Architecture Overview · Park Whisperer Platform

Script-to-Published Video Pipeline

Fully automated video production system that takes a 36–42 word Claude-generated script and produces a published 1080×1920 Reels video with word-pop captions, semantic color-coding, and synchronized ElevenLabs narration — zero manual steps.

● Production · 3 PlatformsElevenLabs + FFmpeg + Claude0 Manual Steps

Click any component to explore it in the full case study.

Input A
Claude REELS Script
10-field structured block from Pass 3: HOOK, SEGMENT_1, CTA, NARRATION (36–42 words enforced), platform captions. NARRATION feeds TTS; all fields feed caption system.
36–42 word contract
Input B
Park Background Images
9:16 aspect ratio images from Azure Blob Storage. Per-content-type image selection (rope_drop vs. weather vs. morning briefing).
Azure Blob · 9:16 images
Input C
Background Music Track
Per-content-type music selection. Energy and tempo matched to content type and time of day.
Per-pipeline audio
5-Stage Video Production Pipeline
Stage 1
Number Normalize
Digits → words for TTS ("60" → "sixty"). Prevents robotic pronunciation in narration.
›
Stage 2
ElevenLabs TTS
NARRATION → MP3 + per-character timestamps (eleven_multilingual_v2). Word-level timing map generated.
›
Stage 3
Caption Align
Word-level timing from ElevenLabs → ASR map (spoken words back to written). Semantic color-coding applied.
›
Stage 4
Frame Compose
Park images + word-pop captions + brand overlays. Per-content-type render profiles (motion, transitions, effects).
›
Stage 5
FFmpeg Encode
1080×1920 @ 30fps. CRF 18. Min 3500k bitrate. Duration-matched to TTS audio. Thumbnail extracted.
🎬
Quality Gates + Safety Checks
Word count (36–42) enforced pre-render. _must_skip_social() check before any platform API call. PREVIEW_MODE and PIPELINE_DRY_RUN gates for staging. Video quality validated (frame count, audio sync).
Word count gate (36–42)_must_skip_social() checkPREVIEW_MODE / DRY_RUNCaption color validationAudio sync verification0 manual steps
Output A
1080×1920 MP4 Video
9:16 Reels/TikTok/YouTube Shorts format. Word-pop captions with semantic color-coding: purple (numbers), mint (ride names), coral (status words). Brand overlay included.
1080×1920 · 30fps · CRF 18
Output B
Per-Platform Captions
IG_CAPTION, TIKTOK_CAPTION, YT_TITLE, YT_DESCRIPTION, YT_TAGS. Platform-native character limits and hashtag strategy applied per publish target.
Instagram · TikTok · YouTube
Output C
Thumbnail + Publish
Extracted thumbnail frame. Published to Instagram Reels, TikTok, YouTube Shorts conditionally. All gated by _must_skip_social() check.
Auto-published · 3 platforms
5
AI systems (Claude + ElevenLabs + Pillow + FFmpeg + Blob)
3
Caption colors (semantic)
10+
Render profiles
0
Manual steps
StackPythonElevenLabs eleven_multilingual_v2Pillow (PIL)FFmpegAzure Blob StorageClaude Sonnet (script)Instagram APITikTok APIYouTube API