← 返回 Skills 市场
FreeAds 随手拍广告
作者
lipeng0820
· GitHub ↗
· v3.7.0
· MIT-0
154
总下载
0
收藏
0
当前安装
17
版本数
在 OpenClaw 中安装
/install freeads-snap-ad
功能描述
🎬 AI 高端广告视频生成器 - 将产品照片转化为 8 秒专业广告视频(含 BGM、Slogan、音效、丰富运镜)。使用 Atlas Cloud API 调用 Veo 3.1 生成视频。触发词:随手拍广告、生成广告视频、产品广告。核心输出:视频文件 URL。
使用说明 (SKILL.md)
FreeAds 随手拍广告 🎬
AI 高端广告视频生成器 | 最终产出:8 秒 TVC 级 MP4 视频(含 BGM、音效、专业运镜)
📦 安装:
clawhub install lipeng0820/freeads-snap-ad
⚠️ 核心原则:产品外观保护(最高优先级)
这是本 Skill 的价值核心,必须严格遵守:
🛡️ 产品保真三原则
- 外观不走样 - 产品的形状、轮廓、设计细节必须与原图完全一致
- 材质不臆断 - 如果无法确定材质,不要描述,让 Veo 3.1 从图片推断
- 品牌不乱改 - 如果看不清品牌标识,不要猜测或编造
📊 置信度判断机制
在产品识别时,模型必须对以下属性输出置信度分数(0-100):
{
"product": {
"type": {"value": "Air Fryer", "confidence": 95},
"material": {"value": "matte white plastic", "confidence": 60},
"color": {"value": "white", "confidence": 90},
"brand": {"value": "unknown", "confidence": 20},
"texture": {"value": "smooth finish", "confidence": 45}
}
}
规则:当 confidence \x3C 80 时,该属性不得在 Veo 3.1 提示词中描述!
🌐 语言处理规则
- 与用户对话:使用用户的语言(中文用户用中文回复)
- 与底层模型交互:始终使用英文
- 翻译输出:将模型返回的英文结果翻译成用户语言后展示
🔑 API Key 获取
当用户没有配置 ATLASCLOUD_API_KEY 时:
❌ 需要 Atlas Cloud API Key 才能生成视频 🎬
请按以下步骤获取:
1. 访问 Atlas Cloud: https://www.atlascloud.ai?ref=LJNA3T
🎁 新用户福利:使用此链接注册,首次充值可获得 25% 奖励(最高 $100)!
2. 登录后进入 Console -> API Keys
3. 创建并复制 API Key
4. 配置环境变量:export ATLASCLOUD_API_KEY="your-api-key"
🚨 执行流程
交互式确认流程
Step 1: 识别产品(含置信度) → 展示结果 → 等待用户确认 ✅
Step 2: 场景设计 → 展示结果 → 等待用户确认 ✅
Step 3: 分镜脚本 → 展示结果 → 等待用户确认 ✅
Step 4: 展示 Slogan → 等待用户确认 ✅
Step 5: 展示 Veo 3.1 提示词(低置信度属性已过滤) → 等待用户确认 ✅
Step 6: 生成视频 → 同时展示 Tips → 返回视频 URL
🎯 产品识别 Prompt(带置信度)
PRODUCT_RECOGNITION_PROMPT = """You are a product identification expert. Analyze the product image with EXTREME PRECISION.
## CRITICAL: CONFIDENCE SCORING
For EACH attribute, you MUST provide a confidence score (0-100):
- 90-100: Absolutely certain, clearly visible
- 70-89: Reasonably confident, mostly visible
- 50-69: Uncertain, partially visible or ambiguous
- 0-49: Cannot determine, guessing, or not visible
## PROTECTION RULES
1. **DO NOT GUESS** - If you cannot clearly see something, confidence should be LOW
2. **DO NOT FABRICATE** - Never make up brand names, logos, or specific details
3. **DO NOT ASSUME** - If material/texture is unclear, say "unknown" with low confidence
4. **PRESERVE ORIGINAL** - Your goal is to DESCRIBE what you SEE, not what you IMAGINE
## OUTPUT FORMAT (JSON)
{
"product": {
"category": {"value": "product category", "confidence": 95},
"name": {"value": "product name", "confidence": 90},
"shape": {"value": "shape description", "confidence": 85},
"color": {"value": "color description", "confidence": 92},
"material": {"value": "material or 'unknown'", "confidence": 55},
"texture": {"value": "texture or 'unknown'", "confidence": 40},
"brand": {"value": "brand name or 'unknown'", "confidence": 25},
"logo_visible": {"value": true/false, "confidence": 80},
"key_features": [
{"value": "feature 1", "confidence": 88},
{"value": "feature 2", "confidence": 75}
]
},
"appearance_summary": "Brief factual description of what is CLEARLY VISIBLE in the image",
"uncertain_elements": ["list of elements that are unclear or ambiguous"],
"overall_product_confidence": 85
}
## EXAMPLES
❌ WRONG (low confidence but detailed description):
"material": {"value": "premium brushed aluminum with anodized coating", "confidence": 45}
✅ CORRECT (low confidence = vague or unknown):
"material": {"value": "metallic surface, specific material unknown", "confidence": 45}
❌ WRONG (fabricating brand):
"brand": {"value": "Xiaomi", "confidence": 30}
✅ CORRECT (honest about uncertainty):
"brand": {"value": "unknown - no visible branding", "confidence": 10}
"""
📝 Veo 3.1 提示词构建(置信度过滤)
关键:只有 confidence >= 80 的属性才能进入 Veo 3.1 提示词!
def build_veo_prompt(product_data, scene_data, shots_data, slogan):
"""
构建 Veo 3.1 提示词,自动过滤低置信度属性
"""
# 提取高置信度属性
high_confidence_attrs = []
for attr, data in product_data.items():
if isinstance(data, dict) and data.get("confidence", 0) >= 80:
high_confidence_attrs.append(f"{attr}: {data['value']}")
# 如果高置信度属性很少,使用通用描述
if len(high_confidence_attrs) \x3C 3:
product_description = "the product exactly as shown in the input image"
else:
product_description = ", ".join(high_confidence_attrs)
# 构建提示词
prompt = f"""Premium 8-second TVC commercial.
PRODUCT APPEARANCE PROTECTION (CRITICAL):
- The product MUST appear EXACTLY as shown in the input image
- DO NOT modify, enhance, or reimagine the product's appearance
- Maintain exact shape, color, proportions, and all visible details
- If there are logos or brand elements, keep them exactly as shown
- The input image is the ONLY source of truth for product appearance
PRODUCT: {product_description}
SCENE: {scene_data.get('setting_description', 'professional studio setting')}
LIGHTING: {scene_data.get('lighting', 'professional studio lighting')}
MOOD: {scene_data.get('mood', 'premium, elegant')}
SHOTS:
"""
for shot in shots_data:
prompt += f"- {shot['timing']}: {shot['description']}. Camera: {shot['camera']}\
"
prompt += f"""
SLOGAN: "{slogan}" appears elegantly in the final shot.
STYLE: Premium TVC commercial, cinematic quality, Super Bowl ad level.
AUDIO: Epic background music with synchronized sound effects.
ABSOLUTE REQUIREMENT: The product's visual appearance must be IDENTICAL to the input image.
No creative liberties with product design, color, shape, or branding."""
return prompt
🎬 分镜脚本 Prompt(保护产品外观)
STORYBOARD_PROMPT = """You are a TVC commercial director. Create a 3-shot storyboard for an 8-second video.
## PRODUCT APPEARANCE PROTECTION (HIGHEST PRIORITY)
⚠️ WARNING: You are NOT allowed to describe specific product details unless you are CERTAIN.
RULES:
1. Use phrases like "the product" or "the item" instead of detailed descriptions
2. Focus on CAMERA MOVEMENT, LIGHTING, and SCENE - NOT product details
3. DO NOT describe materials, textures, or brand elements unless confidence >= 80%
4. Let the input image speak for itself - Veo 3.1 will preserve the product appearance
## GOOD vs BAD EXAMPLES
❌ BAD (too specific, may cause distortion):
"The brushed stainless steel air fryer with digital LED display rotates..."
✅ GOOD (respects product appearance):
"The product rotates slowly, showcasing its design from multiple angles..."
❌ BAD (fabricating details):
"The Apple logo gleams under the studio lights..."
✅ GOOD (factual):
"Any visible branding catches the light naturally..."
## SCENE SELECTION
Based on product category, choose appropriate environment:
- Kitchen Appliance → Modern kitchen, marble countertop
- Electronics → Clean desk, tech workspace
- Fashion → Lifestyle setting, editorial style
- Beauty → Spa-like bathroom, soft lighting
- Default → Professional studio with gradient backdrop
## OUTPUT FORMAT (JSON)
{
"scene": {
"environment": "environment type",
"setting_description": "scene description WITHOUT product details",
"lighting": "lighting description",
"mood": "emotional tone"
},
"slogan": "2-5 word slogan",
"shots": [
{
"number": 1,
"timing": "0-3s",
"type": "REVEAL",
"description": "Shot description focusing on camera and scene, NOT product details",
"camera": "camera movement",
"audio": "audio cue"
},
{
"number": 2,
"timing": "3-6s",
"type": "SHOWCASE",
"description": "Shot description - let the product's actual appearance shine",
"camera": "camera movement",
"audio": "audio cue"
},
{
"number": 3,
"timing": "6-8s",
"type": "HERO",
"description": "Final shot with slogan",
"camera": "camera movement",
"audio": "audio cue"
}
],
"product_appearance_note": "Brief note about what aspects of the product you're uncertain about"
}
"""
📊 用户确认界面(显示置信度)
产品识别结果
## 🎯 产品识别结果
| 属性 | 识别结果 | 置信度 | 是否用于视频描述 |
|------|----------|--------|------------------|
| **类型** | 厨房电器 | 🟢 95% | ✅ 是 |
| **名称** | 空气炸锅 | 🟢 92% | ✅ 是 |
| **颜色** | 白色 | 🟢 90% | ✅ 是 |
| **形状** | 方形立式 | 🟢 88% | ✅ 是 |
| **材质** | 未知(看起来像塑料) | 🟡 55% | ❌ 否(交由 Veo 判断) |
| **品牌** | 未识别 | 🔴 20% | ❌ 否(保持原样) |
| **纹理** | 未知 | 🔴 40% | ❌ 否(保持原样) |
### 🛡️ 产品保护说明
> 为保护产品外观不走样,以下属性将**不会**在视频提示词中详细描述,而是让 Veo 3.1 直接参考原图:
> - 材质(置信度 55% \x3C 80%)
> - 品牌(置信度 20% \x3C 80%)
> - 纹理(置信度 40% \x3C 80%)
>
> 这确保视频中的产品外观与您提供的原图保持一致。
---
👆 以上识别结果是否准确?低置信度属性将保持原样不做描述。
- 回复「确认」继续
- 回复「修改」并告诉我需要调整的内容
🎥 Veo 3.1 最终提示词模板
注意:低置信度属性使用通用表述
Premium 8-second TVC commercial.
PRODUCT APPEARANCE PROTECTION (CRITICAL):
- The product MUST appear EXACTLY as shown in the input image
- DO NOT modify, enhance, or reimagine the product's appearance
- Maintain exact shape, color, proportions, and all visible details
- If there are logos or brand elements, keep them exactly as shown
- The input image is the ONLY source of truth for product appearance
PRODUCT: [仅包含高置信度属性,如 "white kitchen appliance with digital display"]
[低置信度属性省略,由 Veo 从图片判断]
SCENE: Modern minimalist kitchen with marble countertop
LIGHTING: Soft natural morning light
MOOD: Fresh, clean, inviting
SHOTS:
- 0-3s: The product is revealed in its natural kitchen setting. Camera: Slow dolly in
- 3-6s: Smooth orbital rotation showcasing the product from multiple angles. Camera: 60-degree arc
- 6-8s: Hero shot with product centered, slogan fades in. Camera: Static with subtle drift
SLOGAN: "Crisp Perfection"
ABSOLUTE REQUIREMENT: The product's visual appearance must be IDENTICAL to the input image.
No creative liberties with product design, color, shape, or branding.
💡 Tips 生成(视频等待期间)
TIPS_PROMPT = """基于产品类型:{product_type}
生成 5 条关于如何保护产品外观、写出更好广告提示词的技巧。
重点强调:
1. 如何避免产品外观走样
2. 哪些细节不应该过度描述
3. 如何让视频保持产品原始形态
4. 场景选择的技巧
5. 运镜设计的建议
用中文输出。"""
模型配置
| 功能 | 模型 ID |
|---|---|
| 产品识别(带置信度) | moonshotai/kimi-k2.5 |
| 分镜脚本 | moonshotai/kimi-k2.5 |
| Tips 生成 | moonshotai/kimi-k2.5 |
| 视频生成 | google/veo3.1/image-to-video |
版本历史
- v3.7.0 - 🛡️ 核心升级:引入置信度判断机制,低于 80% 的属性不进入提示词;强化产品外观保护
- v3.6.0 - 新增交互式确认流程;Tips 贴士;多语言支持
- v3.5.x - 优化分镜结构;智能场景选择
安全使用建议
What to check before installing: 1) Metadata mismatch — SKILL.md requires ATLASCLOUD_API_KEY but the registry metadata lists none; confirm you are comfortable granting an Atlas Cloud API key. 2) Review the included setup script (scripts/setup-api-key.sh) before running: it will optionally append export ATLASCLOUD_API_KEY to your shell rc (~/.bashrc, ~/.zshrc). If you prefer, set ATLASCLOUD_API_KEY manually instead of running the script. 3) Use a least-privilege / billing-limited Atlas API key if possible and monitor usage/billing after first runs. 4) Be aware the docs include an affiliate/referral link — sign-up via that link is optional and unrelated to functionality. 5) If you will supply images containing third-party brands, confirm you have rights to use them; the skill encourages web searches for logos which may cause the agent to fetch external resources. 6) If you need tighter safety, disable autonomous invocation for this skill (or only invoke it manually) and test with non-sensitive/example images first.
功能分析
Type: OpenClaw Skill
Name: freeads-snap-ad
Version: 3.7.0
The skill is a legitimate tool designed to generate advertisement videos using the Atlas Cloud and Google Veo 3.1 APIs. The included shell script (scripts/setup-api-key.sh) is a standard configuration utility that validates the user's API key against the official service provider and persists it in shell configuration files (.zshrc/.bashrc) for environment variable management. The SKILL.md and documentation contain detailed, task-aligned instructions for the AI agent, including a 'confidence mechanism' to ensure product fidelity, and no evidence of malicious prompt injection, data exfiltration, or unauthorized execution was found.
能力评估
Purpose & Capability
The SKILL.md explicitly requires an ATLASCLOUD_API_KEY and documents calls to Atlas Cloud's upload/generate endpoints (coherent with a video-generation skill). However the registry metadata at the top of the submission lists no required env vars/primary credential — this mismatch is unexpected and should be resolved. The skill also includes affiliate/referral URLs (https://www.atlascloud.ai?ref=LJNA3T) which are not necessary for functionality and suggest monetization/marketing embedded in docs.
Instruction Scope
Runtime instructions stick to the advertised workflow: product recognition, prompt construction, uploading media, and calling Veo 3.1. They also instruct the agent to collect brand/logo images and (optionally) search the web for logos — reasonable for brand-protection features but potentially expands what the agent will fetch or request from the user. The SKILL.md header declares ATLASCLOUD_API_KEY as required while the registry metadata omitted it; that inconsistency may cause permission/consent gaps at install time.
Install Mechanism
There is no install spec (instruction-only) which is low risk. A single included shell helper (scripts/setup-api-key.sh) is present; its behavior is straightforward (prompt for API key, verify via curl, and optionally append/export the env var into the user's shell rc). The script is readable and not obfuscated, but it will modify the user's shell config if accepted — users should inspect it before running.
Credentials
The only credential this skill needs is an Atlas Cloud API key (ATLASCLOUD_API_KEY), which is proportionate to its purpose. However the top-level registry metadata declared no required env vars while SKILL.md and the setup script require and store that API key — an incoherence that may lead to the key being requested or stored without explicit visibility in the registry listing.
Persistence & Privilege
always:false (no forced permanent inclusion). The skill's included setup script can write the API key into the user's shell RC (persistent environment change). Also the skill can invoke external APIs autonomously (default platform behavior); coupling autonomous invocation with access to an API key increases blast radius if the skill were malicious. There is no evidence the skill modifies other skills or system-wide settings.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install freeads-snap-ad - 安装完成后,直接呼叫该 Skill 的名称或使用
/freeads-snap-ad触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v3.7.0
v3.7.0: 🛡️ 核心升级 - (1) 引入置信度判断机制:产品属性需评估 0-100 置信度 (2) 低于 80% 置信度的属性(材质、品牌、纹理等)不进入 Veo 提示词,交由模型从原图判断 (3) 强化产品外观保护:外观不走样、材质不臆断、品牌不乱改 (4) 用户确认界面显示置信度和保护说明
v3.6.0
v3.6.0: (1) 新增交互式确认流程:产品识别→场景设计→分镜脚本→Slogan→提示词,每步都需用户确认后才继续 (2) 视频生成等待期间展示 Tips 贴士 (3) 多语言支持:与用户用中文对话,与底层模型用英文交互
v3.5.2
v3.5.2: 修复 API Key 获取提示,使用带推荐码的 Atlas Cloud 链接 (https://www.atlascloud.ai?ref=LJNA3T),新用户首充可获 25% 奖励(最高 $100)
v3.5.1
v3.5.1: 更新 README 文档,增加 Atlas Cloud 注册引导和新用户首充 25% 奖励说明
v3.5.0
v3.5.0: (1) 优化分镜结构,严格限制最多3个镜头 (2) 根据产品类别智能选择场景环境(厨房电器→厨房,电子产品→科技空间等)(3) 输出结构化 JSON 分镜数据 (4) 完整中间结果报告(产品识别、场景设计、分镜脚本、Veo提示词、视频URL)(5) 分镜脚本模型改用 moonshotai/kimi-k2.5(支持多模态)
v3.1.0
v3.1.0: (1) 视频模型改用 openai/sora-2/image-to-video (2) 增加每一步中间结果输出,方便排查产品不一致问题 (3) 大幅简化文档
v3.0.1
v3.0.1: 强调 Veo 3.1 是 Image-to-Video 模型,必须传入用户的产品图 URL,视频中的产品必须与用户上传的完全一致,不能生成无关产品
v3.0.0
v3.0.0 重大更新: (1) 移除 LOGO 收集功能避免干扰产品识别 (2) 模型从 gemini-3.1-pro 改为 zai-org/glm-5-turbo (3) 增强视频 prompt 要求 BGM、Slogan、音效、丰富运镜 (4) 简化工作流程只用产品图 (5) 作者更正为 lipeng0820
v2.0.9
v2.0.9: 重大更新 - 明确定位为 AI 高端广告视频生成器(不是文案方案生成器);强调最终产出必须是 MP4 视频文件;禁止输出文案方案、设计建议作为最终结果;增加工作流程表格
v2.0.8
v2.0.8: 强化视频生成指令,明确禁止 AI 助手找任何借口(如'没有视频模型'),必须调用 google/veo3.1/image-to-video API
v2.0.7
v2.0.7: 将抠图步骤改为必须执行,确保先提取产品主体再进行视频生成,避免背景杂物干扰广告效果
v2.0.6
v2.0.6: 修复产品识别不准确问题,强化 Gemini 3.1 Pro 的产品分析提示词,禁止猜测,要求准确识别材质和用途
v2.0.5
v2.0.5: 强调最终产出必须是视频文件,增加 API 失败处理规则,明确禁止用其他内容替代视频输出
v2.0.4
v2.0.4: 修正 Veo 3.1 模型名称为 google/veo3.1/image-to-video(无破折号),更新视频费用为官方价格 $0.09-$0.20/秒
v2.0.3
v2.0.3: 修正 Atlas Cloud API 模型名称格式(google/nano-banana-2/edit, google/veo-3.1/image-to-video)
v2.0.1
v2.0.1: 添加快速安装提示,方便用户一键安装
v2.0.0
v2.0.0: 新增品牌LOGO识别保护功能和费用预估
元数据
常见问题
FreeAds 随手拍广告 是什么?
🎬 AI 高端广告视频生成器 - 将产品照片转化为 8 秒专业广告视频(含 BGM、Slogan、音效、丰富运镜)。使用 Atlas Cloud API 调用 Veo 3.1 生成视频。触发词:随手拍广告、生成广告视频、产品广告。核心输出:视频文件 URL。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 154 次。
如何安装 FreeAds 随手拍广告?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install freeads-snap-ad」即可一键安装,无需额外配置。
FreeAds 随手拍广告 是免费的吗?
是的,FreeAds 随手拍广告 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
FreeAds 随手拍广告 支持哪些平台?
FreeAds 随手拍广告 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 FreeAds 随手拍广告?
由 lipeng0820(@lipeng0820)开发并维护,当前版本 v3.7.0。
推荐 Skills