An automated YouTube Shorts system — n8n workflow, host scripts, and a 63-page manual — refined during a documented 35-day production run, with the method to rebuild it for your own channel using AI as a collaborator.
For operators who can run SSH, Docker, and n8n — and who want a documented system to adapt, not a one-click product. No channel-growth or revenue guarantees; the system is what you receive.
The operating channel and its analytics stay private. The product mechanics do not. The manual pages, workflow canvas, script excerpts, and package tree below come directly from the shipped files.
The shipped manual pages, the actual n8n workflow canvas, exact script excerpts, and the full package contents.
Views, revenue, subscriber growth, or any outcome for your channel. This is a system, not a results promise.
Direct page previews from the shipped manual — structure, implementation, and the kind of operational detail you only write after living with the system. Click any page to read it.
Annotated map — redrawn from the shipped 14-node graph to keep the full flow readable:
Delivered inactive and with credentials removed, so you can inspect and configure it before enabling the schedule.
Validates the job and composes the final vertical MP4. Refuses to run while the default channel placeholder remains.
# breakage if a malformed id ever reaches this script. $WORK_DIR ends up in # `rm -rf` later, so this guard matters. if ! [[ "$JOB_ID" =~ ^[A-Za-z0-9_-]+$ ]]; then echo "ERROR: Invalid job_id: $JOB_ID" exit 1 fi # Refuse to render with the default channel placeholder — otherwise every # generated Short would carry the literal text "YOUR CHANNEL", then # upload itself to YouTube. See Chapter 4 for setup. if [ "$CHANNEL_NAME" = "YOUR CHANNEL" ]; then echo "ERROR: CHANNEL_NAME is still the placeholder." exit 1 fi
Watches for completed job folders and controls execution. Drains the whole queue on each event so a render never strands a job.
process_pending_jobs() { for f in "$JOBS_DIR"/*.json; do [ -f "$f" ] || continue JOB_ID=$(basename "$f" .json) # Skip malformed job ids early so they don't reach # process_job.sh as confusing FAILED entries. if ! [[ "$JOB_ID" =~ ^[A-Za-z0-9_-]+$ ]]; then echo "[$(date)] SKIPPED invalid job_id: $JOB_ID" >> "$DAEMON_LOG" continue fi "$SCRIPT" "$JOB_ID" >> "$DAEMON_LOG" 2>&1 done }
Handles authenticated YouTube uploads after composition. The comment is the operator's own note from the production run.
# Reference build note: # chunksize=-1 uploads the MP4 in one request. This worked # for the 35-day production run, but very large files or # unstable networks may benefit from a fixed chunk size. # This script intentionally does not retry; the caller # decides policy. media = MediaFileUpload( file_path, chunksize=-1, resumable=True, mimetype="video/mp4", )
The manual explains the decisions; the scripts and workflow are the implementation; Chapter 8 is the part you reuse across channels. Everything below arrives as one download, with updates provided over the following six months.
Infrastructure, AI pipeline, composition, upload, operational safety, pitfalls, and adaptation. Shipped as PDF and a searchable interactive HTML version.
The FFmpeg composition script, the watcher daemon, and the Python YouTube uploader — the exact files, with configuration guards built in.
Importable JSON, 14 nodes, credentials removed and delivered inactive so you configure it deliberately before the first run.
Standalone HTML you can open in any browser — the same explainers used to reason about the system.
The reusable part: what to keep, what to replace, and how to use AI as a collaborator to rebuild the reference for your own concept. This is the product — it's Chapter 8 of the manual above, not a separate file.
Not part of the initial download — fixes for breakage from dependency and platform changes, provided during your six-month update window. After that, you have the method to adapt yourself.
n8n orchestrates generation; the filesystem hands off to host scripts; FFmpeg composes; the Python uploader publishes. You check a log if you want to.
Each run produces five cards — one generated image and one narration clip per card — composed into an 18–22 second, 1080×1920 Short. A 100-topic pool with 60-day anti-repetition keeps daily topics from repeating. Running cost is roughly $6.69/month (about $5 VPS + API), documented in the manual's cost breakdown.
The reference build is a case study. The method is the product.
Most automation packages hand you a template to copy. This one hands you a working example — and a documented way to rebuild it for your own channel, with AI as a collaborator rather than an author. Chapter 8 is where that lives.
AI handles structured transformations, known patterns, and cause hypotheses. Aesthetic identity, platform intuition, and the last debugging step stay yours. The chapter gives you a framework for telling them apart.
Layout safety, multi-line alignment, dynamic length, identity mismatch, audio sync, title wrap, visual breathing room. When you adapt, you'll hit several — pre-mapped so you don't rediscover them on day twelve.
A real multi-stage cascade documented end to end — deployment miss, field omission, preview-vs-runtime mismatch, retry storm — each with the guard that catches it.
Copy-paste meta-prompts for concept definition, image-prompt adaptation, script tone, and incident diagnosis — with the verified-not-guessed principle baked into each.
The files document one reference build. The method helps you reconstruct it for your own channel concept.
Reading both columns before buying saves everyone a refund conversation. The narrow fit is deliberate, not an accident.
No subscription, no upsells. The Founding price applies during the initial release window and reflects the package as it exists today.
One-time. Lifetime download access. 6 months of compatibility updates.
Instant download of all files after purchase, via Lemon Squeezy. Support is best-effort by email, no SLA. A reference implementation, not a managed service. No performance guarantees — the system is what you receive; your channel's results are your work.
Sales availability: United States, Canada (excluding Québec), Australia, New Zealand, and Singapore only at launch. Other jurisdictions are not currently offered, even if the checkout interface technically displays them.
Documented from a live daily production run. Inspect the manual, the workflow, and the shipped code above — then decide.
This is a mockup. On the live site this opens the Lemon Squeezy checkout for the $349 Founding package.