MyReels Storyboard Design Tool

Integration with myreels-api

CODEBLOCK0

Important: This skill handles storyboard design. After completion, guide the user to use myreels-api skill for actual image and video generation.

Complete Workflow

CODEBLOCK1

Step 1: Collect User Requirements

Collect the following information:

Question	Options
Story content	Plot outline or full script
Video style

cinematic / anime / 3d cartoon / realistic |
| Duration | e.g., 60 seconds, 2 minutes |
| Shot count | e.g., 10-15 shots |
| Aspect ratio | 16:9 / 9:16 / 1:1 (vertical = 9:16) |
| Has dialogue | Yes / No |

After collecting inputs:

• - Use the current OpenClaw session's default LLM to understand the story semantically
• Infer core roles, recurring relationships, and phase-by-phase visual beats from the user's actual plot, not only from surface keywords
• Do not ask the user to configure an extra LLM just to use this skill
• Do not rely on a bundled local storyboard script for core planning; OpenClaw should draft the CSVs directly
• The old helper scripts/generate_storyboard.py is deprecated and should not be referenced in this workflow

Step 2: Character Design (Important!)

Before shot design, establish character consistency using a three-view character sheet (三视图).

Default rule:

• - Treat character design as a single-image turnaround sheet, not as a poster
• The first-pass deliverable should show the same character's front view, side view, and back view on one canvas
• Keep outfit, body shape, hairstyle, and accessories identical across all three views
• Use full-body neutral standing pose and a plain background unless the user asks otherwise
• Recommend nano-banana2 first for generating the base three-view character sheet
• Prefer a character-first workflow: approve the character sheet and tags first, then generate storyboard shots with those tags injected

Character Design Fields

Field	Description
INLINECODE8	Character name
INLINECODE9

Role in story, e.g. protagonist / opponent / companion | | relation_to_main | Relationship to the main character, e.g. self / conflict / ally / romance | | appearance | Physical description (hair, eyes, face, build) | | outfit | Clothing and accessories | | personality | Key personality traits | | reference_model | Base model for the first turnaround sheet, default nano-banana2 | | single_image_turnaround_prompt | Main English prompt for one-sheet front/side/back generation | | front_view | English description for front view | | side_view | English description for side view | | back_view | English description for back view | | character_tags | Consistency tags for AI prompts (e.g., "short hair, left face mole, red earrings") | | negative_prompt | Terms to suppress poster-like, inconsistent, or cropped outputs |

Three-View Diagram Output

Generate character descriptions for AI image generation. The output should be structured for a single image containing all three views:

CODEBLOCK2

Prompt rule:

• - The single_image_turnaround_prompt must front-load the layout requirement: one image, same character, front view, side view, back view, INLINECODE28
• Do not output three unrelated portrait prompts as the default character design result
• Only add expression sheets, weapons, or alternate outfits after the base three-view sheet is approved

Character Consistency

For each character, build a character tag library that will be reused in all shot prompts:

CODEBLOCK3

Character-First Mode

Recommended operating mode:

• - Generate characters.csv first
• Generate relationships.csv for recurring character pairs
• Lock the main character's single_image_turnaround_prompt, negative_prompt, and INLINECODE33
• Only define extra character rows when the story actually implies them; do not invent an opponent, love_interest, or companion just because the genre suggests one
• Reuse the approved character_tags in every INLINECODE38
• Treat the first approved character row as the anchor for downstream storyboard drafts
• In multi-character stories, inject only the characters relevant to each shot phase instead of dumping everyone into every prompt
• Build relationships.csv as a pairwise matrix for all recurring character pairs, not only protagonist-centered pairs
• Reuse relationship_type and prompt_hint when designing confrontation, ally, or romance shots
• Also use visual_relation, emotional_relation, and blocking_pattern to keep staging and mood consistent across repeated pair shots
• Let relationship_type bias the default shot language: conflict leans toward low-angle / confrontational framing, ally toward cooperative full shots, romance toward closer eye-level coverage
• Let theme bias the baseline shot rhythm too: action leans kinetic, romance leans intimate, thriller leans uneasy, sci-fi leans scale/discovery
• In scenes with 3+ recurring characters, choose the active pair/group per shot based on the current beat; do not force every established relationship into every shot

Default OpenClaw behavior:

• - The skill should use the active OpenClaw default LLM to draft characters.csv, relationships.csv, and INLINECODE49
• The CSVs should reflect narrative understanding, not just mechanical keyword expansion
• Keep CSV column names stable in English, but make user-facing cell values follow the user's language whenever possible
• Keep AI-facing prompt fields in English for cross-model portability: single_image_turnaround_prompt, negative_prompt, front_view_prompt, side_view_prompt, back_view_prompt, character_tags, prompt_hint, and INLINECODE57
• Keep IDs and machine-oriented codes stable: character_id, relationship_id, shot_id, INLINECODE61
• INLINECODE62 should preserve concrete story specifics from the user's outline in English, including names, props, locations, and beat-specific actions
• INLINECODE63 and action should vary shot by shot; avoid repetitive generic phase labels when a more specific beat description is available
• For mixed-language input, keep description in the dominant user language; if no dominant language is clear, follow the user's most recent instruction language
• If a downstream automation flow later needs normalized English enum values, create a secondary English export instead of replacing the user-facing review CSV

Step 3: Generate Storyboard

Shot Structure

Each shot includes these fields:

Field	Description
INLINECODE66	Scene + Shot number (e.g., S01-01)
INLINECODE67

Shot size |
| camera_angle | Camera angle |
| movement | Camera movement |
| duration | Estimated seconds (2-5s typical) |
| description | User language description - brief summary in user's input language (Chinese/Japanese/English). NOT used for AI generation. |
| visual_prompt | English prompt for AI image generation |
| action | Character action description |
| dialogue | Dialogue/voiceover (optional) |
| emotion | Target audience emotion |
| sound_fx | Sound effects / music cues (optional) |
| notes | Additional notes |
| difficulty | 🟢 simple / 🟡 medium / 🔴 complex |
| image_url | Generated image URL from downstream tool (optional) |
| video_url | Generated video URL from downstream tool (optional) |
| status | Workflow status in the user's language for review CSVs; normalize in the English export if needed |

Shot Type Reference

Type	Chinese	Use Case
Extreme Close-up / ECU	大特写	Key detail, extreme emotion
Close-up / CU

特写 | Face, key object | | Medium Close-up / MCU | 中近景 | Dialogue, slight conflict | | Medium / MS | 中景 | Conversation, interaction | | Full Shot / FS | 全景 | Full body, relationship | | Wide / WS | 远景 | Environment, establishing |

Camera Angles & Psychology

Angle	Emotion Implication
Eye-level	Neutral, objective
Low-angle (仰视)

Power, authority, threat | | High-angle (俯视) | Vulnerability, submission | | Dutch | Unease, tension, chaos | | POV | Subjective, immersion | | Tilted | Unstable, danger |

Camera Movement

• - Fixed / Static
• Dolly In (推近) / Dolly Out (拉远)
• Pan (摇镜) / Tilt (倾斜)
• Follow / Tracking (跟拍)
• Orbit (环绕) / Crane (升降)
• Handheld (手持) / Breath-like (呼吸感)

Shot Difficulty Grading

Grade	Icon	Description
Simple	🟢	Fixed camera, single subject, no complex interaction
Medium

🟡 | Light camera movement, dual subject, simple effects | | Complex | 🔴 | Fast motion, multiple subjects, complex choreography, special lighting/particles |

Emotion Mapping

Each shot must specify target emotion:

Emotion	Key	Description
Hook	钩子	Attention grab
Tension

紧张 | Building suspense |
| Conflict | 冲突 | Confrontation |
| Sweet | 甜宠 | Romance, warmth |
| Twist | 反转 | Surprise |
| Climax | 高潮 | Peak emotional moment |
| Release | 释然 | Resolution |

Step 4: AI Prompt Engineering

Prompt Structure

CODEBLOCK4

Prompt design rule for this skill:

• - single_image_turnaround_prompt should use layout-first language for INLINECODE83
• INLINECODE84 should carry theme + phase cues so action / romance / thriller / sci-fi drafts do not collapse into the same generic shot language

Required Quality Tags

Always include:

• - INLINECODE87
• INLINECODE88 or INLINECODE89
• INLINECODE90 or INLINECODE91
• INLINECODE92

Control Tags

If needed, add:

• - no text / no watermark / INLINECODE95
• INLINECODE96 / two-shot / INLINECODE98

Character Consistency in Prompts

Always inject character tags from Step 2:

CODEBLOCK5

Scene Consistency

Use scene code +固化环境描述:

CODEBLOCK6

Step 5: Output Confirmation

Multi-Platform Output

Feishu: Use feishu_bitable for online editing
Telegram: Formatted text table or CSV attachment
Other: Default to CSV

Storyboard Table Fields

CODEBLOCK7

Note:

• - The CSV headers stay in English for schema stability
• Human-facing values in all three CSVs should follow the user's language
• The visual_prompt field is always in English for AI image generation
• If automation needs English enums such as shot_type, camera_angle, movement, emotion, difficulty, or status, write a separate English export copy instead of mutating the review CSV

Step 5A: Project Workspace Management

When this skill writes files, it should automatically create a project workspace in the user's current working directory or requested output directory.

Default root pattern:

CODEBLOCK8

Recommended structure:

CODEBLOCK9

Workspace rules:

• - Do not write user project artifacts back into the skill definition folder itself
• Create the project root before saving the first CSV, markdown note, image, or video
• Save editable drafts under INLINECODE106
• After user confirmation, freeze a copy under the matching approved/ folder instead of overwriting history
• Save character three-view turnaround images under INLINECODE108
• Move or copy user-confirmed character reference images into INLINECODE109
• Save storyboard shot images under INLINECODE110
• Save storyboard videos under INLINECODE111
• Move or copy user-confirmed storyboard media into the matching approved/ media folder
• For every myreels-api generation, save the exact request payload under INLINECODE113
• Save task lookup snapshots and task IDs under INLINECODE114
• Save response metadata under metadata/, including model name, mode, aspect ratio, seed if available, source image references, created time, and returned URLs
• Keep request / task / metadata filenames aligned with the output artifact name so reruns can reuse the same context
• Record major review decisions in INLINECODE116
• Summarize the final approved artifact set in INLINECODE117

Versioning rules:

• - Use file names like characters.v1.draft.csv, storyboard.v2.draft.csv, INLINECODE120
• Use *.approved.* for frozen user-confirmed files
• When revising an approved artifact, create a new draft version instead of mutating the approved snapshot
• Put normalized program-facing CSV exports in exports/ with names such as storyboard.program-en.csv and INLINECODE124
• Use aligned sidecar records such as character-C001-turnaround.v1.request.json, character-C001-turnaround.v1.task.json, INLINECODE127
• Use the same pattern for storyboard shots and videos, for example shot-S01-01.v2.request.json and INLINECODE129
• If the user already provided a target directory, follow it and only apply this structure inside that location

Re-run stability rules:

• - Before re-running a generation, read the latest matching request, task snapshot, and metadata files first
• Reuse the last approved prompt, model, aspect ratio, and source asset references unless the user explicitly asks to change them
• If the rerun changes any material parameter such as model, prompt, aspect ratio, seed, duration, or source image, write a new versioned request file instead of overwriting the old one
• In review-log.md, note why the rerun happened and what changed relative to the previous attempt

Step 6: Guide to myreels-api

Guidance Message (Image Generation)

CODEBLOCK10

Guidance Message (Video Generation)

CODEBLOCK11

Step 7: Quality Control Checklist

For each shot, verify:

• - [ ] Character matches design (outfit, hair, facial features)
• [ ] Lighting direction consistent with scene light source
• [ ] Motion trajectory physically possible (no clipping, floating)
• [ ] Shot type serves narrative rhythm
• [ ] Safe zone reserved (vertical: top/bottom 20% no key elements for subtitles)
• [ ] Emotion matches intended audience feeling

Output Files

• - characters.csv - Character design data
• INLINECODE132 - Relationship matrix for recurring characters
• INLINECODE133 - Complete storyboard
• INLINECODE134 - Character CSV starter template
• INLINECODE135 - Relationship CSV starter template
• INLINECODE136 - Storyboard CSV starter template
• Character reference images (via myreels-api)
• Storyboard image URLs (via myreels-api)
• Project review notes and approved snapshots under the project workspace
• Request payloads, task snapshots, and response metadata for reproducible reruns

Notes

• - visual_prompt must be in English with quality tags
• User can input in Chinese / Japanese / English
• In the three CSVs, user-facing values should follow the user's language; only AI-facing prompt fields stay in English
• If the workflow needs machine-friendly English enums, generate a separate *.program-en.csv export instead of changing the review CSV
• Each shot: 2-5 seconds recommended
• Always establish character design before shot breakdown
• Use difficulty grading to manage AI generation expectations
• In OpenClaw, this skill should primarily rely on the active session model for story understanding and CSV drafting
• Do not require users to provide extra LLM credentials just to use this skill in OpenClaw
• No local storyboard-generation script is required; draft the CSVs natively in OpenClaw
• Automatically create a project workspace so drafts, approvals, images, videos, and markdown notes remain organized
• Persist generation requests and metadata so image/video reruns can stay stable
• See references/ for detailed guides