I made a quick test model to further explore this.
I was able to make part of the text become signage, by adding ((architecture sign)) to the prompt:
exterior shot of utopian architecture building, ((architecture sign))
To fix part of the prompt being interpreted as signage, you can use the Render Selection tool. Select the region to change, update the prompt and render. You can optionally turn on the seed to match the original render to try to preserve a better blend.
The blend can also be improved with the Render Selection Blend button:
For the text graphic - Veras uses the base image as input. Here’s a post that expands on how it works, and how it will change in the future: Questions for Better Result