LogInGraswald's Julius Harling has joined 80 Level to tell us more about their new Graswald.ai solution, discuss its capabilities and role in 3D workflows, and explain how they managed to set up a generative AI without using anyone's data.
Please introduce yourself to those who don't know you. How did the story of Graswald begin?
Julius Harling, Founder & CEO Graswald: My background has been on the 3D creative side of things. As a kid, I was a huge animal nerd and started designing 3D models of animals as a hobby. That's how I got into the rabbit hole that is the 3D industry.
When I was fifteen, I spent half a year in Canada as part of an exchange student program. Without much to do, I delved deeper into 3D, started my first professional projects, shared them online, and received great feedback from the community. People asked, "Hey, can we download this? Can we maybe buy this?" I thought that was a good idea, so I started working on a prototype that allowed people to build 3D worlds more easily. I published the first version in 2018, and it became a much bigger success than I expected, with over two hundred users.
At that time, I was still in school. After graduating from high school, I wasn't sure what to do next. I considered doing something artistic and almost started studying classical music but ultimately decided to focus more on 3D art. I learned about startups and company building and, step by step, developed a clearer vision of the company as it exists today.
If my memory serves me right, Graswald first put its name on the map with the release of Gscatter, could you please share a few words about its current state? What features does it currently boast?
Julius: When I started doing 3D, there wasn't a big wave of interest except among the specialists and the artistic community. Creating 3D scenes was (and still is) a lot of work. You need a lot of content and assets that need to look really good to maintain the realism of the scene. You also need tools to build the scene. You can select every single element and move it by hand in 3D space, but that's even more cumbersome than doing it in the real world.
Imagine decorating a 100x100 meter scene in the real world with every single detail – that's a ton of work. In 3D on a computer, it's even more complicated. This led to the idea of providing templates – 3D content that looks great out of the box – and tools that allow you to use that content to build scenes.
With Gscatter, that's exactly what you can do. You get ready-made templates of content and can populate your scene with them. We focused on vegetation because nature is so hard to do on a computer. It's really difficult to render in 3D. We felt specializing in this and focusing on it provides a lot of value for people who want to use it, but people have been using Gscatter for all kinds of projects. Whenever you have objects to scatter around, it's a great tool. Gscatter is a really cool tool, and people are still using it.
We're doing our best to keep it up to date. It's free, so people can use it. We're currently planning to open-source it, but open-sourcing takes time, effort, and planning. It's not one hundred percent ready yet. I know we're not always on top of the latest Blender developments as we are focusing on many things at the moment, but we love Gscatter and our users. We want to continue developing that tool.
Recently, you rolled out Graswald.ai, a video-to-3D model AI solution. In this day and age, when many 3D Artists dislike all things related to generative AI, what encouraged you to launch the new product?
Julius: AI is a huge topic, with generative AI being a specific area within it. The general idea of machine learning is to speed up tasks that are boring and cumbersome, allowing people to focus on more interesting activities. Generative AI is particularly challenging because it aims to produce original content using trained data. However, the content isn't truly original, which concerns artists since their work is often used to train these algorithms, leading to their creations being replicated by generative models.
That's not what we're doing.
Our AI uses machine learning focused on the specific input data we provide. We employ it for tasks typically performed by humans that are difficult and manual, such as masking objects, identifying items, or reconstructing a scene from a video. Our AI does not involve training a foundational model that generates similar data based on large datasets.
We clearly focus on this aspect, which is important since we work a lot with products. When you have a product, its reconstruction needs to resemble the actual product accurately. It’s not useful if it looks cool but has a competitor's logo or a slightly different shape. This applies to AI in general, our usage of it, and my perspective on the future of AI for 3D Artists.
We are now expanding our efforts to reach more people and create tools beyond the VFX and gaming industry. AI excels at automating boring and repetitive tasks that are technically challenging. With the tools we’re developing, we are enabling more people to engage with 3D, become creators, and focus on the creative aspects by automating the tedious parts.
While it might sound cliché, I truly believe that the most effective AI products are not those attempting to handle the entire creative process, but those that empower users by automating repetitive tasks. That’s essentially how we use AI – to make 3D easier and improve the lives of those who work with 3D.
Could you please explain to our audience how Graswald.ai works? Who is your target audience?
Julius: The idea behind Graswald.ai was to take what we've already been building – a combination of content and tools – and make it so easy to use that non-technical people can leverage 3D as a medium to do more and simplify tasks, such as product photography.
The product essentially comprises the same components: a content component and a tooling component. Users can utilize our AI pipeline by creating a video of a product, uploading it, and generating a 3D photorealistic model from that video through our AI pipeline. This enables users to create a wide range of content and marketing material around the 3D object, similar to what people were doing with Gscatter, but easier, more automated, and for a broader variety of use cases. One of the coolest aspects is that everything runs in the browser, so there's no need to download any software.
I mean, I love Blender. Blender is amazing. But it still requires a non-technical person to download a 3D tool, install it, understand navigation, and grasp all the UI elements. Then, inviting someone else to collaborate with you is difficult. Right now, as we're rolling this tool out very carefully, we really care about the quality of our products and how people use them. We're very selective about who gets to use it because we want to spend a lot of time understanding how they use it best and how we can improve it.
Currently, we're working with several larger businesses and enterprises to make this the best tool we can. We plan to gradually make this more publicly available to our community, which is where we come from. The use cases people are currently employing include digitizing entire product catalogs for e-commerce companies, getting 3D at scale very easily but at a very high quality, and optimizing it for the web.
How can the new tool be integrated into regular 3D workflows?
Julius: That's a really interesting question. Right now, many of the people we talk to who are using our tool haven't used 3D before. They don't have established production processes or pipeline stages, and they aren't able to set up anything very complicated. So currently, our tool is the only one they're using, which is beneficial because it keeps things simple for them.
And that's what we're focusing on: being a solution that's easy to use, simple to start with, easy to onboard, and straightforward to work with. Our goal is to provide all the necessary features in one place, such as creating, editing content, building scenes, rendering, sharing online, and even adding music to your scenes. This is our current focus.
The technology we're using behind this is 3D Gaussian Splatting, which we believe is powerful and can support many of our future plans. Gaussian splats can theoretically be integrated into various platforms like Unity, Unreal Engine, and Blender. Consequently, the content created with our tool can potentially be transferred to other projects and pipelines. However, the primary value of our solution lies in its simplicity. While it may not suit very sophisticated production workflows, it is ideal for those just starting with 3D, looking to learn, and wanting to leverage the technology without extensive training.
A question on nearly every artist's mind when it comes to AI is how any particular model was trained. How did you train Graswald.ai? What data did you use? Should Graswald.ai and Gscatter users be worried that their projects will eventually be used as training material due to some fine print in the Terms of Use?
Julius: No. As I mentioned, we're not training a foundational model. Any training we do is with your specific video and input. Our customers upload a video, and we only use that data to reconstruct the model. We're not accumulating data to create the next big diffusion model. So, no, you don't have to worry about your projects or scenes. We're not a data monster trying to collect all your data. It's yours. The models you create are also yours. We're just here to help you create more effectively and efficiently.
How do you approach the business side of things and promote Graswald.ai? What's your current business model?
Julius: As I mentioned before, we're currently working very closely with bigger brands, enterprises, and companies. I believe that one of the significant benefits of our work is establishing 3D capabilities in organizations with little or no experience in 3D or those with 3D experience but lacking the manpower to scale it up to the necessary level for their needs and the content they want to produce. At the moment, these are the types of deals and partnerships we're pursuing.
However, we also believe that the real value lies not only in enabling large businesses but also in empowering small and medium-sized businesses and individuals to create more 3D content. We plan to gradually release parts of our product and solution to the public, gather feedback, fine-tune it, improve it, and adjust it. Our goal is to eventually make it accessible for anyone to use and create a lot of cool stuff.
We are very conscious of the product we're building, who we're building it for, and where the value lies. We want to ensure we do this in a structured, step-by-step manner, so at every stage, we can be open to feedback and make the right choices. In the next few months, you might receive some updates on what we're working on.
What are your future plans? How do you plan to develop and promote Graswald.ai? What should your audience expect?
Julius: Our vision is to enable amazing, simple-to-use 3D content creation. By content, we don't just mean assets, but any type of visual content that originates from some form of 3D space. We aim to make this content easy to create, visually appealing, and of high quality.