mcpField
3
3d-animation-short-generator

3d-animation-short-generator

github.com/sjh00Creative & media4054 installsUpdated 9w ago
Repository

About 3d-animation-short-generator

A creative skill for building shot lists and per-shot H3 prompts for 3D animated shorts. It anchors continuity in character identity and scene landmarks, then specifies action, space, camera, and audio over time.

Key features

  1. 01Creates shot lists and per-shot H3 prompts
  2. 02Maintains continuity with character and scene anchors
  3. 03Plans action, expression, camera, space, and audio by second
  4. 04Checks hook density and shot transitions
  5. 05Provides a style prefix and final prompt structure

Use cases

  1. 01Planning animated shots no longer than 15 seconds
  2. 02Writing prompts that preserve characters and settings across shots
  3. 03Turning a shot list into generation prompts for each shot
  4. 04Checking whether a shot timeline is continuous and actionable

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/sjh00/minimax-h3-storyboard-prompt-skills --skill 3d-animation-short-generator

After installing, type/3d-animation-short-generatorin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

meta.yaml
references/
fallback-policy.md
model-selection.md
qc-checklist.md
shot-table-spec.md
storyboard-guidelines.md

Details

From the project README

3D 动画短片 · 提示词写法

交付：镜头表（可中文）+ 逐镜 H3 prompt。 单镜 ≤15s；更长拆镜。细节字段见 references/。

默认风格前缀（每镜可挂）

Pixar-inspired 3D cartoon, C4D+Octane look, stylized proportions,
warm SSS skin, clear hair silhouette, elastic squash-and-stretch performance,
strong design language — NOT photoreal, NOT flat anime, NOT plastic toy

写法链路（均为写 prompt 服务）

规格(画幅/总时长/台词语) → 一句话 What-if
→ 角色身份锁 / 场景地标锁（可先写卡规格再写进 prompt）
→ 六列镜头表 → 自检 → 文本分镜节
→ 逐镜 H3 终稿（剥离 [char:][scene:][hook:] 等分镜标记）

六列镜头表（草稿）

镜号时长连续性衔接参考锚点Hook每秒指令音频对白

参考锚点： 地标+画面位 · 人物机位位置 · 退场状态 · 光位 · 角色/场景名
每秒五要素： 动作表情 / 运镜 / 空间 / 音频 / 交接下一秒
Hook 例： reveal / reversal / callback / visual-joke / chase / tender…

自检后再写终稿

Hook 密度 · 单镜≤15s · 重要角色≤3 · 地标光可继承 · 每秒无空隙 · 连续链无矛盾

文本分镜节 → H3

每镜一节保留：承接/交接、空间锚、每秒四象限。
生成用 prompt 删除 [char:…][scene:…][shot:…][hook:…]，改为自然语言身份与场景锁。

[规格] S03 · 6s · {画幅} · H3 · {768|2K 意图}

[风格前缀] …

[身份与场景锁] 角色… 地标 door-frame 右侧1/3 … 光…

[时间轴]
0-1s: pose/expression · camera · spatial · audio
1-2s: …
… 结束状态锁定 → 供下一镜继承

[声音] 对白原文进 <d> 或保留中文；SFX；BGM 意图或 N/A
[负面] storyboard lines, labels, watermark, photoreal, stiff anatomy

可再压成 integrated_multimodal_description 等，见 h3-prompt-writing。

失败只改文

强化锚点句 → 缩≤6s 拆邻镜 → 减道具/降 hook。不引入其他模型路径。

自检

  • 跨镜身份与地标一致
  • 每秒可执行
  • 终稿无分镜标签
  • 弹性表演词写到动作上

Comments

Markdown supported