Aimee
One photo in, finished ad creative out.
A production computer-vision system that turns a single product or lifestyle photo into ready-to-run social ad creative, in about the time it takes to make a coffee.
A finished ad set generated by Aimee, in feed and story formats.
Most brands sit on a folder of decent product photos and no fast way to turn them into ads. Aimee closes that gap. Feed it one image and it decides where text can safely go, writes the copy in the brand voice, and renders finished creative in the three sizes social platforms actually want.
The hard part is placement. A roughly 14,000-line vision engine reads each image the way a designer would, finds the faces and the busy areas, and only drops copy where it will not cover something that matters.
- Classifies each image (clean catalog shot vs. real-world lifestyle scene) and routes it down the right pipeline.
- Computes face-aware and saliency-aware safe zones so a headline never lands on a face or on text already in the photo.
- Writes tiered ad copy across several hook angles, tuned to a learned brand voice, with a loop that detects and rewrites generic AI phrasing.
- Renders finished 1080px creative in 1:1, 4:5, and 9:16, streams progress live to the browser, and packages everything as a downloadable ZIP.
- Batch mode runs an entire product catalog in one pass.
- Per-account isolation, sign-in, and metered billing are built in.
A Python and FastAPI backend orchestrates a numbered pipeline (detect, then saliency, then layout, then copy, then render) behind a React single-page app. Computer vision runs on OpenCV, NumPy, and SciPy. Copy and image understanding run on GPT-4o. Typesetting uses a shrink-to-fit engine that guarantees a fixed safe margin, and the whole thing ships as a Docker image.
- Python
- FastAPI
- React
- OpenCV
- NumPy / SciPy
- GPT-4o
- ffmpeg
- SQLite
- Docker
Real figures from the build, not marketing math.