This exercise shows the delightful Valerie from @flowercart.sf and imagines if she also opened stands that sold other joyful goods, such as books, art, and music.

We used Microsoft Copilot and created renderings that show a few small ways to make A Better San Francisco.

This video imagines a viewing platform at the top of the Montgomery Street steps.

The current view is obstructed by trees and a small parking lot.

We used Grok Lite from xAI and GPT Image 2 from OpenAI.

Both image generators created nice viewing platforms at the top of the stairs.

Yes, the trees were rehomed in the process, and they are enjoying their larger, rent-free home together.

In this video, we use three AI image generators to show how a parking lot in the Marina District can be transformed into a neighborhood square or plaza.

The image generators we use are:
1. GPT Image 1, by OpenAI
2. Flux Kontext Pro, by Black Forest Labs
3. Gemini 3.1 (Nano Banana 2) by Google's Deepmind division

While each of these image generators is amazingly powerful and precise, one is the clear winner in this comparison.

Note: These models and versions change frequently, so they may not always have similar capabilities.