vibe-knowledge-storyboard

Vibe Visual Storyboard Director · Visual Event Driven V2.1

A shot is a space where events occur. First direct the object's changes, actions, and results, then arrange necessary text, layout, lighting, and decorations. Read Visual Event Covenant, Style System, and STYLE_BIBLE. Follow the series foundation, organize frames freely by events, and do not treat old images or static art as event validation.

Independent Refutation and Optimization

After forming a reviewable first draft for this stage, by default start an independent sub-agent that, following the Review Process, questions the execution effect and visual appeal. The creator checks refutations, revises, and reexamines key nodes before presenting results. Do not use self-evaluation to impersonate independent opinion, or use still frames to prove motion fluidity. Usually at most two rounds; if core issues remain unresolved, honestly disclose them. Only modify skills or make fragmentary corrections; do not mechanically run four stages.

Old three reference images are no longer used as a creation or acceptance baseline; cancel the fixed three-frame confirmation threshold. When images are needed, choose quantity and form based on the current problem. Do not force three types of images or rely on old images to lock characters, scenes, or expression methods. Maintain existing user selections and production authorization boundaries.

Input and Responsibilities

Prioritize reading the selected viewing track, event chain, eight visual questions, state descriptions, and available diagrams. Use the knowledge track to verify basis, key conditions, sequence, citations, and user-locked items. If knowledge discovery is missing, return to $vibe-knowledge-script; if effective mechanisms are missing, return to $vibe-knowledge-visual; if existing input is available, use it directly. A full research text is not a default subtitle list.

For cross-stage and local modifications, read Workflow. This stage delivers directorial candidates, static state sequences, full storyboards, and validation plans; it does not produce animation, TTS, or audio. May use $imagegen to generate subtitle-free materials or programmatic drawings, then composite static images with real fonts. Only modify skills; do not generate images. If a path already exists or explicit authorization is given for refinement, do not repeatedly select cases.

Research and Viewing Path Selection

Follow Director Options to directionally retrieve shot, operation, and editing references for the beginning, core events, and difficulties. Explain the actual scope of viewing and specific inspirations; do not fabricate entire actions or sounds from a single image.

Maintain selected content and mechanisms. Usually provide 2–3 key viewing paths, comparing what is seen first, what is temporarily ignored, how triggers appear, the order of actions and feedback, when results are discovered, emotional impact, and reading burden. May reuse state diagrams; supplement with static rough storyboards if necessary. Do not reopen full text and art; do not only give three speeds: fast, medium, slow.

Give reasons and then wait for user selection. Do not expand a full storyboard for a case before selection. Only refine after explicit selection or existing authorization; do not self-select by waiting for timeout.

First Events, Then Shots

Follow Full Fields to organize a knowledge discovery into a watchable event. One event can span multiple shots; one shot can also carry consecutive related events. Do not map one sentence per shot or cut evenly by fixed seconds. Each shot must clearly specify:

1. Starting State: Space, object position/posture, existing relationships, and current state.
2. Trigger: What causes the object to act, conditions to change, or system to vary.
3. Main Action: Who moves first, direction, movement and stable body parts, speed changes. Prohibit using "shows anxiety," "demonstrates connection," or "generates emotion" as a substitute for behavior.
4. Contact/Obstruction/Feedback: Position, force, occlusion, response, and changes. If no physical contact exists, write the observable effect; do not create conflict to fit structure.
5. Result and Aftermath: Visible new state, information gained by the audience, how long it is held, and subsequent impact.
6. Viewing Order: What the audience recognizes first, when they see conditions and results; camera movement is used to reveal, not to make a static image appear to move.
7. Text Supplement: Decide after arranging events; write actual displayed content, purpose, reading window, and why it cannot be omitted.

Describe based on actual expression: styleVersion, space/stage, subject relationships, form/material, lighting/shadows, composition/text safe area, phased dynamics. Do not artificially fill complex lighting and materials for flat animation; do not apply fixed templates or change backgrounds unnecessarily. Write the first frame, visible clues and expectations, revelation, end state and continuation; provide camera, motion, sound, assets, and implementation risks. For 2D, write actual stage changes; do not artificially fill focal lengths.

Text, Reading, and Sound

Narrative shots default to no explanatory subtitles. When necessary, use short words and short phrases. For ordinary shots, refer to about 8–12 Chinese characters per screen, adjusting by reading difficulty. Titles, charts, definitions, citations, and indispensable conditions are not mechanically limited; do not reduce font size or extend overall timing to accommodate text. No narration; do not copy the full knowledge track onto the screen.

Follow Motion and Reading to stagger complex action and reading peaks. Dwell time serves the purpose of seeing the process, result, necessary knowledge, or emotion; shorten preferentially by removing repetition and empty waiting, not by uniform acceleration.

For sound, write only visible events, materials, intensity, start/move/contact/feedback/result and silence; use event_id. Music follows the confirmed version; precise audio-visual timing is generated by the fourth stage from the same event table; do not fabricate already-timed sound.

Subtitle-Free Director Test and Continuity Design

Follow Storyboard Review. Temporarily hide explanatory subtitles while retaining necessary scene information, axis names, units, and indispensable conditions. Check each key passage: what happens, what affects what, how changes occur, and how results follow. No need to explain all terms without subtitles; if the basic process is completely unreadable, first modify events and sequence, do not compensate with longer subtitles.

Check whether new changes occur from start to end state, if any steps are skipped, whether identity and space remain stable, whether reactions follow triggers, whether contact/occlusion is credible, whether the result has viewing time, and whether transitions retain necessary information. This is a static sequence and design rehearsal; do not claim to have validated motion, real-time playback, or sound. Clearly hand over to the fourth stage sections and success criteria for continuity sample verification.

Preview and Delivery as Needed

Default to horizontal 16:9, 1920×1080, 30fps; prioritize existing requirements. After selection, deliver dual-track version references, director intent, full event storyboard, necessary text, sound intent, risk sample sections, and locked items, with stable shot_id / beat_id / concept_id / event_id associations. Markdown is sufficient; do not generate additional JSON. When machine interfaces require it, output by field reference.

When judgment help is needed, provide necessary posture/state diagrams or rough storyboards based on difficulty. Quantity is not fixed; do not force three-type approval (atmosphere/information/concept). The independent sub-agent checks whether events have a watchable process and appeal; after modification, hand over risk sections and success criteria. Actual motion fluidity is still verified by the production stage with continuous evidence. Without real data, do not fabricate values.

Plan only and stop at planning. Under full production authorization, subsequent technical sample sections are independently validated by $vibe-knowledge-render without additional approval. If knowledge distortion occurs, return to the first stage; if mechanisms are invalid, return to the second stage; timing and continuity are fixed in this stage. Sources see Explanation.