AI pipeline, web
An AI pipeline that turns one reference post into on-brand visuals for a studio's marketing. Gemini reads the post, three image models redraw it, and every post stays under a one-dollar cap.
- Commits
- 407
- Cap per post
- $1
- Image models
- 3
The problem
A studio promoting several apps needed new visuals every day, and making them by hand didn't scale. AI generation was fast but unpredictable, in quality and in cost.
What I built
Paste the link to a post that performed. CloneOS fetches its slides, has Gemini analyse each one, then rebuilds the post for the studio's own brand: the right image model for each slide (Nano Banana, GPT Image, or FLUX on a rented GPU), captions rewritten in French or English and drawn server-side with real fonts, then an export or a scheduled publish. Every member sees only their own work, and a spend guard caps each post at one dollar, checked in four places.
One pipeline per post: read it, pick the cheapest model that can redraw each slide, rewrite the captions, judge the result, and stop at a hard cost ceiling. Roles and per-member visibility keep everyone's work separate.
