Working with AI in visualisation, and will AI replace 3D artists?
AI has definitely taken over the conversation, and not everything everywhere. My view on it has evolved over the past few years, and I am sure it will evolve again. At the moment I think it is here to stay, but the doom-and-gloom Terminator vision isn’t at all what we’re headed for. Or maybe I’m naive.
Clients have asked me if we use AI in my 3D company. The answer is yes, but only on certain projects, or in certain ways. It can be a way to reduce prices and costs.
AI is currently more of an augmentation of our existing workflows, and that could be said for any industry. It fits in where needed, and as we learn more about it, it speeds things up, makes things easier and gives us more time. Having said that, it doesn’t replace the user and their respective experience. A user still has to know what they are looking for, and writing the prompts takes practice.
For example, I won’t become an accountant just because I can get AI to do it all for me. It still requires situational awareness, relationships, trust and past knowledge that’s acquired over years. We have been augmenting our lives for centuries: fire for warmth and cooking, steel for tooling, calculators for mathing, computers for reading this blog, and email for sending letters.
In the world of art and 3D renders, a few AI workflows are starting to be convincing, but it doesn’t do everything and still lacks personality. Again, like my example above, it still requires experience, the knowledge base to prompt the right changes, and the important one: reading plans. It doesn’t get the big picture, or how it fits within the world. Everything seems to be on an island with just enough ‘environment’ to pass the checklist.
For example, it doesn’t fully understand the context of a site, the surrounding buildings, trees, plants, and project-specific anomalies that are on some plans but not on others, where there are legitimate reasons for it. AI tends to forget things, and it takes constant updates and hacks to get it to remember your situation, your workflow, or even just the current project.
Where it helps, and where it falls down
Both columns are from working on real jobs, not from a demo.
Where it genuinely helps
- Textures, shaders, modelling and even render settings, inside existing software.
- Images from scratch, with good prompting.
- Compiling a briefing package, to about 60 to 80% ready.
Where it falls down
The bulk of the work is still done manually.
- Reading plans.
- The context of a site, surrounding buildings, trees and plants.
- Project anomalies that are on some plans but not others, for legitimate reasons.
- Remembering your project without constant updates and hacks.
- Change. Ask for one thing and it changes several others.
What does AI do to a render?
AI can literally create images from scratch with good prompting, and convince some clients that it’s ok to use them for marketing. Check for watermarks. Or it can be integrated into existing software to help with textures, shaders, modelling and even render settings, but the bulk of the work is still done manually. Which raises the question.
Will it replace Artists and Creatives?
No, I don’t think so. Artists and Creatives have the unique sensitivity and emotions that AI will never be able to recreate. The “je ne sais quoi”, as the French say. Clients don’t have the time to sit on a computer all day figuring out how to prompt Nano Banana into a great-looking render, on minimal briefing, to save a few thousand dollars. The opportunity costs are too high.
A client can use AI to help with the briefing package, though. It can be directed to all the information, compile it, and get it 60 to 80% ready for the Creatives. The final 20 to 40% is clarifying, cross-checking and uploading it to a central hub like Propagio.
What happens when clients change their mind?
AI doesn’t like change. As soon as a camera angle is moved, or something is requested to be replaced, it’s like AI has to start all over again. Context is lost. It was all a facade. Worse still, ask it to change one specific thing and it will change a bunch of other things that weren’t requested. It can be pretty frustrating, unless you have spent some time developing systems and workarounds.
I am sure this will evolve. I can see it getting easier, cheaper and faster. This is certainly a big step since my last blog on AI back in 2023.
I hope this helps someone,
Thanks for reading.
Luke