Response from Kaedim re our AI

Hi everyone! The whole team at Kaedim is humbled by the attention from the community, thank you so much for your excitement and support!

We have seen some scepticism and some messages across different channels about how our technology works so I wanted to be as transparent as possible with everyone.

Here is why I started Kaedim. I was a computer science student and I had taken a lot of 3D modelling units including character and set design and 3D animation. I found the process incredibly hard and the software had a super steep learning curve. I am sure all the 3D artists will feel me on this. I was looking for a way to make my life easier when I started talking to people who were doing this for a living; game dev studios. I sat down and had chats with them looking for a solution. But during these discussions, they unveiled the whole 3D creation pipeline as a huge bottleneck and problem for the gaming industry. I became super interested in the problem so I devoted my masters in Deep Learning for 2D to 3D reconstruction. At the beginning of 2021 I managed to get the first funding through to start Kaedim.

So this is where we are now. We are building ML algorithms that take 2D inputs of art, sketches and photos and turn them into 3D models automatically. We’ve been working on this technology for 3 ½ years, and I’m really proud of where we’ve got to. We have a product that’s starting to produce some exciting results — but it’s far from perfect.

Currently, the process looks like this:

• An input image is submitted and passed through our AI algorithm for reconstruction. The output is generally very accurate, but sometimes it’s not
• A quality control engineer will take a look at the output, and improve where necessary in order for the output to match our quality standards. Yes, our algorithm is not always perfect, and we want to ensure the output quality up to a standard.
• The output is returned to the user for further editing, and the addition of textures.

Today, the whole process takes an average of 15 minutes to be completed from start to finish. For the teams using Kaedim in production today, this represents a 10–20x speedup compared to manual work.

The quality control work that is happening today is vital for the algorithm advancement work that we do in parallel. We are creating a feedback loop for our algorithm that is gathering data in real time of generations and their improved versions.

That’s also why we’re pausing signups today — we’ve had an incredible number of people pay to use the product, and we want to make sure they all have a great experience. We simply don’t have enough quality control engineers to service more demand at the moment (please be patient, we will continue this as we scale).

Our goal is to achieve 1 minute return times in the next 12–18 months based on our product roadmap, at which point we hope to be ready for a general release.

I want to ensure we always listen to the feedback of our users. In this case, it seems that it wasn’t clear enough to everyone that this is our process. I’m sorry about that. We have been very transparent to our customers throughout all our sales calls, but we weren’t prepared for the public attention after sign ups exploded for the last few days. In retrospect, we could have communicated this better in our public-facing communications.

Thank you everyone for being part of the journey. Our team is working really hard to make our vision a reality as soon as possible and we really appreciate everyone who has joined us in these efforts.