Plotloom Shot Prompts

Plotloom Shot Prompts

When to Use

Use after series.md and characters.md exist, with optional episodes/epXXX/episode-card.md, to create continuous video prompts for clip generation.

Inputs

• series.md
• characters.md
• Optional episodes/epXXX/episode-card.md
• Existing character grids such as assets/cast/<slug>/character-grid.png
• Optional previous clip last frame or previous video reference.

Outputs

• episodes/epXXX/video-prompts.md
• episodes/epXXX/video-prompts-en.md only when the adapter/model needs English.

Read These Resources When...

• Read references/visual-continuity.md when character grids, reference sheets, previous clips, or inter-clip handoffs matter.
• Read references/seedance-dreamina-prompt-style.md when targeting Seedance 2.0 / Dreamina / 即梦 style video models.
• Use templates/video-prompts.md and templates/video-prompts-en.md for output shape.

Workflow

1. Read series, characters, optional episode card, available visual references, and prior clip handoff notes. 2. Write continuous narrative prompt tasks, not mechanical shot lists. 3. For Seedance / Dreamina-style models, one clip prompt equals one coherent 8-20s cinematic moment with a single primary timeline. 4. Use the default task shape: clip-01, source beat, duration hint, references and purpose, timeline beats, continuity rules, camera motion, dialogue/audio window, negative constraints, ending frame / handoff point, and rerun notes. 5. If the model/adapter needs English, create video-prompts-en.md as an adapter copy that preserves the source intent and does not add new plot. 6. Stop before calling a model.

Prompt Rules

• Seedance prompt ≠ shot list. It is a director brief translated into a continuous narrative timeline.
• Use visible action over abstract emotion.
• Mention @图片N, @视频N, @音频N only in adapter-targeted copies where that platform syntax is needed.
• Preserve character-grid identity: face, age, outfit, body language, style, and signature props.
• Include entrance/exit/occlusion when continuity could break.
• Avoid conflicting camera instructions. Use one dominant camera idea per clip.
• Dialogue must be serial and sparse; avoid two people speaking over each other in a short clip.
• Reserve head/tail safety: do not place the critical reveal only in the first or last 0.5s.
• Prefer rerun notes over silently changing intent after failed candidates.

Stop Conditions

Stop when video prompts are ready. Do not submit jobs, poll queues, copy selected files, or stitch videos.

Next Skill Handoff

Use plotloom-video-adapter for fake or real model submission.

Failure Modes

• Missing series context: hand off to plotloom-series-brainstorming.
• Missing episode intent: create an episode card or ask for direction.
• Model requires English: produce video-prompts-en.md from the same intent.
• User asks for per-shot model calls: explain that Plotloom uses continuous prompt tasks, not one generation per micro-shot.

Seedance / Dreamina Prompt Style Reference

Sources Borrowed

Patterns are adapted from local readings of Seedance prompt repos, including seedance-prompt-skill, Seedance2-Storyboard-Generator, seedance_prompt, shotcine, and autoflow.

Core Rule

Seedance prompt ≠ shot list
Seedance prompt = director brief translated into one continuous narrative timeline

Each generation task should be one continuous cinematic moment. Dreamina CLI currently supports 4-15 seconds for Seedance 2.0 family commands, so clamp Dreamina-bound tasks to --duration=15 when the creative target says 15-20s. A prompt may include multiple camera beats, but it must have one primary timeline.

Useful Prompt Formula

[规格总纲：竖屏/横屏，时长，风格，情绪]
素材引用：@图片1 = 主角外观参考；@图片2 = 场景参考；@视频1 = 上一段视频用于衔接
0-3秒：建立场景 / 钩子
3-6秒：主体入场 / 冲突出现
6-9秒：动作推进 / 情绪升级
9-12秒：高潮 / 反转
12-15秒：尾帧 / 悬念衔接
声音/台词：串行、简短、可拍
禁止：水印、logo、伪字幕、画面内乱码文字、身份漂移

Reference Syntax

When the target platform supports it, use:

• @图片1 to @图片9
• @视频1 to @视频3
• @音频1 to @音频3

Keep a visible material mapping near the prompt. Do not exceed the provider-specific material count. Some workflows use total multimodal files ≤12; API execution paths may be stricter (e.g. ≤9). Treat limits as adapter-specific constraints.

Prompt Checks

• Is there exactly one primary timeline?
• Does every reference have a purpose?
• Are characters introduced with stable identity anchors?
• Are entrances, exits, and occlusions described where needed?
• Is dialogue serial rather than overlapping?
• Is camera motion attached to a stable space anchor?
• Is there an ending frame / handoff point?
• Does the prompt avoid key information only in the first/last 0.5s?

Do Not

• Submit one model job per micro-shot.
• Convert a storyboard table directly into a video prompt.
• Overload a 15s prompt with more than one major reversal.
• Hide retry rationale; write rerun notes.

Visual Continuity Reference

Character Grid Usage

Use character-grid.png as the identity anchor for face, age, outfit, hairstyle, body type, palette, and rendering style. Refer to the image by path and state its purpose.

Stable vs Mutable Traits

Stable traits should not change without user approval:

• face shape, age, hair, proportions;
• wardrobe lock and signature prop;
• palette and render style;
• relationship to other recurring characters.

Mutable traits may change by episode only if the story requires it:

• expression;
• pose;
• temporary dirt/damage/wetness;
• location lighting.

Reference Sheet / Expression Grid Inputs

GPT Image 2 prompt libraries often use official character sheets, expression grids, and cast boards. When those exist, cite the exact file and describe the intended use: identity, expression, costume, prop, or scene style.

Inter-Clip Handoff

For clip continuity, capture:

• starting state from previous clip;
• ending frame / tail pose;
• prop position;
• whether the character enters, exits, turns, or is occluded;
• whether first/last-frame images should be used by the adapter.

Movement and Occlusion

Describe entrance, exit, partial occlusion, turns, and handoffs when continuity could break. Mention when the character remains visible throughout the clip.

Camera Instructions

Use one dominant camera idea per clip. Avoid mixing contradictory instructions such as locked-off + fast tracking + drone orbit in the same 15-second task.

Common Model Failures

• Face drift across time.
• Age drift.
• Outfit or prop mutation.
• Temporal flicker.
• Impossible hand motion.
• Background or text artifacts.
• Ending frame that does not match the next clip handoff.

Rerun Notes

After failed candidates, write whether to retry the same prompt or revise the prompt. Revision notes should name the failure: character drift, visual continuity break, weak hook, artifact, pacing, unclear ending frame, or dialogue overload.

Episode {{episode_number}} Video Prompts (English Adapter Copy)

Preserve the source intent from video-prompts.md. Do not add new plot during translation.

This file is still a Plotloom artifact, not the exact Dreamina CLI input. The adapter must extract one clip block into a single --prompt string plus CLI flags.

clip-01

• Source clip id: clip-01
• Target adapter: Dreamina CLI
• Recommended command: multimodal2video if reference images exist; otherwise text2video
• Duration seconds: 15
• Ratio: 9:16
• Model version: seedance2.0fast_vip
• Reference images and purpose:
• assets/cast/{{character_slug}}/character-grid.png — identity and outfit continuity; pass as --image only when using multimodal2video
• Prompt string for --prompt:

  [Write one continuous cinematic prompt string here. Do not include Markdown bullets. Include visible action, character identity, camera motion, dialogue window if any, ending frame, and negative constraints such as no subtitles, no watermark, no logo.]

• Continuity rules:
• Camera motion:
• Dialogue / audio window:
• Ending frame / handoff point:
• Adapter-specific notes:
• If using text2video, rewrite reference-image information into the prompt; the CLI will not read local images.
• If using multimodal2video, pass real local paths with repeated --image flags.

clip-02

• Source clip id: clip-02
• Target adapter: Dreamina CLI
• Recommended command: multimodal2video if reference images exist; otherwise text2video
• Duration seconds: 15
• Ratio: 9:16
• Model version: seedance2.0fast_vip
• Reference images and purpose:
• Prompt string for --prompt:

  [One continuous cinematic prompt string. No Markdown bullets in the CLI input.]

• Continuity rules:
• Camera motion:
• Dialogue / audio window:
• Ending frame / handoff point:
• Adapter-specific notes:

Episode {{episode_number}} Video Prompts

clip-01

• Source beat / episode-card anchor:
• Duration hint: 8-20s
• Adapter target: Seedance / Dreamina / fake / other
• Reference images and purpose:
• assets/cast/{{character_slug}}/character-grid.png — identity, outfit, expression continuity
• Optional previous reference:
• previous clip / last frame:
• Intent:
• Single cinematic moment:
• Timeline beats:
• 0-3s:
• 3-6s:
• 6-9s:
• 9-12s:
• 12-15s:
• Character continuity rules:
• Environment continuity:
• Camera motion:
• Dialogue / audio window:
• Negative constraints / avoid:
• Ending frame / handoff point:
• Rerun notes:

clip-02

• Source beat / episode-card anchor:
• Duration hint: 8-20s
• Adapter target:
• Reference images and purpose:
• Optional previous reference:
• Intent:
• Single cinematic moment:
• Timeline beats:
• Character continuity rules:
• Environment continuity:
• Camera motion:
• Dialogue / audio window:
• Negative constraints / avoid:
• Ending frame / handoff point:
• Rerun notes: