All docs

AI Content Creator (Agent Mode)

Updated 2026-04-07

Overview

Flagship feature. A multi-step AI workflow that transforms a YouTube/Instagram/TikTok channel into a video production pipeline. Users connect a channel, AI analyzes content themes, generates ideas, writes scene-by-scene scripts, generates video assets via Fal.ai, and exports platform-ready videos.

User Stories

  • As a creator, I can connect my YouTube channel so the AI understands my content style
  • As a creator, I can generate 12+ video ideas based on my channel's themes and engagement patterns
  • As a creator, I can get multiple script variations with scene breakdowns for any idea
  • As a creator, I can edit scenes inline (dialogue, visuals, duration, reorder, delete)
  • As a creator, I can generate video/image assets for each scene with one click
  • As a creator, I can retry failed scene generations
  • As a creator, I can export the final video for YouTube (16:9), Instagram (1:1), or TikTok (9:16)

Workflow Steps

Channel Setup → Asset Upload (optional) → Idea Generation → Script Generation → Canvas Editor → Preview & Export

Step 1: Channel Setup

  • Connect YouTube/Instagram/TikTok via OAuth or similar-channel URL
  • Manual channel setup with custom description and target audience
  • Auto-analysis extracts: themes, visual style (color scheme, editing pace), engagement patterns (top topics, avg watch duration)
  • Fetches up to 50 recent videos with metrics

Step 2: Asset Upload (Optional)

  • Upload reference images to guide visual generation
  • Select 1-3 channel videos for transcript context

Step 3: Idea Generation

  • AI generates 3-20 ideas per batch (default 12)
  • Each idea: title, description, platform, estimated engagement (0-100), suggested duration, tags (max 8)
  • Duration auto-constrained by platform: YouTube 60-3600s, Instagram 15-90s, TikTok 15-180s
  • Manual idea creation also supported
  • Search and filter ideas

Step 4: Script Generation

  • Generates 2-5 script variations per idea
  • Each variation has 4-5 scenes: hook → content → content → content → CTA
  • Each scene: dialogue (voiceover text), visual description (Fal.ai prompt), duration, scene type
  • User preferences: video type (animated/live-action/mixed), visual style (realistic/cartoon/cinematic), voice, pacing, music, text overlays
  • Inline editing of all scene fields

Step 5: Canvas Editor

  • Visual scene layout with scene nodes showing order, status, duration
  • Trigger batch generation (all scenes) or individual scene generation
  • Track progress per scene (0-100%)
  • Scene statuses: queued → processing → completed/failed
  • Advanced generation modal: custom prompts, model selection, duration override
  • Reference image guidance for asset generation

Step 6: Preview & Export

  • Full preview player with playback controls
  • Scene navigation
  • Export: select platform → auto aspect ratio + user resolution (720p/1080p/4K)
  • Download final composite video