A fully automated ad creative factory. It pulls campaign briefs from Airtable, cross-references them against a benchmark database of top-performing ads (CTR, ROI, impressions, CPC), and uses AI to generate new ad creatives inspired by the best performers.
The system analyzes what makes high-performing ads work — their visual style, hook patterns, and copy structure — then mutates those winning elements into fresh creatives. Each generated ad is uploaded to Google Drive, marked as done in Airtable, and a Telegram notification is sent when the batch is complete.
The workflow starts by reading pending records from the Airtable Creative Factory table — filtering for entries with status "Generate". Each record contains the brief: what product to advertise, target audience, and creative direction.
For each brief, the pipeline looks up reference benchmarks in Airtable — top-performing ads ranked by CTR, ROI, impressions, and CPC. It extracts the "visual DNA" from these winners: hook seeds, creative patterns, messaging angles, and visual style elements.
A system prompt and user prompt are dynamically built by combining the creative brief with the extracted benchmark data. The prompt instructs the AI to generate ad creatives that borrow proven elements from top performers while staying fresh and unique.
OpenAI and Google Gemini generate the ad images. Each image is uploaded to a designated Google Drive folder, the Airtable record is updated with the image URLs and marked as "Done", and a Telegram notification fires with the batch results.
Doesn't generate blind — uses real performance data from past campaigns. High-CTR, high-ROI ads become the blueprint for new creatives, so every generation is informed by what actually works.
Extracts winning visual patterns — color schemes, composition styles, hook angles, and copy structures — then combines and mutates them into new creative variations that feel fresh but carry proven DNA.
Processes creatives in configurable batches with built-in rate limiting and wait nodes. Handles errors gracefully — failed items are logged separately without stopping the entire pipeline.
From brief to final asset — Airtable triggers the workflow, AI generates the images, Drive stores them, Airtable gets updated, and Telegram notifies the team. Zero manual steps in the loop.