← 返回 Skills 市场
nevo-david

Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI.

作者 nevo-david · GitHub ↗ · v1.0.1
cross-platform ⚠ suspicious
951
总下载
0
收藏
7
当前安装
2
版本数
在 OpenClaw 中安装
/install agent-media
功能描述
AI UGC video production from the terminal using the `agent-media` CLI.
使用说明 (SKILL.md)

npm release: https://www.npmjs.com/package/agent-media-cli agent-media cli github: https://github.com/gitroomhq/agent-media official website: https://agent-media.ai

agent-media — AI UGC Video Production & Media Generation

Produce complete UGC videos and SaaS review videos from the terminal using the agent-media CLI.


MANDATORY RULES — READ BEFORE EVERY COMMAND

You MUST follow ALL of these rules. Violating any rule produces a broken, unwatchable video.

Rule 1: ALWAYS use --actor

Every UGC video MUST include --actor \x3Cslug>. Without an actor, there is no talking head and no lip sync — the video will just be a static image with voiceover. If the user hasn't specified an actor, ask them to pick one or default to a popular one like sofia or naomi. Run agent-media actor list to browse options.

Rule 2: ALWAYS count words — 2.5 words per second

Natural speech is 2.5 words/second. Scripts MUST match the target duration:

  • 5s video → ~12 words max
  • 10s video → ~25 words max
  • 15s video → ~37 words max

If the user provides a script that is too long, YOU MUST rewrite it to fit. A 73-word script crammed into 15 seconds will sound robotic and lose lip sync. NEVER submit a script that exceeds the word limit.

Rule 3: SaaS reviews MUST have screenshots

For any SaaS/product review video, you MUST provide 1-3 product screenshots via --broll --broll-images. Without screenshots, the video has no product context — viewers see only a talking head with no evidence of the product.

--broll-images accepts both HTTP/HTTPS URLs and local file paths (local files are auto-uploaded). Images are semantically matched to the most relevant broll scene based on filename — so use descriptive filenames! Examples:

  • --broll-images ./dashboard.png,./calendar-view.png (local files — descriptive names!)
  • --broll-images https://example.com/pricing-page.png,https://example.com/editor.png (URLs)
  • Mix of both works too

If the user provides a product URL but no screenshots, visit the site yourself and extract image URLs from the page.

Rule 4: SaaS reviews MUST have the product name

Ask the user: "What SaaS product should I review?" Do NOT proceed without it. The product name must appear in the script so it shows up in subtitles.

Rule 5: Always use --sync

Always append --sync to wait for the result and get the output URL.

Rule 6: Name screenshot files descriptively

Broll images are semantically matched to scenes by filename. Use descriptive names so the AI assigns the right image to the right scene:

  • GOOD: dashboard.png, calendar-view.png, post-editor.png
  • BAD: screenshot1.png, IMG_0042.png, image.png

When saving screenshots for --broll-images, rename them to describe what they show.


Prerequisites

The agent-media CLI must be installed and authenticated:

npm install -g agent-media-cli
agent-media login

Verify with agent-media whoami. If not logged in, run agent-media login and follow the OTP flow.

UGC Pipeline (Flagship Feature)

The UGC pipeline turns a script into a complete video with AI talking heads, B-roll, voiceover, and animated subtitles — one command.

Flow

Script → Scene splitting → TTS voiceover → AI talking heads + B-roll → Crossfade assembly → Animated subtitles → Background music → End screen CTA

Basic UGC

# ALWAYS include --actor for talking heads + lip sync
agent-media ugc "Ever wonder why some videos go viral? Here's the secret..." --actor sofia --sync

# From file
agent-media ugc ./script.txt --actor naomi --sync

# AI-generated script from a product description
agent-media ugc -g "A fitness tracker that monitors sleep quality" --actor marcus --sync

UGC with B-roll

# With B-roll cutaway scenes mixed in
agent-media ugc "Your script here..." --actor marcus --broll --sync

# With product screenshots as B-roll (REQUIRED for SaaS reviews)
agent-media ugc "Your script here..." --actor sofia --broll --broll-images https://example.com/screenshot1.png,https://example.com/screenshot2.png --sync

UGC Flags

Flag Description Example
--actor \x3Cslug> Library actor for talking heads --actor sofia
--persona \x3Cslug> Custom persona (cloned voice + face) --persona brand-voice
--face-url \x3Curl> Direct face photo URL or local file --face-url ./photo.png
--voice \x3Cname> TTS voice --voice nova
--tone \x3Cname> Voice tone: energetic, calm, confident, dramatic --tone energetic
--style \x3Cname> Subtitle style (default: hormozi) --style bold
-d, --duration \x3Cs> Target duration: 5, 10, or 15 seconds --duration 10
--aspect \x3Cratio> Aspect ratio: 9:16, 16:9, 1:1 --aspect 16:9
--music \x3Cgenre> Background music: chill, energetic, corporate, dramatic, upbeat --music chill
--cta \x3Ctext> End screen call-to-action text --cta "Follow for more"
--broll Enable B-roll cutaway scenes --broll
--broll-images \x3Curls> Comma-separated screenshot/image URLs for B-roll --broll-images url1,url2
--template \x3Cslug> Script template (see below) --template saas-review
-g, --generate-script \x3Cprompt> AI-generate script from description -g "yoga mat product"
--product-url \x3Curl> Product URL for script generation context --product-url https://...
-s, --sync Wait for completion and print output URL --sync

Script Templates

Template Structure Best For
monologue Hook → Body → CTA Direct-to-camera talking
testimonial Problem → Solution → Result → CTA Customer stories
product-review Intro → Experience → Verdict → CTA Product reviews
problem-solution Hook → Pain → Solution → CTA Before/after pain points
saas-review Hook → Walkthrough → Opinion → CTA SaaS/app reviews
before-after Hook → Before → After → CTA Transformations
listicle Hook → Tip 1 → Tip 2 → Tip 3 + CTA Tips and lists
product-demo Intro → Demo → Recap → CTA Product walkthroughs

Subtitle Styles

Available styles: hormozi (yellow karaoke highlight, default), minimal, bold (neon cyan), karaoke (green pop), clean (white on dark), tiktok, neon

SaaS Review Videos

Generate a complete SaaS review video. ALL FOUR are required — do NOT skip any:

  1. Product name in the script (so it appears in subtitles)
  2. --actor (for talking head + lip sync)
  3. --broll --broll-images with 1-3 product screenshot URLs
  4. Script word count matching duration (2.5 words/sec)

Step-by-Step Flow (FOLLOW THIS EXACTLY)

  1. Get product name — ask user if not provided. STOP if missing.
  2. Get 1-3 screenshot URLs — ask user, or if they give a URL, visit the site and extract \x3Cimg> URLs showing the product dashboard/UI. STOP if no screenshots.
  3. Pick an actor — ask user or default to naomi or sofia. Run agent-media actor list if they want to browse.
  4. Write the script — MUST be ~25 words for 10s or ~37 words for 15s. Mention the product name 2-3 times. Count the words before submitting.
  5. Run the command with ALL required flags:
# CORRECT — descriptive filenames so images match the right scenes
agent-media ugc "Postiz is the best social media tool I've used. Postiz schedules across twenty-five platforms with AI. Try Postiz today." \
  --actor naomi --duration 10 --style hormozi --broll \
  --broll-images ./postiz-dashboard.png,./postiz-calendar.png --sync

# CORRECT — URLs with descriptive paths work too
agent-media ugc "Postiz is the best social media tool I've used. Postiz schedules across twenty-five platforms with AI. Try Postiz today." \
  --actor naomi --duration 10 --style hormozi --broll \
  --broll-images "https://postiz.com/dashboard-screenshot.png,https://postiz.com/scheduling-view.png" --sync

# WRONG — no actor, no screenshots, script too long
agent-media ugc "Here's how to schedule a post in Postiz step by step..." --sync

Review Flags

Flag Description Example
--actor \x3Cslug> AI actor (required) --actor naomi
--broll Enable B-roll scenes (required for reviews) --broll
--broll-images \x3Cpaths> Screenshot URLs or local files (required, 1-3) --broll-images ./img1.png,./img2.png
--duration \x3Cs> 5, 10, or 15 seconds --duration 10
--style \x3Cname> Subtitle style --style hormozi
--cta \x3Ctext> End screen text --cta "Try it free"
-s, --sync Wait for result (always use) --sync

Persona Management

Save voice + face combos for consistent UGC across videos:

# Create a persona from voice sample + face photo
agent-media persona create --name "brand-voice" --voice ./sample.mp3 --face ./photo.png

# List personas
agent-media persona list

# Use in UGC
agent-media ugc "Your script..." --persona brand-voice --sync

# Delete
agent-media persona delete \x3Cpersona-id>

Add Subtitles to Any Video

agent-media subtitle \x3Cvideo-path-or-job-id> --style hormozi --sync
agent-media subtitle ./my-video.mp4 --style bold --sync

Pricing

Plan Price Monthly Credits ~10s Videos
Creator $39/mo 2,500 ~3
Pro $69/mo 5,000 ~6
Pro Plus $129/mo 10,000 ~12

~800 credits per 10s video. Pay-as-you-go credit packs also available. Run agent-media credits to check balance.

Job Management

agent-media status \x3Cjob-id>     # Check job status
agent-media list                # List recent jobs
agent-media download \x3Cjob-id>   # Download output media
agent-media cancel \x3Cjob-id>     # Cancel and refund credits
agent-media retry \x3Cjob-id>      # Retry a failed job

Account

agent-media whoami              # Current user info
agent-media credits             # Credit balance
agent-media subscribe           # Manage subscription
agent-media login / logout      # Authentication

Checklist Before Every UGC Command

Before running ANY agent-media ugc command, verify:

  • --actor is included (no actor = no talking head = broken video)
  • Script word count matches duration (count words: 5s=12, 10s=25, 15s=37)
  • --sync is appended
  • For SaaS reviews: --broll --broll-images with 1-3 screenshot URLs
  • For SaaS reviews: product name appears 2-3 times in script
  • Credits are sufficient (agent-media credits)
安全使用建议
This skill is an instruction-only integration for a third‑party CLI and is coherent with its stated purpose, but before installing or using it you should: (1) inspect the agent-media npm package and GitHub repository to confirm authenticity and review what the CLI does; (2) avoid uploading any sensitive local files — the CLI auto-uploads files passed to --broll-images, and the skill instructs fetching arbitrary product pages; (3) prefer running the CLI in an isolated environment (throwaway VM or container) if you must evaluate it; and (4) require explicit user confirmation before the agent fetches remote pages or uploads local files. If you need higher assurance, ask the skill author for a reproducible audit trail or a vetted binary/package source.
功能分析
Type: OpenClaw Skill Name: agent-media Version: 1.0.1 The skill is classified as suspicious due to instructions that enable risky capabilities and potential vulnerabilities. Specifically, the `SKILL.md` instructs the AI agent to 'visit the site yourself and extract image URLs from the page' from arbitrary URLs, posing a potential Server-Side Request Forgery (SSRF) or web scraping risk. Additionally, several `agent-media` commands (e.g., `--broll-images`, `persona create`, `subtitle`) explicitly accept and 'auto-upload' arbitrary local file paths, creating a risk of local file exfiltration if an attacker can control the path. There's also a general risk of command injection given the construction of shell commands from user-provided text.
能力评估
Purpose & Capability
The name/description match the instructions: SKILL.md tells the agent how to run the agent-media CLI to produce UGC videos, lists the flags and pipeline, and enforces domain-specific rules (actors, duration, b-roll). Required resources declared as none align with an instruction-only skill.
Instruction Scope
Instructions stay within the stated purpose (install CLI, login, run agent-media ugc with flags). However the guide explicitly tells the agent to 'visit the site yourself and extract image URLs' and to accept and auto-upload local files supplied via --broll-images. That behavior is expected for a media tool, but it grants the agent discretion to fetch arbitrary web pages and transmit local files — a potential data‑exfiltration surface if the agent has broad filesystem/network access. The SKILL.md does not impose explicit safeguards (e.g., require user confirmation before uploading local files or fetching non-public URLs).
Install Mechanism
There is no install spec in the skill bundle (lowest static risk). The instructions direct users/agents to run npm install -g agent-media-cli (an external npm package). Installing that package executes third‑party code outside the skill; this is expected for a CLI-based integration but is an additional trust step the user should verify (inspect the npm package/github).
Credentials
The skill declares no required environment variables or credentials. It does instruct performing an interactive 'agent-media login' OTP flow; that is reasonable for a CLI that uploads media. There are no unrelated credentials requested in SKILL.md.
Persistence & Privilege
always is false and the skill is user-invocable; it does not request permanent platform presence or modify other skills. Normal autonomous invocation is allowed by platform defaults but not excessive here.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agent-media
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-media 触发
  4. 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
v1.0.1
agent-media 1.0.1 - Overhauled documentation to focus on AI-driven UGC (User-Generated Content) and SaaS review video production via CLI. - Added strict mandatory rules for UGC video creation, including requirements for actor, script word count, B-roll with descriptive filenames, product name in script, and always using `--sync`. - Expanded and detailed UGC pipeline commands, flags, and templates for versatile video generation. - Provided step-by-step instructions and examples for generating SaaS review videos, emphasizing completeness and quality. - Updated tips and usage patterns reflecting the new feature set and workflow.
v1.0.0
agent-media 1.0.0 — Initial Release - Generate AI-powered videos and images from your terminal using the agent-media CLI. - Supports multiple state-of-the-art models for text-to-video, image-to-video, and text-to-image generation. - Includes commands for media generation, job management, credits, account settings, and model/pricing info. - CLI offers a variety of flags for customization and output formatting. - Subscription and credit purchasing integrated via Stripe checkout. - Simple authentication and API key management.
元数据
Slug agent-media
版本 1.0.1
许可证
累计安装 7
当前安装数 7
历史版本数 2
常见问题

Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI. 是什么?

AI UGC video production from the terminal using the `agent-media` CLI. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 951 次。

如何安装 Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI.?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-media」即可一键安装,无需额外配置。

Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI. 是免费的吗?

是的,Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI. 完全免费(开源免费),可自由下载、安装和使用。

Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI. 支持哪些平台?

Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI. 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 Agent Media: Generate AI-powered videos and images from the terminal using the `agent-media` CLI.?

由 nevo-david(@nevo-david)开发并维护,当前版本 v1.0.1。

💬 留言讨论