only the basic visual style and script-confirmed time and basic lighting and shadows are retained |
| Prompts rely on “last shot” or “that thing just now” | Each prompt word independently describes necessary information |
Core Competencies
- Script Reading: Identify scenes, characters, actions, dialogue, narration, inner OS, and ambiguity.
- Audio Ledger: Count all vocal content and estimate duration, checking for omissions and duplicates.
- Fixed Asset Extraction: Pre-fixed characters, scenes, and standalone items that appear repeatedly, across 15 seconds, or non-consecutively.
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add arch-dog/video-prompt-engineer