<aside> 🎬

One form to finished AI UGC ads.

Import this n8n workflow, connect your keys, and every submission returns up to 3 finished, captioned, 9:16 UGC style video ads: AI creator avatar, natural voiceover, lip synced video, product b roll, word by word captions. Delivered to Google Drive, logged in an Ad Library sheet, and emailed to you.

Roughly $1 to $2.50 in API fees per finished ad. A UGC creator typically invoices $150 to $300 for the same deliverable and takes days.

</aside>

<aside> ⬇️

The workflow file

AI_UGC_Factory_v1.json

It has been scanned and is safe to publish. No API keys, no email addresses, no real Drive or Sheet IDs. Credentials point at readable placeholders (OPENAI, HIGGSFIELD, JSON2VIDEO, GDRIVE, GSHEETS, GMAIL, ELEVENLABS) rather than anyone's real n8n credential IDs, and the Config node already ships with REPLACE_WITH_DRIVE_FOLDER_ID and REPLACE_WITH_GOOGLE_SHEET_ID in place.

69 functional nodes plus 11 sticky notes, 10 stages. Built for on demand use: it runs when you submit the form, not on a schedule. Works on n8n Cloud or self hosted, current version.

</aside>

What it actually does

1. Intake. A hosted n8n form: product, audience, pain, offer, optional product page URL and product image, how many variants (1 to 3).
2. Page mining. If you gave a product URL, it fetches and strips the page so the script agent works from your real positioning, not guesses.
3. Creative Director (GPT-5.5). Writes 5 distinct ad concepts: hook, 15s body script, 10s CTA script, avatar casting brief, b roll shot, caption. Each concept attacks a different angle. Hard rule baked into the prompt: no invented stats, no fake claims.
4. Hook Scorer (GPT-5 mini). Scores all 5 for scroll stop, native feel and clarity. Only the top N go to video. You never pay to render a weak script.
5. Avatar (Higgsfield Soul). Photoreal creator, amateur selfie aesthetic, 9:16, 1080p.
6. Voice. OpenAI TTS by default, fewest keys, or flip one Config switch to ElevenLabs for the most realistic delivery. ElevenLabs returns raw PCM, so the workflow wraps it into a WAV in flight with a Code node, because Higgsfield only accepts WAV. No ffmpeg, no extra service.
7. Lip sync (Higgsfield Speak v2). Two passes per ad: the 15s hook and body segment, and the 10s CTA close. Same avatar, same voice, full continuity.
8. B roll (Higgsfield DoP). Optional. Animates your product photo into a moving product shot, cut between the talking segments. If it fails, the ad still ships without it.
9. Assembly (JSON2Video). Stitches hook, b roll and CTA at 1080x1920 and burns in word by word TikTok style captions, auto transcribed from the voiceover.
10. Delivery. MP4 saved to your Drive folder, a row per ad in your Ad Library sheet with scripts, captions, scores and estimated cost, and a summary email with links.

<aside> 🛡️

Every generation step is async with polling, timeouts and per variant failure isolation. One failed variant never kills the batch.

</aside>

The economics per finished ad