AI Collage
From a sentence of goals to a finished collage.
Product screens
Overview
I built an autonomous agent that turns a sentence of goals into a finished photo collage.
Problem statement
Turning goals such as “freedom, discipline, financial independence” into a good-looking vision board normally takes hours of searching and arranging.
Solution
Prompt design.
I use GPT-4 to expand a short statement of goals into themed search prompts.
Retrieval.
I retrieve about sixty candidate images per run from Pexels using those prompts.
Filtering.
I filter candidates by orientation and by semantic similarity between image and prompt, so the board stays on theme.
Layout.
I arrange the remaining images with composition rules, so the board needs no manual curation.
Interactive front end.
I render the result in a React front end, so the finished collage can be viewed in the browser.
Outcome
I delivered the complete prototype, backend and front end, in about three weeks. A one-sentence brief becomes a finished collage without manual curation.
Technologies used
Data volumes & scale
- 1 prompt in, 1 collage out
- ~60 candidate images per run
- 5 pipeline stages
- 3 weeks to deliver