From Image to 3D Model Using AI
What if you could send in a straightforward drawing and receive a 3D model of it in return? Kaedim does just this. And the biggest names in technology are following suit with similar 2D to 3D solutions. Meet Konstantina Psoma, CEO of Kaedim, who wants the game industry to spend more time on creativity and less on geometry.
How did you identify the game industry as a focus? What other industries were you looking at?
– The idea for Kaedim came from personal frustration after using Autodesk Maya for a 3D modeling project while studying at university. I was then surrounded by hundreds of fellow computer science students that were all playing video games in their free time.
After being struck by the steep learning curve of the software and the lengthy hours I had to put in to complete the 3D project, my first response was to go and talk to game developers as, in my mind, there was no way they hadn’t solved the problem for their scaled 3D asset productions.
I was shocked to get tours in studios that had rooms of 50-60 3D artists all modeling from scratch, from a cube, every single 3D object for their scenes. Fast forward, having become really interested in the problem, I decided to do my Masters on Deep Learning and my thesis on 2D to 3D reconstruction, and this is how Kaedim got born. Starting from the gaming industry was the obvious thing to do since more and more people were coming out saying that games dev has become unsustainable, titles were getting postponed, and there is a huge crunch and employee turnover because of that.
The next steps, after games and metaverse creators, include verticals such as e-commerce, product design, and architecture.
As a user putting in an image, how much time does it take to generate a 3D model using Kaedim’s platform?
– With our current AI model, which optimizes for quality capture, the process end to end takes an average of 15 minutes. But the time really depends on the input image and can go as low as 1-2 minutes.
Is there a particular type of asset or object that usually outputs more precise models than others?
– Generally, if more than 1 point of view of your art or object is provided, the more accurate the outputs. Also, our technology is best for simple inputs of art or objects without an excessive amount of fine detail.
When talking to game companies, what are their biggest concerns with this type of 3D automation?
– Game companies are super excited to hear about our tech since we are the only AI tool that provides 100% ready-to-use 3D assets for production pipelines. One of the requests that we often get, which can be considered a concern as well, is the detail loss that very fine detail inputs might experience. Studios constantly want us to build new features and upgraded functionality. For example, we are working hard to increase the quality capture without increasing the time so that higher end productions can use Kaedim as well. They also ask for a texturing extension. A lot of work to be done.
If you look at other uses of AI and machine learning in digital entertainment, what are you most impressed with and where do you see a fit at scale?
– I love all the generative AI and text-to-image applications more specifically. They are really complementary to Kaedim. We see a lot of AI-generated 2D inputs coming through Kaedim to be transformed into 3D models and it’s pretty impressive. An end to end AI powered pipeline. For small studios with low budgets, an AI-powered pipeline can be a game changer for producing quick proofs of concepts, ideate and even accelerate the production phase of making games!
video by Kel Savage
If I was an executive in the game industry, seeing all these upcoming AI applications, challenging the current ways of doing things – maybe even facing internal resistance from staff that is currently doing a lot of work by hand. How would you advise me to strategize around it?
– From all the companies and studios we are working with, there is no single 3D artist that thinks our AI software is replacing them. When you have to make 100 3D models by the end of month, and you know that you’ll need to work by hand 14 hours per day to MAYBE make it, then you are looking for a painkiller. Having a tool to augment you is a life saver. If you are a business owner that is about to announce to their publisher that they need to postpone the launch, you are looking for a painkiller. This is what Kaedim is. We are working with companies that want to do more or do better, with companies that want to be ahead of the curve.
In general, how do you see the game industry changing in the next decade?
– In the next decade I hope to see the following changes:
More variety in games. The difficulty to make content, and especially 3D content, has led to little to no variety in the style and content of the games that are released. In the next decade I hope that tools like Kaedim will make the creation of 3D content easier, so more people can start creating niche content and more end-users will be able to find an experience that they enjoy.
Games getting majorly de-risked, both financially and in terms of content. The ability to make much cheaper and quicker proofs of concepts as well as rapid prototyping and iteration of ideas will be crucial for getting more games to publishers and launched.
Lower barriers to entry. It shouldn’t be so hard to make a 3D game or experience. The whole world is moving towards an immersive digital 3D format, so anyone should have the tools that allow them to contribute to our 3D experiences. Right now, you’d need to assemble a team of experts to make something worth-while and this is definitely something that has already started changing, but it’s only the beginning.
For Kaedim, I want us to be the force moving these changes forward.
Papers have recently been published providing similar 2D to 3D technology from both NVIDIA and Google Research.
Kaedim’s software has sparked debate on a number of social forums, where the authenticity of the technology has been questioned. Some claim the 3D models are manually created, and not by an AI, looking at their geometry. Konstantina has responded to the public criticism the following post.
