agent-media

Imported from yuvalsuede/agent-media-skill (SKILL.md). Install upstream with npx skills add yuvalsuede/agent-media-skill. Copyright stays with the author.

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, Product Acting UGC, 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 — PICK A RANDOM ONE

Every UGC video MUST include --actor <slug>. 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:

1. Run agent-media actor list to get the full list of available actors
2. Pick a random actor from the list — do NOT always use the same one (e.g., don't always default to sofia or naomi)
3. Tell the user which actor you picked and suggest they can browse with agent-media actor list or pick a specific one with --actor <slug>

NEVER hardcode a default actor. Every video should feel different — variety in actors is key to quality UGC content.

Rule 2: ALWAYS count words — 2.5 words per second

Natural speech is 2.5 words/second. Scripts MUST match the target duration exactly — too many words sounds robotic, too few words creates awkward pauses and silence:

• 5s video → 10–12 words (not fewer!)
• 10s video → 22–25 words (not fewer!)
• 15s video → 33–37 words (not fewer!)

CRITICAL: Count the words before submitting. If the script is too short, ADD more content. If too long, CUT words. A 15-word script on a 10s video = 5 seconds of dead silence. A 50-word script on a 10s video = rushed robotic speech. Both are broken.

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

PIP Mode (Picture-in-Picture)

PIP mode creates a full-frame talking head with rotating B-roll image overlays in the lower portion. The actor speaks to camera while relevant visuals slide in and out below. Subtitles appear just above the overlay.

# Basic PIP video — actor speaks to camera with auto-generated B-roll overlays
agent-media ugc "Stop scrolling. If you struggle to grow on social media, consistency beats perfection every time." \
  --actor adaeze --pip --duration 15 --style hormozi --sync

# PIP with specific style
agent-media ugc "Three things I wish I knew before starting my business..." \
  --actor sofia --pip --duration 10 --style tiktok --sync

PIP Options:

# PIP with bottom-right overlay, large size, slide-left animation
agent-media ugc "Your script here..." \
  --actor adaeze --pip --pip-position bottom-right --pip-size large --pip-animation slide-left --sync

# PIP with rounded overlay and scale animation
agent-media ugc "Your script here..." \
  --actor sofia --pip --pip-style rounded --pip-animation scale --duration 10 --sync

PIP Rules:

• --actor is required (PIP needs a talking head)
• Max duration is 15 seconds (longer videos are split into ≤10s clips with seamless continuity)
• B-roll overlays are auto-generated from the script — no --broll-images needed
• Script word count rules apply: 2.5 words/sec (15s = ~37 words max)
• Do NOT combine --pip with --broll — they are separate modes

Script Templates

Rule 7: ALWAYS use --style — PICK A RANDOM ONE

Every UGC video MUST include --style <name>. Without a style, you get the same hormozi subtitle every time — boring and repetitive.

If the user hasn't specified a style:

1. Pick a random style from the list below — do NOT always default to hormozi
2. Tell the user which style you picked

NEVER hardcode a default style. Variety in styles makes each video feel unique.

Subtitle Styles (17 styles)

Popular

Bold & Energetic

Aesthetic & Soft

Colorful

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 <img> URLs showing the product dashboard/UI. STOP if no screenshots.
3. Pick an actor — ask user or pick a RANDOM one from agent-media actor list. Never always use the same one.
4. Write the script — MUST be 22-25 words for 10s or 33-37 words for 15s. Too few words = awkward pauses. Too many = robotic. 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

Product Acting UGC

Generate creator-style product-in-hand UGC from a product image, actor, scenario template, and short spoken script. Use this when the user wants an actor to hold, react to, smell, present, or demonstrate a physical product.

Requirements:

1. Product image URL — must be public HTTPS, or upload a local product image first and use the returned URL.
2. Actor — --actor <slug> is required. Run agent-media actor list if the user has not chosen one.
3. Script or product context — provide either --script for exact words or --about so the API can generate a short script.
4. Word count — script is capped at 3 words/second × duration (≤15 words at 5s, ≤30 at 10s, ≤45 at 15s).

CLI

agent-media product-acting \
  --product-image https://cdn.example.com/product.png \
  --actor sofia \
  --about "A premium perfume with a warm vanilla dry-down" \
  --template product-in-hand \
  --acting-style honest-review \
  --duration 5 \
  --sync

agent-media product-acting \
  --product-image https://cdn.example.com/product.png \
  --actor naomi \
  --script "I did not expect this perfume to smell this expensive." \
  --template car-selfie \
  --acting-style shocked \
  --sync

REST API

curl -X POST https://api-v2-production-2f24.up.railway.app/v1/generate/product_acting_ugc \
  -H "Authorization: Bearer ma_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "product_image_url": "https://cdn.example.com/perfume.png",
    "actor_slug": "sarah",
    "product_name": "Rose Noir",
    "product_description": "Premium rose perfume with a warm vanilla dry-down.",
    "template": "product-in-hand",
    "acting_style": "honest-review",
    "duration": 5,
    "subtitle_style": "hormozi"
  }'

Flags

Credit cost: 30 × duration + 50 credits, plus 5 credits when the API generates the script. Runtime: usually several minutes.

Show Your App Videos

Generate a video where an AI actor holds a phone that displays your app screenshot and reads your script, with Hormozi-style word-by-word subtitles burned in.

Requirements (enforced server-side):

1. Vertical app screenshot — PNG, JPEG, or WebP, height > width (phone portrait). The API rejects landscape.
2. Public URL — screenshot must be reachable. Host on R2, S3, your own CDN, etc. Local files are not accepted by the CLI for this command.
3. Script word count — capped at 3 words/second × duration (≤15 words at 5s, ≤30 at 10s, ≤45 at 15s).

CLI

# Random actor (recommended — variety is key)
agent-media show-your-app \
  --app-screenshot https://cdn.example.com/my-app.png \
  --script "You really need to try this app — it generates UGC videos in seconds." \
  --duration 5 --sync

# Specific actor
agent-media show-your-app \
  --app-screenshot https://cdn.example.com/my-app.png \
  --script "Try this app, it changed everything for me." \
  --actor sarah --duration 10 --sync

REST API

curl -X POST https://api-v2-production-2f24.up.railway.app/v1/generate/show_your_app \
  -H "Authorization: Bearer ma_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "app_screenshot_url": "https://cdn.example.com/my-app.png",
    "script": "You really need to try this app.",
    "duration": 5
  }'

Flags

Credit cost: 75 flat. Runtime: 4–8 minutes (GPT Image + Seedance 2.0).

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 <persona-id>

Add Subtitles to Any Video

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

Pricing & Credit Deduction

Plans

Pay-as-you-go credit pack: 3,900 credits for $39 (one-time purchase, never expires).

How credits are deducted

Rate: 30 credits per second of video. 1 credit = $0.01.

Deduction order: Monthly credits are used first (they expire at period end), then purchased credits (never expire).

Refunds: If the video generation fails, credits are automatically refunded. Canceled jobs are also refunded.

Run agent-media credits to check balance before generating.

Job Management

agent-media status <job-id>     # Check job status
agent-media list                # List recent jobs
agent-media download <job-id>   # Download output media
agent-media cancel <job-id>     # Cancel and refund credits
agent-media retry <job-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

REST API (v2)

agent-media also has a REST API for programmatic access. Interactive docs at https://agent-media.ai/docs/api-reference

Endpoints

Auth

Authorization: Bearer ma_YOUR_API_KEY

SDKs

TypeScript SDK: npm install @agentmedia/sdk

Python SDK: pip install agent-media

curl Example

# Generate a video
curl -X POST https://api-v2-production-2f24.up.railway.app/v1/generate/ugc_video \
  -H "Authorization: Bearer ma_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "script": "Your 50+ character script here...",
    "actor_slug": "sofia",
    "tone": "energetic"
  }'

# Check status
curl https://api-v2-production-2f24.up.railway.app/v1/videos/{job_id} \
  -H "Authorization: Bearer ma_YOUR_KEY"

MCP Server (Claude Code / Cursor) — coming soon

{
  "mcpServers": {
    "agent-media": {
      "command": "npx",
      "args": ["@agentmedia/mcp-server"],
      "env": { "AGENT_MEDIA_API_KEY": "ma_xxx" }
    }
  }
}

OpenAPI Spec

Download at https://agent-media.ai/openapi.json — import into Postman, Insomnia, or any OpenAPI tool.

Checklist Before Every UGC Command

Before running ANY agent-media ugc command, verify:

• --actor is included (no actor = no talking head = broken video). Pick a RANDOM actor if user didn't specify one!
• --style is included. Pick a RANDOM style if user didn't specify one — never always use hormozi!
• Script word count matches duration EXACTLY (5s=10-12, 10s=22-25, 15s=33-37 words — too few = pauses, too many = robotic)
• --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
• For PIP mode: --pip flag is set, --actor is included, duration ≤ 15s
• For PIP mode: do NOT also pass --broll (PIP auto-generates overlays)
• For Product Acting: --product-image, --actor, and either --script or --about are included
• Credits are sufficient (agent-media credits)