Flova Costume Romance Minidrama

Costume Romance Minidrama

Mission

用于古风甜宠短剧：围绕古装甜宠类型的反差人设、身份悬念、高密度糖点、站位连续和情绪节拍，制作 1:30-2:00 左右的短剧视频。

This is a genre template on top of scripted short-film production. It should coordinate with flova-scripted-short-production or canvas-long-video-production for long multi-clip execution.

JarvisHub Execution Model

• 主 Agent 负责编排：读取画布事实、整理任务 brief、按阶段同步 TodoWrite 进度，并把已确认的文本成果写入画布；媒体生成、等待、拼接和评审交给具备相应能力的执行 agent 或当前可用工具。TodoWrite 不是画布写入前置条件。
• 图像、视频和拼接交给具备媒体能力的执行 agent：brief 给出稳定输出身份、用途、真实参考 URL 和关键约束；需要下游引用时必须等待真实 imageUrl / videoUrl，不能把提交态当完成态。
• 多模态验收交给 critic sub-agent：只读取真实媒体并评审，不生成、不补素材。
• 当前已知 canvas 工具集没有通用音频生成、音频驱动口型、末帧抽取或视频直改工具；旁白、BGM、字幕、末帧承接和精准 lip sync 只能作为后期合成计划或 blocked 项，除非本轮工具列表明确暴露对应能力。

When To Use

Use this skill for:

• 古风甜宠短剧,
• costume romance minidrama,
• palace/jianghu/school/courtyard romance,
• meet-cute, misunderstanding resolution, intimacy escalation,
• identity-secret romance hooks,
• short video romance drama.

Do not use it for serious historical drama, action wuxia, or non-romance costume videos unless the user wants sweet-romance genre grammar.

Canvas-Native Boundary

Follow the JarvisHub Execution Model. Do not assume fixed models, exact video length, or reference_video support.

Use exposed role-image, scene-image, per-shot video, text overlay, and assembly capabilities through the proper sub-agent. Treat BGM/VO as a plan or bound user asset unless this turn exposes a real audio tool.

Workflow

1. Confirm script source: user-provided or AI-generated.
2. Review/confirm script.
3. Confirm character references or generate character designs.
4. Confirm aspect ratio, BGM, and subtitle policy.
5. Build storyboard.
6. Generate/confirm scene images.
7. Generate videos shot by shot.
8. Create BGM/VO/subtitle plan if requested; generate or attach real assets only when supported by this turn's tools.
9. Assemble and export.

Pause after script, character images, production preferences, storyboard, scenes, video batch, and final cut.

Genre Requirements

A strong short costume romance needs:

• contrast personas: each lead has public/private contrast.
• identity hook: each lead has a secret or hidden status.
• dense sweet beats: touch, gaze, near-distance, teasing line, protective gesture, accidental closeness.
• emotional landing every 10-15s.
• no empty transitional padding.

For AI-generated scripts:

• target 10-14 shots,
• 1:30-2:00 total,
• scene + dialogue + action format,
• each scene has an emotional landing,
• at least one explicit identity hint or reversal,
• each lead gets at least one persona-reveal moment.

Character Policy

Use user-provided references first.

Do not hard-code one beauty standard. Character design should follow:

• user reference images,
• user-described appearance,
• genre expectations only as a secondary guide.

Each lead should have:

• face/hair/wardrobe,
• costume palette,
• accessories,
• emotional bearing,
• public persona,
• private contrast,
• secret/identity clue.

If references exist, do not alter face, hair, costume, or color without user approval.

Scene Policy

Core scenes may include:

• palace courtyard,
• garden corridor,
• bamboo path,
• study room,
• riverside bridge,
• lantern street,
• inn room,
• academy courtyard,
• mountain pavilion.

Scene images should be realistic cinematic costume-drama spaces unless the user requests stylization:

• natural light,
• candles/lanterns,
• mist/morning/evening atmosphere,
• real architectural/material texture,
• no game-CG look by default.

Storyboard Rules

10-14 shots by default.

Each shot:

• 8-15s if the tool supports it,
• contains at least one action beat or line,
• includes start and end positions for both leads,
• preserves continuity from previous shot,
• specifies scene, action, dialogue, camera, emotion.

Every shot must state:

• starting position/facing/hand state,
• ending position/facing/hand state.

From Shot 2 onward, opening state should reference the previous shot's ending state unless there is a clear scene/time cut.

Sweet-beat camera language:

• close-up for eye contact,
• slow push for emotional reveal,
• hands/fabric/hair ornament detail for tension,
• two-shot for distance closing,
• over-shoulder for hidden gaze,
• foreground veil/flowers/curtain for layered romance.

Avoid empty atmosphere shots unless they include micro-action, environmental motion, or emotional setup.

Prompt Rules

Use Chinese prompts by default for this genre unless this turn's tool list performs better with English.

Video prompt structure:

1. reference binding for characters and scene,
2. continuity opening state,
3. camera move,
4. lead actions and micro-expressions,
5. dialogue with {...} only if video model should produce speech,
6. scene/environment motion,
7. ending continuity state,
8. negative constraints.

If dialogue/VO/subtitles are handled in post, do not force the video model to generate readable subtitles.

Negative constraints:

无随机文字，无水印，无多余人物，无现代物件，无服装错乱，无脸部漂移，无字幕（若字幕后期添加）

Audio And Subtitles

BGM default if requested:

• light costume-romance instrumental,
• gentle strings/pipa/dizi,
• restrained sweetness,
• no vocal lyrics by default.

Subtitle policy:

• if user wants subtitles, add in assembly/timeline, not in video generation prompts.
• align subtitles to dialogue timing.

Dialogue clarity is more important than BGM.

Quality Review

Check:

• romance genre beats are dense enough,
• both leads remain visually consistent,
• references are preserved,
• no modern artifacts,
• start/end positions maintain continuity,
• identity hook is visible,
• sweet beats are not empty posing,
• subtitle policy followed,
• real media URLs exist.

If a shot lacks emotional landing, rewrite the beat before regenerating.

Output Contract

Return:

• script_source,
• genre_plan,
• characters,
• scene_list,
• storyboard,
• prompts,
• media,
• audio_subtitle_plan,
• assembly_plan,
• review.

Do not present a script as a finished minidrama.