I have rendered a ride-on street-sweeper/street-vacuum from three different angles, on a pure white background, with a matching shadow channel.
Which currently available AI tool (commercial is fine) would be best suited to generate different urban sceneries, matching the rendered product (scale, light direction/shadows, reflections, camera lens and bokeh)?
NanoBanana from Google (Gemini) is the state of the art at the moment.
A good option could be RhinoBanana that uses API to export Rhino3D viewport directly into Gemini and render it.
Commercial options could be Vizcom (really focused on car and product design) and also Vectary3D.
Weavy, Flora and similar gives you access to a wider model to make from images to short videos.
Thanks, I’ll benchmark NanoBanana and Vizcom.
Also check out RunChat, which allows you to use different models.
All right, thanks; the RunChat website is a bit terse regarding how to faithfully montage isolated renderings of products into appropriate sceneries.
I recently started using Claude instead of ChatGPT. The difference is that Claude, through Visual Studio Code, writes what I want in the language programmers use. So far I’ve built a website from scratch, I’m creating an RPG game (pixel art), and now I’ve connected Claude to Visual Studio Code, to Rhinoceros and to Grasshopper… I’m going crazy hahaha.
Hi @Lagom
I’ve been using Google’s Flow (which uses NanoBanana) and think it works great. Here’s a quick example just using a random 3d model from our library. First rendering is 500 samples in screen res directly out of Rhino and second is the result from Flow using the prompt: “Please hang this lamp over the table in a modern, minimalistic dining room. Do not change the design, colour or shape of the lamp.”.
Thanks, that’s exactly what I meant @Normand and it works very well, changing the reflections on shiny and metallic parts of the product. Took an old project to try things out. The various plastics/coatings that have some relief look a wee bit off, but with some more restrictive prompting rules, I guess that can be avoided. Adobe’s Photoshop or Firefly don’t get perspective, shadows, and bokeh right. Not even worth posting here for comparison.
Gemini Nano Banana 2 test
Vizcom test
For HD up to 4K resolution, one has to subscribe to the commercial version I suppose.
Yup 
@Lagom Is the rendered 3d object faithful to your model, or is it a reinterpretation?
@Lagom this is one of my prefered example for Surface modeling class:
Is it you behind this amazing design?
The product must still be rendered by a renderer; AI renderers still cannot understand the various product parts and their CMF specification. The context, however is AI, and looks very credible. Only when you know you can spot the AI glitches. At €4,99 per month, Gemini is very competitively priced.
From my small understaing it’s actually improving the apperance of the light shade 
No, that’s not designed by me.
Ach, I want you give me back my "man of the day " cup . 
It could be a problem of positioning the herbs and “weight” of the terms in the prompt.
Also you can change the creativity slider to keep the image more stady.
Another option could be to use inpaint intead of rebuilding the whole image.
…but all of this requires new trail.
Gemini Nano Banana 2 does the right thing with nothing more than an dunce’s prompt: “Keep everything exactly as it is, but add a few herb pots behind the scissors towards the top left of the image.”
If you really want to cut down rendering time, the process must be very swift and reasonably predictable. The bare product CMF and on-white rendering process of three different camera perspectives takes 20 minutes at HD resolution. With Gemini, I’ve spend maybe a minute of thinking about the scenography, styling, and then typing two prompts.
