Growth Creative Factory is an open-source (MIT) "creative production factory" for advertisers: input your ad performance data, receive new copy written by AI, and a set of brand-appropriate images for every display position. Create in bulk, all with a single command.

Problem Being Solved

Creative is the biggest leverage of social advertising but also the slowest process. Each ad requires a copywriter, designer, and resizing for each placement, so teams typically only test a few variations each week. GCF wraps that whole loop into one pipeline.

Main Features

• Find underperforming ads: upload a CSV, Google Ads, or Meta Ads, automatically flag underperforming ads based on CTR / CPA / ROAS, and diagnose the cause using the ad's own data.
• Create new copy: Claude (or an available offline copywriter) writes headlines/descriptions based on 5 approaches: benefit, urgency, social proof, problem/solution, and curiosity. Automatically checks character limits, filters out policy-violating words, and removes duplicates.
• Generate images instead of spreadsheets: 6 design templates × 5 formats (Feed 1:1, Feed 4:5, Story/Reels 9:16 with safe zone, Link 1.91:1, 16:9), customized brand kit via YAML file. One demo run: 5 underperforming ads turn into 75 images, rendering in about 16 seconds on a 4-core laptop.
• English – Vietnamese bilingual: accompanying font fully supports Vietnamese accents, CTAs, and badges in both languages.
• Easy review and delivery: gallery.html page filters by format/template, contact sheet, TSV file for Figma plugin, push to Google Sheets. Test results are fed back into memory to avoid previously lost approaches in the next cycle.

Who Should Use It

Performance marketers, agencies, and growth teams running Facebook/Instagram/TikTok/Google Display who need to test many creatives; online shops and startups wanting a week’s worth of social posts from a product brief without needing a designer.

How to Get Started

Runs on Windows, macOS, Linux (Python 3.10–3.13). Clone the repo, run pip install -e ".[ui]" then gcf run --input examples/ads_sample.csv --mode dry. Dry mode runs offline, no API key needed; for sharper copy, add ANTHROPIC_API_KEY and switch to --mode live. Includes CLI, Streamlit app (Creative Studio + Ads Wizard), and Python API.

What Makes It Different

Unlike generic AI image generation tools, GCF starts from real effectiveness data, has guardrails to comply with advertising policies, and outputs ready-to-upload images for each placement. Renders using pure Python (Pillow), no browser or Figma needed, data resides on your machine. All demo images in the README are rendered directly from the repo and can be recreated with the command gcf demo.