Video Content Pipeline

Generate social media content: static image cards, animated chat-bubble videos, AI video clips, and full composite videos with music.

Quick Reference

Content Formats

1. Static Chat Card (Image Post)

The signature format: fake Telegram-style chat between Rees and Atlas.

1. Write HTML using chat-card template (assets/chat-card-template.html)
2. Screenshot with Puppeteer (scripts/generate-video.mjs → screenshotCard())
3. Compress: `convert output.png -quality 90 -resize 1200x output.jpg`
4. Post via GetLate API

2. Animated Chat Video

Messages appear one by one with typing indicators.

1. Write HTML using vertical template (assets/chat-card-vertical.html) 
2. Add id="msg1", id="msg2" etc to messages, id="typing-1" etc to indicators
3. Generate frames with generateChatFrames()
4. ffmpeg stitches to MP4

3. Full Composite Video

AI intro → animated chat → AI outro, optionally with music.

1. Generate AI clips via KIE (generateKieVideo + pollKieTask)
2. Generate chat frames in parallel
3. Stitch with stitchVideo()
4. Optionally add music with ffmpeg audio mixing

Pipeline Script

Main script: scripts/generate-video.mjs

Exported functions:

• screenshotCard(htmlPath, outputPath) - Static image
• generateChatFrames(htmlPath, outputDir, fps) - Animated frames
• generateKieVideo(apiKey, prompt, {model, aspectRatio}) - AI video
• pollKieTask(apiKey, taskId, maxWaitMs) - Wait for AI video
• stitchVideo({introPath, chatVideoPath, outroPath, outputPath}) - Combine clips
• postToX(apiKey, accountId, content, mediaPath) - Post to Twitter

KIE API

Base: https://api.kie.ai/api/v1 | Auth: Bearer token

For model details, parameters, and endpoints:

• Video models: Read references/video-models.md - Veo 3.1, Runway Gen-4, Kling, Wan, Bytedance
• Audio models: Read references/audio-models.md - Suno V3.5-V5
• Image models: Read references/image-models.md - Imagen 4, FLUX 2, Grok Imagine, GPT Image

Recommended Models

• Video (social clips): veo3_fast - $0.40/8s, 720p, includes audio
• Video (high quality): veo3 - higher fidelity, slower
• Music: Suno V5 via /api/v1/generate
• Images: Nano Banana Pro (default, best quality) — FLUX 2 is backup only

Aspect Ratios

• X/Twitter feed: 16:9 or 1:1
• Instagram Reels / TikTok / Stories: 9:16
• Profile pic: 1:1 (400x400+)
• X banner: 3:1 (1500x500)

Posting

GetLate API: https://getlate.dev/api/v1/posts

• Auth: Bearer token
• Account ID for X: 698c65144525118cee8bc851
• Supports image and video media items
• Upload media to tmpfiles.org first, use /dl/ URL

HTML Template Structure

Chat cards use this CSS class structure:

• .chat - outer container (dark bg)
• .header - Atlas avatar + name
• .msg - individual message bubble
• .msg-right / .msg-left - alignment
• .msg-name.rees / .msg-name.atlas - colored names
• .typing - typing indicator with .dot children
• .visible - class added to show elements (for animation)

For animation, messages start with opacity: 0 and get .visible added sequentially.

Writing Style (X Posts)

Follow these rules for post copy:

• Lead with insight, not features
• Write like telling a mate something interesting
• Self-aware, not self-congratulatory
• No numbered feature lists disguised as insights
• No "So I built" energy - more "here's what surprised me"
• If it sounds like a LinkedIn post, bin it
• Captions: simple, not cringy. Let the card do the work.

ffmpeg Cheat Sheet

# Resize to vertical
ffmpeg -i input.mp4 -vf "scale=720:1280:force_original_aspect_ratio=increase,crop=720:1280" output.mp4

# Fade in/out
-vf "fade=t=in:st=0:d=0.5"  # fade in first 0.5s
-vf "fade=t=out:st=3.5:d=0.5"  # fade out at 3.5s

# Concatenate clips
echo "file 'a.mp4'\nfile 'b.mp4'" > list.txt
ffmpeg -f concat -safe 0 -i list.txt -c copy output.mp4

# Add music under video
ffmpeg -i video.mp4 -i music.mp3 -filter_complex "[1:a]volume=0.3[bg];[bg]adelay=0|0[delayed]" -map 0:v -map "[delayed]" -shortest output.mp4

# Strip audio
ffmpeg -i input.mp4 -an output.mp4

# Add text overlay
ffmpeg -i input.mp4 -vf "drawtext=text='hello':fontsize=48:fontcolor=white:x=(w-tw)/2:y=h-60" output.mp4

Dependencies

Required on the system:

• puppeteer (npm) - screenshots and frame capture
• ffmpeg - video stitching, audio mixing, effects
• imagemagick (convert) - image compression/resizing
• curl - API calls and file uploads