With better prompting and processing each rendering in sequence, Chat GPT delivers fairly good consistency across the different views. Unfortunately, it refuses to add environmental reflections in the blade and screw. Nano Banana 2 is great for one image, but totally fails to sequence renderings, not generating a similar background.
I am 100% convinced that rendering as we all knew it is dying right before our eyes and AI rendering engines, as they get more clever, will be the way all 3d rendered images are generated in the future.
It’s not often you get to a see such a dramatic upheaval in an established workflow, kinda fun to watch it happen.
i am glad for it. i spent too many hours of my life trying to master arcane pseudo-scientific rendering technology, all for what, the sake of an image? all that time spent on an image could have gone towards another iteration. after all, we’re designing something not an image
Still the fundamental problem is far from being solved - how to render the product itself with the correct CMF. Even Vizcom fails in that respect, or, rather, it takes far too long compared to “manual” clipped renderings, always three views, I am using to benchmark the various available services for consistency and CMF fidelity.
Imagine prompting for all the different parts and materials used in this product. Or consider all the leather, textile, and laminate textures from suppliers when designing a business or first class aircraft seat. With “classic” drag-and-drop-materials rendering, you’re there in no time at all.
After experimenting for only two days (unpaid tier for all three), I would say the ranking is Vizcom, Gemini Nano Banana 2, ChatGPT. Quickly render with highly accurate CMF in Lumion, Keyshot, or Maxwell, and pipe into Vizcom. For only €43 per month, Vizcom offers a very structured approach and, if told, does not change the formal aesthetics of the product to be contextualised, and does not hallucinate much. There is also a free student/educational use tier.
Workbench
Low resolution preview
have you played with the paint in edits in vizcom?
you can paint in an area and give that specific selection a prompts to edit it.
Make sure to check out their You Tube channel there is some golf there.
this video is worth watching, they touch on the cmf and paint to edit process.
this one is good too-
Not yet. I first want to find out what the most reliable and also fastest way for credible contextualisation is. The CMF doesn’t need changing, because it’s already rendered.
Vizcom (so far) seems to excel in keeping consistency across views. Literacy, being able to write good prompts, is becoming an essential skill for industrial designers, as far as contextual and other types of visualisation are concerned. It would be interesting to learn how it is discussed and taught in leading industrial design educations, and in what kinds of courses.
I saw amazing example usign ComfyUI that has the advantages of being local.
Of course you need computationl power and a steep learning curve.
Recently they added the “app” version that should make the installation easyer.
The big pro is that you don’t belong to payperuse anymore using free models or you can adopt payed model without loosing your workflow.
Whether in a small design studio or an in-house design department, you always need speed and ease of use. Looking at an exemplary ComfyUI workflow, it seems you need a dedicated person to construct, maintain, and update workflows, besides computing power.
Using ComfyUI takes quite a bit of work. You need an OBJ file of, say, a street sweeper, and an MTL file to go along with it. However, a discussion revealed this: "If you require physically accurate materials and textures, the industry standard is to skip 3D rendering inside ComfyUI entirely. Instead, render a clean, multi-material base image inside your renderer, export a depth map and a material ID pass, and load those flat images directly into ComfyUI for the AI detailing and background generation.
Which is exactly what I’m exploring with ChatGPT, Gemini, and finally Vizcom. The question now rather is whether for productive (read fast, carefree, and billable) use in contextualisation, it makes sense to go down the Comfy UI or Vizcom route. An interesting topic for sure.
Managing the design discovery process with AI is much more efficient. I’m curious to see how Rhino will evolve in this new era.
So, you rendered the building in the renderer of your choice, with the client’s CMF or your architect’s specification applied to each building component, and then you created an appropriate context in Vizcom? Or what does the above show?
@Lagom The experimentation process went like this: I started with several architectural images that I liked, which I found on Instagram. That was my starting point. Then I asked the AI to develop different variations of the style based on the location. I requested a style that would fit the architecture of Northern Europe, Greece, and Italy, depending on the setting. After that, I defined the overall mood using reference images that had nothing to do with architecture. They were simply images, colors, and visual elements that I liked, with the goal of conveying a specific mood and giving the project a clear creative direction.
Like here: desert, The Great Gatsby, 1920s.
All right, I see; that’s the “common” use case for AI, different from what I am exploring at the moment - trying to show rendered products in different contexts from various viewing angles, in 2K/4K resolution, without AI hallucinating or changing the product’s CMF, product graphics, etc., and the whole shebang done in max. 10 minutes ; )
Yeah, I think I’ll probably do that too in the future—use a 3D model that I’ve designed myself, then assign materials, lighting, a background, and so on, and see if it preserves the original design of my model. But for now, I’m just trying to understand how everything works in Vizcom.
Should be perfect for the huge amount of footwear you’re doing by the looks of it.