While testing Patch, I created this 3d model, and the display image in a back and forth between Rhino and RunChat. (Note: Runchat doesn’t require Rhino WIP. In my case I prefer running it standalone in a browser)
I’m attaching a screenshot from the workflow in Runchat to give you an idea of the process.
From a top view, I ran a nano 2 to create the display image. I copy pasted the image to Rhino and cropped it / sheared it a bit to get what I wanted, and allowed another for some refinement, then repeating the copy paste and cropping in Rhino.
Then I had a good rendered view that I could further process the image to a contextual image. In the first image you see the first nano 2 attempt, where the model is a bit too big.
So I prompted to make it 20 percent smaller and remove the crossed item.
The first renderings had dirt on the white plastic, so that was removed in the third image.
For the final image I used nano banana pro, higher res and added prompts to match the lighting color to the display.
This looks pretty sweet Gijs, are those initial renderings from Rhino? I’ve found that usually nano banana doesn’t love screenshots as inputs but it definitely does a good job of re-lighting those renders. Maybe it works better with products than it does with architecture.
@Gwyl_Jahn I’ve only used screenshots from a rendered display of the product (the first and the third on top left). The first 3 scenes are from nano 2 and the last was nano pro. I found this one turned out pretty nice, but there’s a bit of a luck factor in this workflow. What I don’t like is that you are never going to get exactly what you want, and that the next attempt can sometimes be worse than the previous. I would still be very hesitant to use this for client work, because you can’t tune it precisely enough. But for exploring a mood before diving into building a scene to render, I can see this as an interesting addition.
Also, it depends on the product you’re working on. If the product is very niche, it starts to halucinate.
