Flova Wool Felt Animation

Wool Felt Animation

Mission

用于羊毛毡动画短片：输入文案或旁白后，按羊毛毡材质世界的统一风格生成角色/场景元素图、逐镜头视频、可选 BGM/旁白和最终合成。

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:

• wool felt animation shorts,
• narration-driven felt stories,
• warm handmade texture videos,
• 9:16 social animation,
• real-world scenes rebuilt entirely in wool felt material.

Do not use it for miniature model stories; use flova-miniature-world-short for diorama/tilt-shift miniature scale.

Core Visual Rule

The world is full-size but made of wool felt.

It is not:

• miniature,
• diorama,
• puppet stage,
• toy figurine,
• glossy CG,
• realistic human skin.

Everything should have:

• visible wool fibers,
• soft rounded edges,
• handmade uneven surface,
• muted warm palette,
• soft diffused light.

Workflow

1. Receive script/narration.
2. Confirm specs: 9:16 default, duration estimate, theme emotion, narration language.
3. Build storyboard.
4. Generate character/scene/prop element images.
5. Generate one video shot at a time in order.
6. Ask whether BGM and narration are needed after clips are confirmed.
7. Create optional audio plan; generate real audio only if this turn exposes a real audio tool.
8. Assemble.

Pause after spec, storyboard, element images, each shot video, optional audio plan or assets, and final output.

Storyboard Rules

Each shot:

• 4-15s,
• standard target 10-15s for narration alignment,
• contains 2-4 internal beats,
• maps directly to narration meaning,
• has scene, action, emotion, camera movement.

Do not let narration and image talk past each other. If narration says "dumplings are served", the shot should show that action or a direct emotional equivalent.

Character And Scene Rules

Characters:

• choose style grade by role:
  • realistic-cute for main characters,
  • moderately cute for secondary characters,
  • simplified/chibi only for crowds or symbolic groups.
• faces must be readable: eyes, nose, mouth visible.
• keep clothing and body style consistent.

Scenes/props/food/buildings:

• true-to-shape real-world proportions,
• fully covered by wool felt material,
• not cartoon-deformed unless user requests.

Prompt Rules

Every image/video prompt must anchor the style:

wool felt animation style, real-world scene and characters rendered entirely in wool felt material, visible fiber texture covering all surfaces, soft rounded forms with no sharp edges, smooth and fluid motion, natural movement, warm muted color palette, full-size felt world at realistic scale, NOT miniature, NOT diorama, NOT puppet, NOT toy figurine, readable facial features

Element image prompt:

• style anchor,
• subject,
• felt material detail,
• colors,
• three-view/clean reference if needed,
• white or simple background.

Video prompt:

• reference bindings,
• style anchor,
• scene atmosphere,
• ordered internal beat script with time ranges if supported,
• camera movement,
• SFX if relevant,
• no subtitles/no text overlay.

Avoid:

• sharp edges,
• realistic skin,
• glossy material,
• plastic doll look,
• miniature scale,
• puppet stage,
• text rendering.

Audio

BGM is optional and asked after clips are approved:

• gentle instrumental,
• warm,
• no vocals by default.

Narration is optional:

• generated only if user confirms,
• follows storyboard shot mapping,
• do not duplicate narration inside video prompts if it will be added separately.

SFX may be embedded in video prompts if the video model supports sound, using simple natural effects.

Quality Review

Check:

• all surfaces read as wool felt,
• world is full-size not miniature,
• main characters are readable and consistent,
• narration maps to shots,
• one shot was reviewed before moving to next in interactive mode,
• no unwanted text/subtitles/watermark,
• real media URLs exist.

If a clip becomes plastic/toy-like, regenerate with stronger wool fibers, handmade surface, matte fabric, and negative constraints.

Output Contract

Return:

• spec,
• storyboard,
• key_elements,
• shot_prompts,
• media,
• audio_plan,
• assembly_plan,
• review.

Do not call the animation complete before all shot clips are generated and chosen audio is either bound, generated by an exposed audio tool, or explicitly marked as post-production pending.