Respecting rights of users, fundamental to building AI: Canva’s Robert Kawalsky

For Canva, which remains a one of its kind visual suite for creators, there come no indications that the Australian tech company is about to slow down. Even temporarily. They’ve just announced Dream Lab, which they reference as a high-fidelity visual AI tool, as an addition to the Canva app that we use on PCs, on the web and on smartphones. In addition, the Magic Studio and Visual Suite products, which were strengthened significantly earlier in the year, see an addition of new features too. The approach towards keeping the fundamentals constant, whilst bolting on new functionality that often has artificial intelligence (AI) underlining it, has its advantages.
“I think that is there’s two major benefits for our community. They get to learn Canva once, and then they can create a whole variety of things. For us, we can build a new feature and make it available on across the whole suite,” says Robert Kawalsky, Head of Product at Canva, as he sits down for a conversation with HT. Canva now has 200 million active users, up from the 180 million number the company had shared with HT in April, at the time when they announced new Magic Suite updates at the Create keynote.
Recently, Canva has continued its trend of acquiring AI expertise, by bringing Australian tech company Leonardo.ai within its fold. Kawalsky himself co-founded and was CEO of Zeetings, a presentation startup which was acquired by Canva in 2018. That strategy reaped rewards rather quickly, with Leonardo’s Phoenix foundation model defining Dream Lab and more. Edited excerpts.
Q. Canva’s taken significant forward steps in terms of its feature set. How challenging was it to shortlist and build on everything that’s been added to Canva in the last year, with the latest features?
Robert Kawalsky: We’ve added a huge amount of functionality across the course suite, on the app and web, including whiteboard in presentations, video and different design types. We’ve been investing in our platform and infrastructure for over a decade now. We’ve built a horizontal suite. When you think about Canva, the one thing that’s really quite unique is a whole visual suite, whether it’s websites or presentations or whiteboards and video, while being completely interoperable. I think there’s two major benefits for our community. They get to learn Canva once, and then they can create a whole variety of things. Also for us, we build a feature once, and make it available on across the whole suite.
There is a complexity to building at this velocity. I think we’ve really built these muscles over a long time now, about how to move really quickly and we think it’s essential particularly with everything happening in the world of AI. We’re really proud of being the first movers, with features such as bringing a magic editor right product into a document that’s editing directly in line. The velocity has always been a critical part of what we’ve done. I hope we don’t slow down, and I don’t see any reason to slow down. We’ve got an amazing team of builders to serve our community with amazing functionality.
Q. Canva made the absolutely correct bets with AI over the last few years. First Kaleido and Affinity, and now clearly with Leonardo.ai at rapid speed. Do you believe the strategy of AI acquisition helps quicken innovation?
RK: We’ve always recognised that there are incredibly talented teams doing amazing things all over the world. We did acquire Affinity and then Leonardo in a relatively short succession. That’s in a context, relatively few acquisitions. If you look over the past decade, we’re really diligent about each acquisition we make, because it is a big bet. Fundamentally, when we come across a world class team doing amazing things which we can kind of accelerate with our team, our distribution and by baking their technology into a product, it is just such a harmonious combination.
Leonardo, for example, is a truly world class team building a world leading foundational model, and one that allows on the image creation side, functionality including more control and precision. It’s worth seeing, and to take that and to put it in the hands of the Canva community, is just incredibly powerful. We look to those opportunities where there’s an incredible team thats created something really special and unique, and where the combination of that product fits within Canva. It is certainly a quick, way of bringing in new capabilities into Canva.
Q. How does the weightage work now, in terms of focusing on in-house models, versus using models that are available from other companies?
RK: We’ve been building in-house AI capabilities for many years. Kaleido’s background remover was our first in house model, which was and continues to be the market leading product in that space. We’ve always believed that one of the prongs, is to build in house. There are an amazing array of models and capabilities that are now available from all the leading AI companies that everyone’s talking about. I think the area that is unlocked by being able to take those capabilities, it’s never just a matter of plugging them in and turning them on. There’s a lot around really making that work for our audience. But taking the best of what’s out there and making them available for the community is the second pillar.
The third is Canva itself, which is a true platform today. We have hundreds of apps built by third party developers on the Canva platform, and many of those are AI powered. We think that’s another way of making the best and most cutting-edge AI capabilities available to our community. There are amazing companies we can partner with, and there are incredible things we could build in-house as well. We choose to start with the point about the value we want to bring our community. We have just crossed over 200 million monthly active users, which gives us an incredible opportunity to bring new technology to the community.
Q. Where does Dream Lab fit in Canva‘s overall landscape?
RK: Dream Lab is a completely new surface in Canva. It’s a brand new image creation capability that is able to it gives users an incredible amount of control and precision when creating images. It uses a foundational model called Phoenix, that Leonardo built. It’s fast and creates incredible fidelity images. It also allows a user to reference styles. This is important as to move image generation from a capability that you have, to a capability that actually allows you to get real work done. You need to be able to control an image, and style referencing is important to doing that. It’s worth mentioning on the capability side is that it’s also deeply integrated into the rest of Canva. So it’s one click to go from an image you created in Dream Lab into a design, and what that means is you then get all the image editing capabilities there and then.
One of the things the Phoenix model is incredibly good at is text generation. You can create an image with text embedded in it, if you then take that image into a Canva design, you can use our grab text functionality to identify that text using AI and edit it as if it was a normal text box. You actually now can create images with AI and then use AI to edit those images, which is a new capability that we can put into people’s hands. One of the amazing things we noticed with Canva is, as soon as you make technology accessible and easier to use, the number of ways in which people use it are surprising. I think we’re going to find the same with Dream Lab.
There are use cases that we’re already starting to hear from users, whether it’s in creative agencies or small businesses. A powerful element of Dream Lab is that because it’s its own surface, every time you put in a prompt, you get a number of results. A particularly powerful aspect with Dream Lab because of its Phoenix model, is its prompt adherence. The Leonardo team has got really good at taking a prompt that people input and ensuring the output really adheres to what that person was thinking. It’s hard to think of designs which won’t be enhanced with this capability, whether it’s a presentation or a social media post or even a kid’s birthday invitation
Q. Will there be guardrails to ensure the AI generated images don’t go wrong with human faces, skin colour or the use of certain phrases in a prompt? Canva’s always had a strong and clear data usage policy, has anything changed on that front?
RK: Absolutely, we have invested in trust and safety capabilities on Canva from an AI perspective, for a while now. We’re applying that same capability into Dream Lab which ensures there’s no harmful content, offensive content, and so on. Ultimately, we are committed to providing the safest experience while ensuring that there are those rails in place.
We believe that for users, it should always be opt in. When we train on data, it’s really about thinking holistically how we use AI. One of the things we care deeply about is always having guardrails in place, ensuring any data we use is always on an opt in basis. Even then, we must continue to respect the rights of creators. We have our $200 million fund for our creators (these are designers who create on Canva) and we found almost 99.9% of our creators have opted in to taking part in that fund and and helping to train our models. RRespecting the rights of creators, reward them appropriately, and respect the rights of users, are fundamental to building AI capabilities.