Canva’s Danny Wu on Giving 230 Million Users Superpowers With AI - Ep. 265

0:10Hello, and welcome to the NVIDIA AI Podcast. I'm your host, Noah Kravitz. Our guest today is from Canva. Canva is a visual communications platform. You've probably heard of it. They're really well-known, I'll put it that way. They serve all kinds of customers, big, small, Fortune 500, small businesses. Their goal is to make it easy for anyone to design and publish anywhere. So most of the time, I read a little pre-written introduction to welcome the guest on, because most of the time I get to talk to people who are doing really cool things, But I don't get to experience them firsthand because it's not feasible to, you know, go to the lab and watch them make new drugs or what have you.

0:47But in this case, in the pre-planning, I was encouraged to go check out some of the new AI features we're going to be talking about on Canva. So I carved out some time before this taping, and I was just telling Danny, our guest, that I got so excited. I actually did. I didn't tell you this part, but I hit kind of blank page syndrome at first because we were saying I'm not a, you know, I was telling you, I'm not like a visually creative person naturally. But then there was an example and I ran with it. And so I'll tell you more about that in a second. But first to welcome you on, Danny Wu is head of AI products at Canva.

1:20Danny, thank you for taking the time to join the AI podcast. No problem, Noah. I'm really excited to be having this chat and thank you as well. Thank you. And, you know, you can tell it's Friday as well as we're taping this. So that makes it extra exciting. But it was really cool because I just went on to play with it. and I went to Canva code and there was a sample, you know, you have little ideas up there for inspiration. One of them was to create an interactive timeline based on history. And so I saw it and I was like, oh, I can adapt that. Oh, I'll make one for the podcast. And so, you know, I figured out how to change the prompt and I did it and it coded it.

1:57And then we had the back and forth of adjusting some things. And then I wanted to hook it up to an RSS feed. And, you know, having that experience firsthand is what it's all about. So it made me extra excited to talk to you. So welcome. Thank you. Canva told us that's really, really exciting. And that's a great use case. So if you would, tell the audience a little bit about your background, about Canva, and then we'll get into talking about AI. Yeah, sure thing. So I actually joined Canva about nine years ago, back when I was a lot smaller. And I started off as a front-end engineer. year and um over the years it was really really kind of the vast potential of just like how wide like the design spaces and how much we can be helping helping the world realize their creativity and empower them to design that just um that just made me really excited really excited to really excited to do all the cool things we do at canva um i transitioned to becoming a product manager um and led our first data science and machine learning team which was a natural fit But not just for AI, especially with a journey with AI, just getting more and more capable and advanced and at an incredible pace.

3:13So best that you can remember, nine years is a long time for an AI podcast. This world moves so fast, right? But best you can remember, what was the product or the user experience like nine years ago? And what kinds of technologies were you working with to build those experiences? Yeah, so actually the core elements of Canva haven't changed so much on the surface to the end user. We've obviously went through a few rewrites and obviously Canva has been expanded with so much more functionality. Even nine years ago, a lot of the things we were thinking about was to help me break down complex and tedious tasks and use technology to make it as easy and simple as possible.

4:02So everyday users, just anyone, don't have to worry about it. And they can just focus on the creative process and creating. Right. And so that's kind of a natural, it's almost a natural answer to my next question. But when you started looking at AI and how to incorporate AI into the camp experience, what was inspiring you? What were you and your teams looking to achieve? How did you get started integrating AI? Yeah, so I guess even before our earliest was really we were calling it machine learning back then. And the earliest was about maybe seven or so years ago when we started using ML to power our content search and recommendations.

4:43Not to interrupt you, Danny, but let me just give a little shout out because we've had a run of guests lately who the first chance they got said machine learning when I said AI, and I love it. So shout out to ML always. Anyway, sorry. Continue. No, Pops. It is kind of amazing seeing how this world progresses. But yeah, so we started with using ML to make it easier for our users to find matching and relevant graphics and photos, things that go well with the rest of the design or what they're trying to do. And that was definitely really useful. Match recommendations has been used quite literally like billions of times.

5:22Then we started getting into features like background remover, which lets people just with one click. if you've got a normal photo, just remove the background. It is something that is like, you know, very like standard and expectations we stay. But when we launched it, it was definitely like one of our favorite features and continues to be. And I think just to go back to your question, I think a really pivotal moment was around the release of StyleGAN and generative LSO networks. And actually like Navidia's research on StyleGAN, like this person that needs this, was kind of a pivotal moment into generative content.

5:58So we kind of just really saw the potential, like, you know, it's not just around base generation, but around we kind of see like the magic of Canva is integrating all the different steps and different parts of design into a simple page, as I like to call it. And so we really invested in our content library, in millions of templates that make it easier to start. And what we saw and got really excited about AI was that, firstly, we can offer all the amazing high-quality content for people to use. When the user might want something, they might want to have an idea that didn't necessarily exist. Maybe it has actually never been created in the world.

6:41Like, AI just gives us this superpower and ability to actually create things on demand specifically for what someone has in mind or in mission. And just kind of turn that idea, turn that search term or prompt into something they can use to express themselves. Right. And this is getting kind of into more of a meta conversation a little bit, which is fine. You're smiling, so we'll go there. But as I was saying, as somebody who primarily uses words to communicate, especially in work, speaking and writing, I have visual ideas. I just have the hardest time expressing them in a way that's close to what I'm thinking, let alone satisfying for the person I'm talking to.

7:22And so what you were just saying in my experience with being able to use words to try to express something and then the iterative process that you go through using an AI tool. It's a different take on the word accessible in the sense that we talk about accessibility, democratizing whatever the tool is. It got me all excited, as you could tell when I came in, because that ability that it translated it to something visual, but it was interactive and it could tweak it. And I just, when you were beginning to add AI features, and I don't mean to reduce it to, you know, but thoughtfully and going through that process, did you have an inkling into how much and how quickly the way that we think about creativity and access to these superpowers, as you called them, how quickly this would take off?

8:09Yeah, so I think probably for maybe the majority of people in the industry, the pace of advancements in AI can always be incredibly surprising. But I don't know creativity specifically. I haven't mentioned this before, but before I joined Canva, I actually was a freelance graphic designer. So I would do, I would, you know, I would retouch photos, I would design graphics and et cetera. And what I think is really, really interesting about AI how like there's a couple levels like for creative professionals, like there can be incredibly time-taking tools that lets you really scale your output, that lets you, that lets you, you're always working with a time constraint, you're always working with a budget.

8:53And what AI allows you to do and has allowed me personally to do is actually like utilize my creativity and deliver better or more scaled outputs with less time and also with a little bit less pain. and I think that area is really exciting yes but simultaneously like as we as you know as we think about camera and um all the 230 million plus users we have it's also a way that enables people who don't necessarily have so much creative like technical skills to do things that they just couldn't do before like like no maybe like a lot of people like um I'd like to think myself as a fairly fairly visual person but I know that so many people like they have ideas they have what want to create that they don't necessarily they didn't necessarily have the tools the pencil the paintbrushes to be able to do that right right can you get into some examples of tools that users have how canvas letting users tap into this creativity with you know an assist from ai how do you do you talk about co-pilots do you just talk about tools and now the tools have ai behind them how do you talk about it yeah so we have a lot of really popular tools and that's under our Magic Studio Suite, which is like a collection of integrated features and tools that really help people throughout the entire design process.

10:09The way we really think about it internally is a bit less about, say, co-opers, for example, but we really think about it as like AI is a superpower. Like this is something that helps our users and helps our people achieve tasks. What do they want to do? What do they want to get to? And this is a tool that helps them get there. Fantastic. Did Do you want to go into specific functionality in the Canva app or do you want to keep it high level? No, I'm really happy to. So some examples are really popular amongst our users, et cetera, for creating is really our photo editing features in Magic Studio.

10:44You can, for example, restyle a photo. You can change your background, change your shirt, add some balloons, make it really neon and cyborg, et cetera. So you can take, I guess, real photos, like something, maybe a product shoot, or maybe just kind of a shot of your office, for example, and just transform it and also experiment and brainstorm with some ideas really, really quickly. And it works on mobile as well, even a mobile app. Absolutely. What's the divide of mobile users? You have to go into specifics, but they're a strong mobile base. I'm just imagining people on their phones taking photos.

11:21Yeah, absolutely. It's roughly half-half, like half mobile, half desktop. Yeah, man, I'm old. I'm thinking in, you know, 70, 30, like there's no power in a phone. Yeah, it's amazing. My guest today is Danny Wu. Danny is the head of AI products at Canva. And we have been talking about the role of AI. I'm going back in Danny's career as an engineer, as a graphic designer, along his journey to now heading up AI products at Canva. kind of the development of machine learning and what we now call AI and how it's transforming superpower to use your words, giving everybody superpowers, creative superpowers to do things that just weren't possible before.

12:02Getting in a little bit to the technical aspects of how this all works and how Canvas is going about making it work, what's in your stack? What are some of the key AI technologies, some of the apps that are powering Canvas design tools? Yeah, that's a great topic. So we have a three-pronged and flexor approach to our tech stack and how we power all of our AI features. So first, like for specialized and unique and really high-value problems to us, we developed our own proprietary AI models and research. And last year, we acquired LittleNada.AI, an Australian financial model startup with the Phoenix model.

12:40And that's kind of an area that we really invest in for the problems that are quite, I guess, quite specialized and unique to our community and our users. But we're also very open and I'm very excited to be integrating all the best-in-class technology that's available in the world. Like it's really about how we, it's really about choosing the best tech available and bringing that into our product, solving problems, giving our users superpowers and making it accessible to all of our community. and that makes up for many, many of our AI features in Magic Studio. And finally, we also have our app ecosystem where developers around the world, they can use our APIs and SDKs to build things on camera, including AI cloud apps and features to kind of really expand what's possible to do on camera.

13:28And so the developers, they're leveraging your AI models, your AI infrastructure for the functionality of their apps that you were just talking about. Yeah, I mean, they can leverage our AI infrastructure, but they also don't have to. It is an open platform. And that's what makes it really flexible. That's fantastic. Yeah. That being said, still, when you were talking about your unique use case, I mean, offering, you know, it's uncharted territory in a lot of ways. It's a huge data problem in so many ways. And then you're operating at this scale. I think you said 250 million users? 230 million.

14:0230 million, excuse me. Yeah. Yeah. It's, yeah, you must be sort of figuring out, you know, building the plane as you fly it in some respects. just out at that frontier. So when you're working at that kind of scale with so many images and things being generated and modified and everything, how does Canva ensure kind of integrity, a level of quality, and how do you approach everything around AI and originality? Yeah, so scale is definitely like, it's a great place to be, but it's also an interesting challenge on so many levels. I think what's kind of really important to us is when we first started really cropping joint of AI into Canva, like we set out to really define like what are the values that we have and what are really our beliefs around AI and what all these areas like really came to, you know, how do we ensure that this accelerates human creativity, but also how do we ensure that like so all of our users, like they entrust Canva with the data, they entrust Canva with the designs and media and that's something that we've always like hold like really top of our mind, which is why, for example, like with a kind of creators program, we draw our priorities program where creators can, they can choose to opt out or if they want to stay opted in, they'll get paid priorities every single month if they want to make their works available for helping us improve our AI.

15:22Or additionally, like, you know, we, when you upload something on camera, you design something on camera, we don't train your AI on your data or designs unless you have given us consent. And many of our users like do give us consent. We spend a lot of time making sure that the the wordings they have, the dialogues, they're actually transparent and understandable, for example. But I think these kinds of values and principles is really important. And that's even more important when you're in this fast-paced environment and you have so many different teams that are building all sorts of these features.

15:54So just to go back and kind of clarify, confirm, it's an opt-in, it's off by default. Your data, your images you create, generate on Canva will not be used in training unless you opt-in. Yeah, so for users, it's opt-in. We don't train AI on your content without your explicit content. Fantastic. So how are the users liking it? How are, you know, like Canva's been around before? I mean, before a lot of these things were possible, let alone everybody's using Gen AI and everything. How's your user base doing with it? Yeah, so we have had pretty phenomenal adoption of our AI features. and the chart is actually always a bit crazy to look at.

16:36So far, we've had over 16 billion users of our AI products and features and that's continuing to grow. And we recently launched a few upgrades to the Visual Suite, including AI features. And I think what's really interesting is the adoption of these features in terms of usage and also in terms of what people create and publish with them is every new launch is just being adopted faster than before. So are there any kind of looking back, whether recently or further back, something internal related to building the tools and the infrastructure and research and everything that goes into it? Or something from the community that somebody did with one of the tools that just made you say, wow, success stories, feedback from users, things jump out as you kind of look back?

17:19Yeah, I mean, so one of the things we really like to do is to firstly like internally test and try out like all of the features that we're cooking up and working on very actively and just like use them for real work and everyday lives. And that's often actually helped us realize what we're building is actually like far bigger than what we thought in the first place, if that makes sense. It does, if even from afar, you know, but it does. Yeah, yeah. I think Time of Color is a pretty great example. We actually started that project with the goal of making it easier for designers and product managers to make interactive prototypes of new features.

18:00And then we started trying out internally and we started having some internal use cases. People started making things like thank you cards. Our education team got their hands on it and saw how powerful this is for learning and educational content. And that's when we realized, wow, this world is actually so much bigger. And when it comes to our users, a lot of us are very average just searchers and lurkers around all of our social channels and community channels, whether it's X, whether it's at the events we throw. And there's actually so many examples just about every week. Like we kind of realized new use case or some like new way that like how AI features are doing something we didn't necessarily imagine.

18:40So what comes next? Speaking of things we can't imagine, how do you see AI continuing to impact art and design visual communications, to use that term? How is AI going to shape the industry to come? I think that this intersection is like one of the fastest moving intersections. And I guess like maybe going back to how we think about it, which is on multiple levels of both increasing how just like everyone in the world can express and realize their creativity and turn that into outposts as well as how this can help creative professionals do their work bigger, better with more scope and more scale.

19:16I think like this thing probably like one very common theme that's been happening this year is really around increased control. So you kind of have, you know, image generation models or video generation models where you type in a text prompt, right? You type in a little description and chop down a few results. You get a few images. That's kind of a very low level of expressiveness and fidelity. They're trying to tell like very little information into something in mind. I think what's really exciting is like giving people the ability to add more inputs and add more direction and just like input more of creative control with whatever medium, whether it's, you know, whether it's reference images, whether it's sketching, whether it's just inspirations or like, oh, I really like this.

20:00I really like this part about this. I'm going for this vibe. And then memory, allowing these AI features to be less one-shot machines, but more a creative partner that works with you, that learns you, your preferences, your goals. I think that will really transform this interaction. Yeah, it's interesting to think about the split between professional work and other creative work and just the different perspectives on AI. But then to think about the way, you know, I use it differently, right, in those different things. And to your point at the end, having that longer memory to reference back, certainly in professional and personal as well, but it's a different kind of experience.

20:40Yeah, I know. Speaking of a personal use case, there was actually a story I heard a week ago, which I thought was really fun. One of our users was telling me how they were doing some landscaping work and they were using one of our features, Magic Edit, to actually mock up ideas for their yard and gardening. And I thought, wow, that's actually really, really cool and creative and would have never thought of it. I mean, I guess as long as you have the skills to actually, you know, trim and shape and cut the designs, that's sort of that last model, I would fall short myself. But that's amazing. Danny, this has been really fun and fascinating conversation that I'm excited is ending only because then I can get back to working on my interactive timeline on CanvaCode.

21:23But before we get to wrapping up, I wanted to ask you, in your own life, in work, life, personal life, what have you. Any AI tools or maybe tips and tricks for working with a chatbot, a prompt that you've found? Anything that you've been using lately, AI related, that you might want to share with the listeners because it's just really been great. Yeah, absolutely. So there's always so many like fun and just like new and exciting things in the AI world. And I personally love to really try everything. But something I can play around with a bit and just getting a lot of like delight form is actually really creating like personal agentic workflows using zapier and just connecting things like um you know like especially for things like whether it's reminding me or just like helping me like schedule things or actually even like planning out trips with ideas like from my friends and mates about what to do so it's a bit of an experimentation process and there's some moments where it's like incredibly magic and there's some moments where like you know it's all i need to do more work on this but kind of like moving from moving from chatbots to more like agents is something that just that has a lot of like really interesting and unique applications yeah excellent and you're able to to experiment in zapier yeah and it's and very cool that the kind of getting ai to be i'm not you know just replying to you with like text or maybe an image but actually being able to look up and take real world actions as you like that's something that feels magical it's a step forward absolutely danny for listeners who want to learn more about Canva, about any of the things we've been talking about, get hands on and experience it.

22:57It's fun, don't you? Where can they go online? Where would you directly start? Yeah, I'd actually really suggest our design school. So this is a free online resource. We can learn everything about both design, storytelling, marketing, but also especially AI and how to really like leverage all of our AI tools and non-AI tools to just do things, whether it's like storytelling, whether it's like teaching. And we've got lots of lessons and activities there. Canva Design School. Yes. Awesome. Danny Wu, again, thank you so much. This has been a pleasure. And best of luck on everything you're doing and your travels.

23:35And we hope to catch up again soon. Thank you so much, Tawa.

23:50Thank you.

24:22Thank you.