10.10.23-Showcase-Seoul-Common_Sense_Machines

Startup Exchange Video | Duration: 6:50
October 10, 2023
  • Interactive transcript
    Share

    DHRUPAD KARWA: Good afternoon, everybody. It's such a pleasure to be here. My name is Dhru, and I'm the product lead at Common Sense Machines, also known as CSM. And our company was founded by Tejas Kulkarni and Max Kleiman-Weiner, who are both computer science PhDs at MIT.

    So the problem that CSM is solving relates to creating 3D worlds. The demand for 3D content today is skyrocketing across different industries, like gaming, animation, simulation, prototyping, wireframing, and 3D printing. But to produce a single 3D asset, it's really, really complicated. It can cost thousands of dollars and take multiple teams many days, sometimes even weeks, of collaboration.

    Now, because of this complexity and the difficulty of this pipeline and the prohibitive cost and time, it makes it difficult for companies to build large and diverse 3D datasets. Take Shutterstock, for example-- one of the largest content marketplaces in the world. They have just less than a million 3D assets listed on their website, which is really nothing compared to their number of stock photos and videos. So it's really high time that someone solved this problem.

    Meet Cube by CSM, an AI product that lets anyone anywhere generate game-engine-ready 3D assets using simple prompts, like a line of text, a single image, or just a short video. And it's available anywhere. There's an app, there's a mobile application, there's a web application, there's a Discord bot, and you can also use it through our APIs.

    Now, the real power of Cube lies in its simplicity. It works in just three clicks. Click one, you enter your prompt, like this image on the left of this cool-looking sheep in a denim jacket and sunglasses. Click number two, you just hit the Generate button and wait a while as our AI does its magic. And click three, you export the 3D model on the right and take it into your pipeline, going to build your project.

    Now, this entire generation process today takes just two hours, not two days, or two weeks like in traditional 3D modeling. And we expect that already short time to dramatically decrease as our team continues to make more improvements to our AI model both in terms of speed and quality.

    Here's another example, this time showing a video prompt captured through our mobile application. Here you can see how Cube has transformed this physical humanoid object into something a lot more lifelike, a 3D model that's ready to enter reality.

    Now, this character on the right could be used in so many different ways. It could be used in a mobile action game. It could feature in Marvel's next superhero franchise. It could be used in synthetic training data for robotics. It could even be 3D printed and turned into a real-life robot filled with intelligence and common sense. So the possibilities are truly, truly exciting.

    Now, in terms of quality, Cube is state-of-the-art. Compared to anything else on the market today, our outputs are significantly higher quality, including the MIT OpenAI and its 3D model called Chappie, which was trained on millions of proprietary data points. And our team intends to maintain this competitive advantage, not just through making continuous model improvements as we are, but improving each and every aspect of the end-user experience.

    Now, returning to the theme of robots, I wanted to share with you this recent use case that came out of Tokyo, Japan. So there was a game development studio that organically discovered Cube online, and they used Cube to generate 3D characters from single images. They then animated the characters, using programs like Mixamo and Unity.

    They animated them to music, producing these lovely dance videos. And these videos went pretty viral online on Twitter. And I wanted to share one with you.

    So there should also be a really lovely soundtrack. Unfortunately, it's not playing. In the last two presentations, it's actually worked. I was thinking three for three is just hoping for too much. But anyway, it's hopefully still a nice little spectacle.

    So for those who can't see, it's basically a robot doing a robot dance. [LAUGHS]

    Now, the artist who made this, they went on to share a complete workflow on their personal blog. And I find that community energy like this is not only very inspiring for us as a product team, but it's also often a leading indicator of great things to come.

    So it's honestly a bit difficult to believe that we only launched our product a couple of months ago. And it's already grown to over 50,000 users, without a penny of marketing spent.

    Our cohorts in Korea and Japan are particularly exciting. They're growing 20% week over week, with users performing thousands of thousands of interactions on Cube daily. And the average user in Korea generates 2.6 3D models every 24 hours. So it's very much an exciting time for us.

    And we're also seeing use cases emerge and applications emerge from a range of different industries with 3D workflows, like gaming, animation, simulation, and also 3D printing. And we have enterprise plans, studio plans, as well as APIs, to support these applications and take these industries into a really beautiful 3D generated future.

    So let's partner, let's build [? PoCs, ?] and let's make new worlds together. Thank you very much.

  • Interactive transcript
    Share

    DHRUPAD KARWA: Good afternoon, everybody. It's such a pleasure to be here. My name is Dhru, and I'm the product lead at Common Sense Machines, also known as CSM. And our company was founded by Tejas Kulkarni and Max Kleiman-Weiner, who are both computer science PhDs at MIT.

    So the problem that CSM is solving relates to creating 3D worlds. The demand for 3D content today is skyrocketing across different industries, like gaming, animation, simulation, prototyping, wireframing, and 3D printing. But to produce a single 3D asset, it's really, really complicated. It can cost thousands of dollars and take multiple teams many days, sometimes even weeks, of collaboration.

    Now, because of this complexity and the difficulty of this pipeline and the prohibitive cost and time, it makes it difficult for companies to build large and diverse 3D datasets. Take Shutterstock, for example-- one of the largest content marketplaces in the world. They have just less than a million 3D assets listed on their website, which is really nothing compared to their number of stock photos and videos. So it's really high time that someone solved this problem.

    Meet Cube by CSM, an AI product that lets anyone anywhere generate game-engine-ready 3D assets using simple prompts, like a line of text, a single image, or just a short video. And it's available anywhere. There's an app, there's a mobile application, there's a web application, there's a Discord bot, and you can also use it through our APIs.

    Now, the real power of Cube lies in its simplicity. It works in just three clicks. Click one, you enter your prompt, like this image on the left of this cool-looking sheep in a denim jacket and sunglasses. Click number two, you just hit the Generate button and wait a while as our AI does its magic. And click three, you export the 3D model on the right and take it into your pipeline, going to build your project.

    Now, this entire generation process today takes just two hours, not two days, or two weeks like in traditional 3D modeling. And we expect that already short time to dramatically decrease as our team continues to make more improvements to our AI model both in terms of speed and quality.

    Here's another example, this time showing a video prompt captured through our mobile application. Here you can see how Cube has transformed this physical humanoid object into something a lot more lifelike, a 3D model that's ready to enter reality.

    Now, this character on the right could be used in so many different ways. It could be used in a mobile action game. It could feature in Marvel's next superhero franchise. It could be used in synthetic training data for robotics. It could even be 3D printed and turned into a real-life robot filled with intelligence and common sense. So the possibilities are truly, truly exciting.

    Now, in terms of quality, Cube is state-of-the-art. Compared to anything else on the market today, our outputs are significantly higher quality, including the MIT OpenAI and its 3D model called Chappie, which was trained on millions of proprietary data points. And our team intends to maintain this competitive advantage, not just through making continuous model improvements as we are, but improving each and every aspect of the end-user experience.

    Now, returning to the theme of robots, I wanted to share with you this recent use case that came out of Tokyo, Japan. So there was a game development studio that organically discovered Cube online, and they used Cube to generate 3D characters from single images. They then animated the characters, using programs like Mixamo and Unity.

    They animated them to music, producing these lovely dance videos. And these videos went pretty viral online on Twitter. And I wanted to share one with you.

    So there should also be a really lovely soundtrack. Unfortunately, it's not playing. In the last two presentations, it's actually worked. I was thinking three for three is just hoping for too much. But anyway, it's hopefully still a nice little spectacle.

    So for those who can't see, it's basically a robot doing a robot dance. [LAUGHS]

    Now, the artist who made this, they went on to share a complete workflow on their personal blog. And I find that community energy like this is not only very inspiring for us as a product team, but it's also often a leading indicator of great things to come.

    So it's honestly a bit difficult to believe that we only launched our product a couple of months ago. And it's already grown to over 50,000 users, without a penny of marketing spent.

    Our cohorts in Korea and Japan are particularly exciting. They're growing 20% week over week, with users performing thousands of thousands of interactions on Cube daily. And the average user in Korea generates 2.6 3D models every 24 hours. So it's very much an exciting time for us.

    And we're also seeing use cases emerge and applications emerge from a range of different industries with 3D workflows, like gaming, animation, simulation, and also 3D printing. And we have enterprise plans, studio plans, as well as APIs, to support these applications and take these industries into a really beautiful 3D generated future.

    So let's partner, let's build [? PoCs, ?] and let's make new worlds together. Thank you very much.

    Download Transcript