We started Prexi because we were tired of the same problem. Open a new model, sit in front of a blank prompt field, spend 45 minutes on Reddit and Discord trying to figure out what actually works, get okay results, ship anyway.
The models got better every two months. The prompt advice did not keep up.
Rishit Patel built the original Prexi in 2022 as a Stable Diffusion gallery. It was the right idea at the right time. We bought the domain, wrote to him, and relaunched it in 2023 with a simple rule: nothing goes in the library until it has been tested across every active model we support.
Three people. No automation pipeline. No bulk-import from other datasets. Every prompt in here was either written by us or submitted by someone who attached an output image as proof.
The library is free to browse. It always will be. That is not going to change when we eventually run out of runway or get acquisition interest. We have it in writing.
No VC. No employees. No quarterly OKRs.
Former Pentagram Berlin. Has opinions about type that she will share whether you asked or not. Curates the library aesthetic and keeps us from publishing anything visually embarrassing.
Ex-Runway Tools. Wrote the prompt-model matching layer from scratch over a long weekend in November 2023. Still fixing the bugs from that weekend.
Former Are.na editorial. 4 years of building visual libraries by hand. Every prompt in Prexi has been through Sid. If it is mediocre, it is not here.
Prompt submissions, press, or just hello from Lisbon.