关于：视觉人工智能之思丨3DsMax,Corona,SD,PS

Re: Vision

Type/Project: Minds

Location: Worldwide

Year: 2024

Work by: 3DsMax/Corona/SD/PS

Hi! I wanted to share my experience with AI, Stable Diffusion in particular. It will help you understand what it is, its advantages and disadvantages and how it work in real projects.

The use of AI in architecture for visualization became widespread only a year ago, and it has quickly become an essential tool. At the same time, I started exploring neural networks and created my first project: I improved one of my first exteriors. At first I only improved the original render...

...and then completely redrawn

I was captivated by the result, and I further explore, and realized that there is so much I can do with it and completely redo greens of any kind, both from draft renders...

...improvements to fully finished renders...

...and makes masterpieces from several lines

But also to work with complex bird's-eye views, not only correcting such complex matters for 3D editors as the sea and sand, but also completely re-creating huge tropical landscapes!

However, there are also some drawbacks: you can't create your own fully-fledged interior design or replace objects from scratch. It's also impossible to recreate the exact materials, such as bricks.

But for creating full-fledged images from sketches, altering nature, enhancing the appearance of asphalt, earth, leather, and much more, where there's no clear order, improving the atmosphere of the frame by adding people and animals, turning a rough render into a finished picture, and much more, which I've mentioned as examples, neural networks are perfect!

It's also worth noting that for qualitative improvement,you may need up to 2 hours per image, and for a full-fledged redesign in large exterior projects, it can take up to 4 hours. We also had to upgrade our graphics card. For me, the minimum requirement was an NVIDIA card with 8GB of video memory. Additionally, you need patience when going through options and learning how to communicate effectively with a neural network to ensure it understands you correctly. The whole process is like stitching together a beautiful Frankenstein, but the result is truly worth it!

In conclusion, I'd like to say that it's not fair to criticize AI-generated images. They are just another tool, that can produce incredible results. There's no magic button to make an image beautiful — each frame takes several hours to create. You need to understand the settings of the neural network and process the image correctly. But thanks to this, you can get a picture that is often impossible to achieve with traditional methods.This means you will have an advantage over those who resist change. You will become better than them, and customers will be interested in your work. This is a major step forward for architectural visualization, which has been lacking for so many years!

Thanks! Stay tuned!