← Back to Skills Marketplace
zmy1006-sudo

FMT高清动态图生成器 v6.0

by mingyuan · GitHub ↗ · v6.0.0 · MIT-0
cross-platform ⚠ suspicious
50
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install fmt-highdef-dynamic
Description
FMT肠菌移植科普高清动态图生成器(插画版+4K静态图+1080P动态视频+批量模式+风格一致)。生成配套数字手绘插画静态PNG(4K超清),并转化为高清晰度1080P动态视频。风格活泼生动,适合科普教育。触发词:生成FMT动态图 / 生成高清动态 / 插画转高清视频。
README (SKILL.md)

FMT高清动态图生成器 v6.0

专为《探秘人体微宇宙 · FMT肠菌移植健康科普(插画版).docx》文档设计 v6.0(2026-05-12):新增风格参考图+批量生成+质量核查+降级策略


⚠️ 四大核心问题修复(必读)

问题1:图片清晰度不足 → 4K分辨率

  • 必须在 image_synthesize 时指定 resolution: "4K",输出4096px宽超清图像
  • 4K分辨率下文字边缘更锐利,AI渲染更精准

问题2:中文字清晰准确 → 双轨策略

  • 策略A(首选):Prompt中正确引导,文字放在色块背景上,白字蓝底对比
  • 策略B(保底):AI生成无文字版 → Python PIL叠加精准中文字层

问题3:动态视频文字消失 → 文字保留强制指令

  • 每条 gen_videos Prompt必含文字保留模块(见第四章)

问题4:内容单薄 → 三层场景结构强制要求

  • 前景(装饰粒子)+ 中景(多角色互动)+ 背景(场景延伸)

一、工具与输出规格

工具 用途 输出规格
image_synthesize 静态插画PNG(首选) 4K(4096px),16:9
nano-banana-pro 静态插画PNG(备选,中文更好) 1K/2K/4K
gen_videos(I2V模式) 动态视频(首选) 1080P,6秒,MP4
batch_image_to_video 动态视频(备选) 768P,6秒,MP4

⚠️ gen_videos 默认选6秒1080P;10秒1080P会自动降级为768P。


二、风格一致性保障(新增·改善1)

首次使用本技能时,必须先生成"风格参考图",作为后续所有图片的风格锚点。

步骤0:生成风格参考图(每批次只做一次)

目的:锁定标准角色外观、线条风格、配色体系,确保20张图风格一致

生成命令:

image_synthesize({
  requests: [{
    aspect_ratio: "16:9",
    output_file: "/workspace/fmt_style_guide.png",
    prompt: "[风格参考图Prompt,见下方标准模板]",
    resolution: "4K"
  }]
})

风格参考图Prompt(固定不变,每次套用):

[4K ULTRA HIGH DEFINITION - 4096px width, 16:9]

[STYLE REFERENCE SHEET - All future images must follow this exact style]

[TITLE AREA] "FMT肠菌移植 · 风格标准" in bold white Chinese on #B5DDE5 blue banner at top

[CHARACTER REFERENCE ROW - 5 chibi characters side by side]:
Character 1: Doctor - blue medical coat, stethoscope, round glasses, kind smile, #B5DDE5 outfit
Character 2: Nurse - light cyan uniform, medical cap, rosy cheeks, waving hand, #A8D8C8 outfit
Character 3: Patient (adult) - casual clothes, slightly worried expression turning hopeful, #FDE3D1 blush
Character 4: Child patient - big head, small body, bright curious eyes, brave smile, colorful clothes
Character 5: Gut bacteria (good) - small round colorful cute characters, different shapes/colors representing diverse flora

[STYLE RULES BOX]:
- Line width: 6-8px rounded thick strokes, soft and friendly
- Eye style: Bean eyes with white highlight dot, black pupil center
- Cheek: Soft peach blush oval on both cheeks
- Body proportion: Big head (50% of body height), small body, tiny limbs
- Shadow: NONE - flat design only
- Border: None, pure shapes

[COLOR PALETTE REFERENCE]:
Primary blue: #B5DDE5 (light lake blue)
Accent cyan: #A8D8C8 (soft teal)
Warm gold: #E8C98A (soft earth yellow)
Peach blush: #FDE3D1 (warm peach)
Deep text: #4A5E6F (gray-blue for text/icons)
White: #FFFFFF (background and text on dark)
Background: #F7FBFC (very light blue-cream)

[DECORATION REFERENCE]:
- Bubbles: Semi-transparent light blue circles, sizes 15px-60px
- Particles: Tiny white dots with soft glow
- Medical icons: Shield, cross, heart - #4A5E6F color, rounded style
- No shadows, no gradients, flat only

[LINE STYLE SAMPLE]:
Draw a sample line style card showing: thick rounded lines, no sharp corners,
all elements connected smoothly, children's illustration aesthetic

[BRAND AREA] Small: hospital icon + "苏州市立医院" in small Chinese text bottom left
[VERSION] "v6.0 Style Guide" bottom right in small grey text

保存路径/workspace/fmt_style_guide.png

后续所有图片Prompt开头必须附上风格锚定指令

[FOLLOW STYLE GUIDE: /workspace/fmt_style_guide.png]
Style rules: chibi characters, thick rounded lines, flat design, pastel palette as defined.
Use the same character designs, line width, eye style, and color palette as the reference.

三、批量生成模式(新增·改善2)

批量指令格式

用户可用以下任一格式一次性提交多页需求:

格式A:范围型

批量生成第3-7页,主题分别是:肠道菌群失衡、临床表现、诊断标准、治疗方案、康复管理

格式B:列表型

批量生成:p1=肠菌移植概述,p2=适应证,p3=禁忌证,p4=供体筛选,p5=制备流程

格式C:全文型

为《FMT肠菌移植科普.docx》全文生成插画,共20页,每页独立生成

批量处理流程(自动执行)

Step 1: 解析需求 → 生成页面列表 [p1,p2,...pN]
Step 2: 检查是否已有风格参考图
         - 无 → 先生成风格参考图(仅1次)
         - 有 → 跳过
Step 3: 逐页生成(每页独立):
         3a. 构建Prompt(锚定风格参考图)
         3b. image_synthesize → 4K PNG
         3c. 质量核查(见第四章)
         3d. 若通过 → gen_videos → 1080P视频
         3e. 若中文不准 → 策略B(PIL叠加)
         3f. 整理到 /workspace/fmt_pages/p[N]_主题/
Step 4: 完成后汇总报告

页面输出目录结构

/workspace/fmt_pages/
├── style_guide.png                    ← 风格参考图(只生成1次)
├── p01_肠菌移植概述/
│   ├── static.png                     ← 4K静态插画
│   ├── HD.mp4                         ← 1080P动态视频
│   └── metadata.json                  ← 页面元数据
├── p02_适应证/
│   └── ...
└── ...

metadata.json格式

{
  "page": 1,
  "title": "肠菌移植概述",
  "status": "completed",
  "static_generated": "2026-05-12T12:00:00",
  "video_generated": "2026-05-12T12:02:00",
  "chinese_quality": "pass",
  "notes": "中文清晰准确"
}

四、质量核查清单(新增·改善3)

每张图生成后、发送给用户前,必须执行以下核查:

核查A:图片清晰度

  • 图片尺寸 ≥ 4096px 宽(4K)
  • 无明显模糊区域
  • 线条边缘锐利,无锯齿

核查B:中文文字(策略A使用时)

  • 标题文字清晰可读,无乱码
  • 文字与背景对比度高(白字蓝底 / 深字白底)
  • 字号足够大(标题≥72px,正文≥48px)
  • 无文字错误或缺字

核查C:三层构图

  • 前景层:存在装饰粒子/气泡(5-8个)
  • 中景层:存在多角色互动(≥2人)+ 核心场景
  • 背景层:存在场景延伸,非空白

核查D:风格一致性

  • 与风格参考图(fmt_style_guide.png)风格一致
  • 角色外观符合参考图规范
  • 配色在标准色板范围内

核查E:动态视频(生成后)

  • 视频时长6秒
  • 分辨率1080P
  • 文字/图标/编号全程可见,未消失
  • 动效柔和,无剧烈抖动

核查结果处理:

  • 全部通过 → 发送给用户
  • 任一项失败 → 记录失败项 → 重新生成或切换策略B → 重新核查

五、降级策略(新增·改善4)

降级路径总览

image_synthesize 失败 → nano-banana-pro 备选
gen_videos 失败 → batch_image_to_video 降级768P
中文不准 → 策略B:PIL叠加字层
风格漂移 → 重新锚定风格参考图

详细降级规则

降级1:静态图生成失败

触发条件:image_synthesize 返回错误或图片质量极差(无法识别内容)
降级动作:切换到 nano-banana-pro
  uv run ~/.codex/skills/nano-banana-pro/scripts/generate_image.py \
    --prompt "[原Prompt精简版,不含resolution参数]" \
    --filename "/workspace/fmt_page_[N]_[主题].png" \
    --resolution 4K
修复后:继续原流程

降级2:动态视频生成失败

触发条件:gen_videos 返回错误或输出文件损坏
降级动作:切换 batch_image_to_video
  batch_image_to_video({
    count: 1,
    image_file_list: ["/workspace/fmt_page_[N]_[主题].png"],
    output_file_list: ["/workspace/fmt_page_[N]_[主题]_HD.mp4"],
    prompt_list: ["[动态Prompt精简版,不含分辨率指令]"],
    duration_list: [6],
    resolution_list: ["768P"]
  })
修复后:告知用户视频为768P备选版本

降级3:中文不准(PIL叠加字层)

触发条件:image_synthesize 生成的中文出现错误/乱码/模糊
修复动作:
  Step A: 生成无文字版本(在原Prompt后追加 "NO TEXT, NO CHINESE CHARACTERS")
  Step B: 使用Python PIL叠加中文字层
    python3 -c "
    from PIL import Image, ImageDraw, ImageFont
    img = Image.open('/workspace/fmt_page_[N]_[主题].png').convert('RGBA')
    w, h = img.size
    scale = w / 4096  # 4K基准
    # 标题栏
    bar = Image.new('RGBA', (w, int(h*0.10)), (181,205,214,245))
    img.paste(bar, (0,0), bar)
    draw = ImageDraw.Draw(img)
    try:
        font_title = ImageFont.truetype('/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc', int(64*scale))
        font_sub = ImageFont.truetype('/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc', int(36*scale))
    except:
        font_title = ImageFont.load_default()
        font_sub = ImageFont.load_default()
    draw.text((w//2, int(h*0.05)), '[标题文字]', fill=(255,255,255,255), font=font_title, anchor='mm')
    draw.text((w//2, int(h*0.08)), '[副标题文字]', fill=(200,220,240,255), font=font_sub, anchor='mm')
    img.convert('RGB').save('/workspace/fmt_page_[N]_[主题]_cn.png')
    "
  Step C: 用 _cn.png 作为基础重新调用 gen_videos

降级4:风格漂移

触发条件:多张图风格不一致(角色外观、线条粗细、配色偏差)
修复动作:
  重新调用 image_synthesize,在Prompt开头强制附上:
  "[MUST FOLLOW STYLE GUIDE: /workspace/fmt_style_guide.png]"
  "Exact same line width, eye style, body proportion, color palette"

六、中文字解决策略(核心)

策略A:AI直出中文(首选)

关键技巧:

  1. 用英文括号包裹中文:"肠菌移植" in bold white Chinese
  2. 强化指令:"render Chinese text accurately, crisp and clear"
  3. 指定字体风格:"Chinese text in clean sans-serif font, thick strokes"
  4. 文字放在色块背景上,白字蓝底,对比度最高
  5. 字号占图高3-5%(4K图上约120-200px)

配色对比强化中文清晰度:

背景色 文字色 效果
#B5DDE5(浅湖蓝) #FFFFFF(白色) ✅ 最清晰
#4A5E6F(深灰蓝) #FFFFFF(白色) ✅ 清晰
#E8C98A(金色) #4A5E6F(深色) ✅ 清晰
#FFFFFF(白色) #4A5E6F(深色) ⚠️ 可用
透明/渐变 任意颜色 ❌ 模糊,禁止

七、静态插画生成规范(4K超清)

构图层次要求

[前景层 - FOREGROUND] 5-8个元素
- 彩色气泡(半透明浅蓝色,多种尺寸)
- 光点粒子(白色微光点)
- 小装饰图标(医疗符号、爱心、星标)

[中景层 - MIDDLE GROUND] 核心内容层
- 主要角色(3-5个Q版卡通人物互动)
- 核心场景主体(器官/设备/流程道具)
- 信息色块(带中文标题的色块卡片)
- 编号圆圈①②③标注流程步骤

[背景层 - BACKGROUND]
- 场景延伸(肠道内壁纹理/医院走廊/自然环境)
- 渐变色调(从浅到深,暗示景深)

视觉风格规范

属性 要求
风格 数字手绘插画(Digital Hand-drawn Illustration)
人物 Q版大头,豆豆眼(深色+白色高光),淡淡腮红,粗圆润线条
线条 柔和圆润,6-8px粗,无锐角,无阴影
色彩 低饱和马卡龙色系,纯扁平无阴影
背景 极浅纯色(#F7FBFC),充足留白
画布 16:9,4K(4096px宽),横版

配色方案

颜色 色值 用途
浅湖蓝 #8ECCD6 / #B5DDE5 主色调/标题栏
浅土黄 #D4A96A / #E8C98A 强调、点缀
沙色 #F2E4C9 / #F5E8D3 背景辅助
淡青 #A8D8C8 / #B8E4D4 辅助色调
雾蓝 #D6EAF0 / #E8F4FA 装饰、留白
深灰蓝 #4A5E6F 中文字、图标底色
白色 #FFFFFF 背景/白字底色

八、动态视频生成规范(1080P+文字保留)

动效Prompt结构

[FOLLOW STYLE GUIDE: /workspace/fmt_style_guide.png]
The chibi cartoon medical illustration comes gently to life with rich layered motion.

[LAYER 1 - BACKGROUND ANIMATION]:
- Background texture gently breathes with soft color waves, subtle parallax drift
- Organic gut-wall pattern slowly shifts, very subtle

[LAYER 2 - MIDDLE GROUND ANIMATION]:
- [核心角色] gently bob up and down, arms wave slowly, smile widens gradually
- [主要场景] twinkle with warm glow, pulse gently, never disappear
- [流程步骤] ①②③ numbered circles pulse in gentle sequence
- [中文字区域] stable with soft breathing glow, text always visible

[LAYER 3 - FOREGROUND ANIMATION]:
- Bubbles float upward slowly with slight wobble
- Light particles twinkle and sparkle gently
- Small medical icons pulse with warm glow

[CRITICAL - TEXT STABILITY]:
- Every text element, Chinese label, number badge remains FIXED in position
- NO text fades, disappears, or becomes blurry at any point
- Text may have gentle shimmer/glow but must always remain legible
- Chinese characters stay crisp and stable throughout entire animation
- All ①②③ circles stay clearly visible with steady light pulse

[ANIMATION QUALITY]:
- High definition 1080P, ultra crisp and clear
- Gentle looping animation, dreamy and warm, never frantic
- Dreamy atmosphere, educational and friendly tone
Duration: 6 seconds. All characters and text in place throughout.

九、中文优化Prompt片段库

标题型

[TOP BANNER] Full-width #B5DDE5 light blue banner at top:
"肠菌移植" in large bold white Chinese, very large font, crisp and clear
"重建肠道健康" in smaller white Chinese below
Chinese font: bold rounded sans-serif, thick strokes, white-on-blue high contrast

步骤标注型

[STEP INDICATORS] Three numbered circles:
Circle 1: "① 评估" white text on #4A5E6F dark blue circle
Circle 2: "② 供体" white text on #B5DDE5 light blue circle
Circle 3: "③ 移植" white text on #E8C98A gold circle
All: bold font, crisp Chinese, high contrast, evenly spaced

对话气泡型

[SPEECH BUBBLE] White rounded rectangle, soft #B5DDE5 border:
"医生:您好,我来为您详细说明。"
in #4A5E6F dark blue Chinese text, 48px on 4K image, clean rounded font

信息卡片型

[INFO CARDS] Three cards with colored top strips:
Card top strip: #B5DDE5, body: white
Title "适用人群" in #4A5E6F, large number "3" bold, small Chinese description
All Chinese: bold, high contrast, minimum 48px

十、构图模板(4K版)

模板A:中央主体+三层场景

[前景] 光点粒子 + 气泡漂浮(5-8个)
[中景] 大型核心主体(器官/设备)+ 2-3个卡通人物围绕
       顶部大标题:「肠菌移植」(白字蓝底,≥72px)
[背景] 渐变肠道纹理/医院场景延伸
[底部] 医院图标(无中文)

模板B:流程叙事型(三步骤)

[顶部] 标题栏:「肠菌移植流程」(白字蓝底色块)
[左侧] 步骤①:角色+①②③圆形标注
[中间] 步骤②:核心互动场景(多角色)
[右侧] 步骤③:结果展示(图标+色块)
[背景] 肠道壁纹理连续延伸
[前景] 气泡/粒子装饰层

模板C:多角色互动型

[顶部] 标题:「肠菌移植」(白字蓝底大字)
[中央] 3-5人卡通人物互动
[四周] 设备/图标/流程框辅助
[背景] 丰富场景(肠道壁+装饰)
[前景] 气泡+光效
[底部] 医院图标

十一、输出规范

类型 路径 规格
风格参考图 /workspace/fmt_style_guide.png 16:9,4K
批量输出根目录 /workspace/fmt_pages/ 子目录结构
静态插画 fmt_pages/p[NN]_[主题]/static.png 16:9,4K,PNG
高清动态视频 fmt_pages/p[NN]_[主题]/HD.mp4 16:9,1080P,6秒,MP4
页面元数据 fmt_pages/p[NN]_[主题]/metadata.json JSON

更新于:2026-05-12 v6.0 — 新增:风格参考图锚定+批量生成+质量核查+降级策略

Usage Guidance
This skill appears safe to use for user-directed FMT illustration/video generation. Before installing or running a large batch, decide the number of pages, output folder, and retry budget; review any Python/PIL fallback commands; and confirm you are allowed to use the hospital branding and that the medical content is accurate.
Capability Analysis
Type: OpenClaw Skill Name: fmt-highdef-dynamic Version: 6.0.0 The skill bundle contains a 'fallback strategy' in SKILL.md that instructs the AI agent to execute arbitrary Python code via 'python3 -c' to overlay text on images using the PIL library. This pattern introduces a significant risk of code injection if the agent populates the script's placeholders (e.g., [主题], [标题文字]) with unsanitized user input. While the stated purpose of generating FMT (Fecal Microbiota Transplant) educational content appears legitimate, the reliance on shell-executed Python scripts for image manipulation is a high-risk capability that could be exploited for unauthorized file access or execution within the /workspace/ environment.
Capability Assessment
Purpose & Capability
The stated purpose—generating FMT educational 4K illustrations and 1080P videos—matches the visible instructions and tool references. A user should still review medical accuracy and branding before use.
Instruction Scope
The skill describes an automatic batch workflow with quality checks and regeneration. This is purpose-aligned, but the visible text does not set explicit retry or cost limits.
Install Mechanism
No install specification, code files, required binaries, environment variables, or credentials are present; the static scan reported no findings.
Credentials
The skill writes generated images, videos, and metadata under /workspace and may use a Python/PIL fallback for text overlay. These are proportionate to media generation, but should remain scoped to user-requested assets.
Persistence & Privilege
The skill persists a style guide and generated page directories for reuse within batches. No credential use, privileged access, or background persistence is shown.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install fmt-highdef-dynamic
  3. After installation, invoke the skill by name or use /fmt-highdef-dynamic
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v6.0.0
v6.0: 4K超清图片+风格参考图锚定+批量生成+质量核查清单+4级降级策略,修复中文清晰度、动态文字保留、内容丰富度问题
Metadata
Slug fmt-highdef-dynamic
Version 6.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is FMT高清动态图生成器 v6.0?

FMT肠菌移植科普高清动态图生成器(插画版+4K静态图+1080P动态视频+批量模式+风格一致)。生成配套数字手绘插画静态PNG(4K超清),并转化为高清晰度1080P动态视频。风格活泼生动,适合科普教育。触发词:生成FMT动态图 / 生成高清动态 / 插画转高清视频。 It is an AI Agent Skill for Claude Code / OpenClaw, with 50 downloads so far.

How do I install FMT高清动态图生成器 v6.0?

Run "/install fmt-highdef-dynamic" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is FMT高清动态图生成器 v6.0 free?

Yes, FMT高清动态图生成器 v6.0 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does FMT高清动态图生成器 v6.0 support?

FMT高清动态图生成器 v6.0 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created FMT高清动态图生成器 v6.0?

It is built and maintained by mingyuan (@zmy1006-sudo); the current version is v6.0.0.

💬 Comments