# AI Design Still Feels Like a Slot Machine

Markdown copy of https://realalpha.app/v/heE80uxOmus/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Stefano Corazza says Canva’s custom models and agents make design AI faster, more editable and built for iteration.

- Show: The TWIML AI Podcast
- Hosts: [Sam Charrington](https://realalpha.app/people/sam-charrington/)
- Guest: [Stefano Corazza](https://realalpha.app/people/stefano-corazza/) is head of AI research at Canva.
- Published: Nov 10, 2026
- Length: 57 min

## The best minutes

- [5:09](https://realalpha.app/v/heE80uxOmus/?t=309) [Canva's data makes smaller design models more economical](https://realalpha.app/s/canvas-data-makes-smaller-design-models-more-economical/) (3 min)
- [16:34](https://realalpha.app/v/heE80uxOmus/?t=994) [Detailed scene descriptions help smaller models create better designs](https://realalpha.app/s/detailed-scene-descriptions-help-smaller-models-create/) (5 min)
- [39:35](https://realalpha.app/v/heE80uxOmus/?t=2375) [Owning the model stack creates more room to cut inference costs](https://realalpha.app/s/owning-the-model-stack-creates-more-room-to-cut-inference/) (4 min)

## Also worth watching

- [10:58](https://realalpha.app/v/heE80uxOmus/?t=658) [Layered designs let people move elements before regenerating pixels](https://realalpha.app/s/layered-designs-let-people-move-elements-before-regenerating/) (4 min)
- [22:51](https://realalpha.app/v/heE80uxOmus/?t=1371) [Design editing needs benchmarks beyond single-shot generation](https://realalpha.app/s/design-editing-needs-benchmarks-beyond-single-shot/) (4 min)
- [27:04](https://realalpha.app/v/heE80uxOmus/?t=1624) [Style mismatches can be fixed without retraining from scratch](https://realalpha.app/s/style-mismatches-can-be-fixed-without-retraining-from/) (3 min)
- [31:14](https://realalpha.app/v/heE80uxOmus/?t=1874) [Brush revises designs through tools and visual feedback](https://realalpha.app/s/brush-revises-designs-through-tools-and-visual-feedback/) (5 min)
- [35:48](https://realalpha.app/v/heE80uxOmus/?t=2148) [A simpler agent harness lets the model use familiar code paths](https://realalpha.app/s/a-simpler-agent-harness-lets-the-model-use-familiar-code/) (4 min)

Skip the video-model ecosystem detour, the learning analogy about children, and the closing research outlook.

## Full analysis: Stefano Corazza says Canva is 40X cheaper on the same task

- Corazza says Canva is 40X cheaper than the most expensive frontier models on the same design task.
- He ties that figure to smaller models, design data and vertical integration, but gives no comparison method.
- For teams, the test is cost and quality across revisions, not image price alone.

Its 6 sections, what it means for you and what to watch are on [the page](https://realalpha.app/v/heE80uxOmus/#analysis), free with an account:

- [3:54](https://realalpha.app/v/heE80uxOmus/?t=234) Creative AI must make revision cheap
- [6:14](https://realalpha.app/v/heE80uxOmus/?t=374) Canva's 40X cost claim rests on narrower models
- [11:11](https://realalpha.app/v/heE80uxOmus/?t=671) Craft moves revisions into editable structure
- [23:21](https://realalpha.app/v/heE80uxOmus/?t=1401) Current benchmarks still underweight design editing
- [31:24](https://realalpha.app/v/heE80uxOmus/?t=1884) Brush uses a tool loop, not one prompt
- [41:16](https://realalpha.app/v/heE80uxOmus/?t=2476) Full-stack control can lower serving costs

## Summary

Corazza leads AI research at Canva. His team built Canva Craft and Canva Brush to support image generation, design editing and iterative creative workflows.

Corazza says design AI needs to support fast iteration and human control, since models may get 70% or 80% of the way to a result but people still handle the last mile. He argues Canva’s design data and vertical integration let smaller models deliver quality at lower cost, while Craft separates layout planning from pixel generation and Brush acts as an agent across design tools. He describes evaluation as a gap in the field: current benchmarks focus on single generations, while design editing needs structured rubrics and human judgment. He also says simpler agent harnesses, faster inference and APIs for agents can make Canva’s capabilities more useful and affordable.

### Minute by minute

- [1:15](https://realalpha.app/v/heE80uxOmus/?t=75) **Design AI needs iteration.** Corazza says Canva had roughly 100 AI models in production, with Craft and Brush marking larger steps in its design workflow. He argues that AI may get 70% or 80% of the way to a creative result, so people need a fast, enjoyable way to make revisions.
- [4:21](https://realalpha.app/v/heE80uxOmus/?t=261) **Custom models can cut costs.** Corazza says Canva’s unique design data and deep integration of models into its product made building its own models worthwhile. He says a model with two orders of magnitude fewer parameters can beat a larger one when trained on the right data, and that Canva’s models cost 40X less than the most expensive frontier models for the same design task.
- [9:34](https://realalpha.app/v/heE80uxOmus/?t=574) **Craft separates planning from pixels.** Corazza says Craft was trained from scratch, while Brush started from an open-weights model. Craft’s three stages amplify the prompt, convert it into structured scene data, then generate pixels; an LLM can intervene in the scene plan or produce layouts without generating images.
- [12:27](https://realalpha.app/v/heE80uxOmus/?t=747) **Layers make edits more controllable.** Corazza says Magic Layers decomposes flat images into elements that people can move or resize before regenerating pixels. He says this gives designers more control than asking an image model to reposition something when the image itself has no information about its size or location.
- [14:57](https://realalpha.app/v/heE80uxOmus/?t=897) **Design data shapes model quality.** Corazza says Canva’s decade of design data, including licensed material and user data used with permission, helped produce a strong training set. The team also expanded its structured scene descriptions, adding details such as bounding boxes and relative positions, and found that a larger diffusion model could capture them.
- [21:40](https://realalpha.app/v/heE80uxOmus/?t=1300) **Diversity and editing need better tests.** Corazza says Craft’s training mix includes stock images, synthetic data, user-generated data and millions of Canva-created or licensed designs, with diversity intended to avoid repetitive outputs. He says most benchmarks test single generations, not design editing, and calls for tests that start from existing designs and assess edits, speed and cost.
- [24:57](https://realalpha.app/v/heE80uxOmus/?t=1497) **Rubrics help assess design quality.** Corazza says vision-language models can judge qualities such as readability and composition when given clear rubrics, alongside human evaluation. He says Craft initially made cartoon images too realistic and 3D-looking, and that the team can tune the model’s taste and style through fine-tuning and reinforcement learning.
- [31:02](https://realalpha.app/v/heE80uxOmus/?t=1862) **Brush works across design tools.** Corazza describes Brush as an agent that understands layered designs and can edit their elements, make them more readable or professional, and respond to broad instructions such as making something feel more modern. He says it can also handle higher-level tasks, including creating a cohesive 30-slide deck.
- [35:37](https://realalpha.app/v/heE80uxOmus/?t=2137) **A simpler harness gives agents room.** Corazza says Brush is on its third harness, and the team has worked to make it thinner and simpler so the LLM can use design tools more freely. Brush can write code for tool operations and API calls; the team chose representations partly by considering what kinds of code LLMs are already trained on.
- [39:35](https://realalpha.app/v/heE80uxOmus/?t=2375) **Inference efficiency expands access.** Corazza says lower inference costs can let Canva’s free users, including students and nonprofits, use more AI. He says the team has seen improvements such as fivefold speed gains over two weeks and lists optimization across hardware, kernels, model steps, distillation, serving, batching and caching.
- [43:03](https://realalpha.app/v/heE80uxOmus/?t=2583) **Model ownership unlocks optimization.** Corazza says Canva runs its models on cloud infrastructure and manages their optimization and serving, without owning the hardware. He says access to model weights and collaboration across AI research, graphics and cloud engineering enable more optimization; he expects inference costs to keep falling.
- [47:56](https://realalpha.app/v/heE80uxOmus/?t=2876) **Distilled Craft models trade speed for fidelity.** Corazza says Canva plans to release Craft Fast and Craft Ultra Fast, using distillation and other optimizations. He says owning both parts of Craft lets the team tune the language and diffusion models together, while giving users a choice between faster, cheaper versions and higher fidelity.
- [50:20](https://realalpha.app/v/heE80uxOmus/?t=3020) **Canva is opening models to agents.** Corazza says Craft is available through Canva Connect APIs and Canva’s MCP, alongside other capabilities. He describes demand from enterprise customers building custom workflows and from people using assistants through MCP, and says agents need structured information, easier previews and less friction to test services.
- [54:18](https://realalpha.app/v/heE80uxOmus/?t=3258) **Canva will adapt as AI changes.** Corazza says Canva’s direction is to make AI more accessible, embed it in iterative creative work and remove workflow barriers. He expects specific projects and models to change every few months as the team looks for bottlenecks and opportunities.

## Transcript

The TWIML AI Podcast, hosted by Sam Charrington, with Stefano Corazza.

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[0:00](https://realalpha.app/v/heE80uxOmus/?t=0) As good as modern image generation models have become, designing with them can still feel like playing games. Put the right magic words in the prompt and you might win at the slot machine. Try to iterate on the result and you're playing whack-a-mole. That's fine for memes, but when design is your job, the game gets old. Canva's AI research team wants to make that process more predictable, with greater control over layouts, easier revisions, and models fast enough to keep up with the creative process. To get there, they've trained their own family of models using Canva's design data and integrated those models right into their tools.

[0:35](https://realalpha.app/v/heE80uxOmus/?t=35) Here's Stefano Corazza, head of AI research at Canva, on what his team is trying to change. You have to build around iteration. AI will get 70%, 80% there. It will never get to 100%. And so you have to factor in that for the last mile, you need a human in the loop and be able to iterate in a way that is fast and also fun. I'm Sam Charrington, and this is the TwiML AI podcast. For over a decade, I've been exploring the ideas and innovations shaping the future of AI through conversations like this one that help you understand what's real, what's next, and what matters. Let's jump in. This has been a pretty busy week for you and your team.

[1:23](https://realalpha.app/v/heE80uxOmus/?t=83) Well, first of all, the team was announced. Was this like the debut of AI research at Canva? Yeah, that was the big debut last week, right? I think 12 hours before we actually announced Canva Craft, the model. And we also announced Canva Research. We had like a blog post, a website with a lot of information about who we are and what we do. So it was really fun. And then this month has been the busiest month probably of my life, where like every week we are announcing a new model. And so exactly 40 minutes ago, we had this announcement about Canva brush. So very exciting. This visual language model for design, creation and editing.

[2:08](https://realalpha.app/v/heE80uxOmus/?t=128) Excellent. And so have there been more models than just those two over the course of the month? Yeah, so we have a total of 100 AI models in production roughly right now. And some are smaller and may not be a huge large language model or anything like that. But we started injecting, infusing AI in all parts of the product and to really help this design workflow in every possible way. But now we are taking bigger steps with Craft, the foundational model for image generation. And then with Brush, we are completely revolutionizing the way people can create content And also, we are revolutionizing how fast and affordable that is.

[2:53](https://realalpha.app/v/heE80uxOmus/?t=173) Because those two parameters, pioneers using AI are okay waiting a minute for the magic moment where something magical gets generated. But if you want to really be using the day-to-day, that's not an option. In game development, there's a rule. Anything that takes more than 10 seconds is completely out of the question. and we wanted to get to like instantaneous AI. Yeah, I mean, particularly with the way that you need to iterate when you're creating, like those delays really stack up. Yeah, absolutely. And people, I think there's a little bit of Silicon Valley BIOS that I've seen myself being part of that BIOS that you think like the new tech comes and it's going to be single shot, perfect, and then people would just adopt it. But the reality is never the case, especially if the task has some creativity in it.

[3:50](https://realalpha.app/v/heE80uxOmus/?t=230) It's not just a purely technical task. And so you have to build around iteration. AI will get 70%, 80% there. It will never get to understand. And so you have to factor in that for the last mile, you need a human in the loop and be able to iterate in a way that is fast and also fun. So that's a key part, I think, of what we do. Until the moment when we will have telepathic helmets that you can wear and you can transfer the full amount of your intent and your vision. But that is a little bit far out, I think. You know, we're going to dig into the models and what you've created there, talk a little bit about how, but I'm trying to decide how interesting it is that you even did this.

[4:38](https://realalpha.app/v/heE80uxOmus/?t=278) Like, on the one hand, you know, there are lots of, you know, their Frontier Labs have models and they, a big part of their businesses are allowing companies like Canva to use those models as part of broader processes. There are open weights models that you can build on. And so there's a lot that you, Canva, could have chosen to build on. On the other hand, it's like, maybe it's obvious that a Canva with all of the data and user interactions that you have access to would build a model. Do you think it's interesting or do you think it's obvious that you did it? No, I think it's very interesting because I think it also applies to other companies maybe of similar size.

[5:20](https://realalpha.app/v/heE80uxOmus/?t=320) when basically there's two factors. So first of all, you have unique data that nobody else has. And then the second part is deep vertical integration of AI models in your stack is going to provide incredible performance and cost reduction. Those two factors are basically the driving input. And I'm sure there's a lot of companies that are like in the, you know, thousands of employees that are even smaller that will do that and where it makes sense. And so for us, it was kind of the discovery of the fact that you can have,

[5:56](https://realalpha.app/v/heE80uxOmus/?t=356) **Stefano Corazza:** you can fine tune or train a model that is two orders of magnitude fewer parameters if you have the data and you can beat the bigger model.

[6:06](https://realalpha.app/v/heE80uxOmus/?t=366) And then because it's smaller, you're going to be like another magnitude faster and cheaper, right? So that is the belief that we had. And right now we are in the 40X cheaper than the most expensive frontier models. They can do the same design task, but they are using others or many more parameters to do it. Instead, we have smaller models that we can fine-tune, we can do reverse learning with unique data that we have, and we can basically squeeze a lot more. Same level of quality, sometimes better, but most importantly incredible speed and cost reduction. So that's one part. The second part is that the product itself with AI is evolving dramatically, right? And the way AI is helping humans generate design is really changing. And so there has been a lot of like, I would say siloed innovation where like, okay, I have an image model, the image model creates an image. Then I have a next step, an LLM looks at it and thinks, oh, we need to change something, goes back to the image model and says, hey, can you make something different?

[7:16](https://realalpha.app/v/heE80uxOmus/?t=436) And like in that back and forth, you lose control and you're also throwing away information. And so we have, with Craft, we have created like an image model that most of the work actually is creating a scene graph of what the design should look like. And then the conversion to pixel is just the last step that we train from scratch. But at the layout, let's call it the design layout stage, the LLM can already intervene and make changes before the pixels are generated. And so we are seeing that there's incredible product differentiation by integrating the LLM world and the diffusion model world, if you want, in a more cohesive way.

[8:00](https://realalpha.app/v/heE80uxOmus/?t=480) like the iterations can happen earlier and you have all non-destructive modifications with humans in the loop instead of having to bake the result, show it, iterate, and then to basically throw away in that process a lot of information every single time. So those were, I think, the two main factors that drove our decisions. And also, if you look, the top models are now open source or open weights on the image generation space. I think video is a little bit of a different ecosystem. Actually, this will be probably an episode on its own, how the ecosystem for video model and image generation model is completely different, sometimes upside down.

[8:43](https://realalpha.app/v/heE80uxOmus/?t=523) It's kind of fascinating. In what sense? Without making it a full episode, in what sense? Something happened about, I think, four months ago where some of the companies developing video models started releasing open-weights versions. And that had a fantastic domino effect where it forced all the competitors to do it. And so now you have an ecosystem where a lot of the image models are not open-weights, but the video models are. And then because of that, I think, also because of that, you see the cost per second of AI video generated like really plummeting because there's a whole community that is trying to figure out how to crack that problem.

[9:31](https://realalpha.app/v/heE80uxOmus/?t=571) And because there's a lot of models that are open weights, more people can apply to that problem. And so it's a fascinating different dynamics on the two ecosystems, if you want. With Kraft, did you start from an open weights model and train it, tune it, or is it a from scratch train? So for craft, we started, it's like train from scratch because we realized that there were some several breakthroughs that we believe were going to unlock a much more powerful model and those were not just there in the existing model. So that one we trained from scratch. For Canva Brush, which is our vision language model, We started from an open source, open weights model, and then we built on top of that.

[10:20](https://realalpha.app/v/heE80uxOmus/?t=620) So you're using whichever of those makes the most sense given the time or given the particular use case. And it sounds like with Kraft, I have covered this idea of a kind of two-stage generation with layout and then rendering. It was in a conversation with Qualcomm, who has some papers in that area. But I don't believe, to my knowledge, there are no popular open weights versions of that. Are the two parts of a model like that, are they trained independently or end-to-end? Or what's the system approach to training a model like that? Yeah, that's a great question. So the answer is both. So we train them individually, but then there's always a time where we're trying to look at the end-to-end.

[11:11](https://realalpha.app/v/heE80uxOmus/?t=671) But I would say that the steps are actually three now. First, we take all the input and the prompt or what the context may be, and we try to amplify and make it as powerful as possible. Then there's a process of taking that and converting into basically a structured form of data, more like a scene graph, if you want. And then the last step is generation of the pixels where our model training from scratch has shown really, really good performance. And by breaking down into these steps, you're also allowing for a better integration with agents, with LLMs, because there could be some components of that pipeline.

[12:00](https://realalpha.app/v/heE80uxOmus/?t=720) They could be useful for some tasks. If I say, hey, can you just make me 10 different layouts of this single page, right? I don't need the LN to generate pixels. You can use, you know, the craft design planning, can create different plans, can have a human in the loop to confirm and then can do all the generations, for example. Right, so there's actually, we are seeing new workflows coming out from breaking down into these steps. It seems like that would also help with editability um yeah historically you have a scene and you say oh make the person a little bit bigger and the thing you get back is totally different like we're a lot better now uh like the gpt 2.5 image models are really really good at consistency between uh revisions but it does strike me that if you could change the scene graph and move the thing over or make the thing bigger and not have to re-render, that that would help with consistency as you're trying to move things around.

[13:05](https://realalpha.app/v/heE80uxOmus/?t=785) Like 100%. We believe on this so strongly that we release magic layers at the beginning of the year, which is basically taking a flat image decomposing in layers and elements so that people can really have that fine control. I mean, how many times have you tried to say to any image model generation Oh, can you move that thing a little bit to the right or like two feet? But in that image, there is no information to know what is two feet, you know, or two inches or centimeters.

[13:37](https://realalpha.app/v/heE80uxOmus/?t=817) **Stefano Corazza:** And so it's a lot easier for a designer to just move that element to the right and then regenerate.

[13:43](https://realalpha.app/v/heE80uxOmus/?t=823) And so our beliefs is with craft, basically, the humans and the designers can edit elements, can make them bigger, smaller, move them to the left or the right and then regenerate the pixels later. And we also are extending that capability to images that are generated outside Canva using magic layers. So actually the majority of images that we process with magic layers come from outside Canva and people generate them somewhere else and then they come into Canva and they want to edit. And so there we can decompose into layers and elements and people can start moving things around and have that fine level of control.

[14:25](https://realalpha.app/v/heE80uxOmus/?t=865) They always want, I mean, people want consistency and control, right? Those are the two things that any creative has asked AI for the last, I would say, two years. And we came quite far along, but like, it's still, you know, we don't want to provide slot machines, you know, where like, hey, if you generate 20 times, you might get lucky, right? We want to actually make sure the time investment that people are putting into it is really controlled and the result is predictable. You kind of referenced earlier two dimensions that were important, quality and efficiency. Talk a little bit about each of those.

[15:04](https://realalpha.app/v/heE80uxOmus/?t=904) We can start with quality. The model performs well, like on Design Arena is one of the number four now. Is that right? Yeah. What you brought to bear to create a high quality model, what went into that result? I think we underestimated the power of like a decade of design generated and design data that we had. I think we were even ourselves a little bit surprised to rank that high because it was also the first submission that we did. So we have, of course, internal evals and we do a lot of also human evaluation internally, but you never know until you actually get into the real world arena. But yeah, it was basically a lot of work over the last year, really collecting data that makes sense for designers and really harvesting data that Canva has been.

[16:00](https://realalpha.app/v/heE80uxOmus/?t=960) created for over a decade now, including also data from users with their permission or data that we license from the outside. And I think we put together like a pretty good body or a pretty good training set. And I think deliver a lot of value. And also we had like several breakthroughs about how to use the data and how to train the model. So I think that was another, you know, the ability to run, hundreds of experiments and really figured out what worked and what provided the quality. I think that was a great several breakthroughs, but one on top of the other that I think made the difference.

[16:41](https://realalpha.app/v/heE80uxOmus/?t=1001) Can you give us some examples of the kinds of things you learned that really moved the needle for you? Without getting too much into the weeds,

[16:51](https://realalpha.app/v/heE80uxOmus/?t=1011) **Stefano Corazza:** we underestimated how detailed of a scene description you can give to a diffusion model actually if the diffusion model is above some number of parameters can handle like incredible

[17:05](https://realalpha.app/v/heE80uxOmus/?t=1025) detail of the scene so you can the size of our scene description really grew in the last six months significantly as we realized oh actually you can be a lot more accurate a lot more precise and so we expanded basically not only the initial prompt of course but like the description of the seen is now to a level of detail that I think no human will want to actually go and look through and edit, you know, unless it's like visually editing, right? And so there's so much detail and by increasing a little bit the size of our diffusion model that we trained from scratch, we were actually able to capture all that detail, all those nuances and really, you know, squeeze the maximum quality out of it.

[17:52](https://realalpha.app/v/heE80uxOmus/?t=1072) And is that a unique property of your particular diffusion model that you trained? Or is that just a general observation about diffusion models? I think it can be generalized for more cases. In our case, I think because we understand so well the coupling between the generation of the scene and the diffusion model, You know, we really work to make sure, you know, we didn't over generate on the scene description or, you know, or under train on the diffusion model. And so the two are like perfectly calibrated to be the input and the ideal input and output model of each other. Are the scene descriptions only unstructured text or are there also some degree of structure, coordinates, bounding boxes, sizes, things like that?

[18:44](https://realalpha.app/v/heE80uxOmus/?t=1124) Yeah, oh, they are absolutely structured. Yeah, it's a fully structured, it's still text-based, but it's fully structured with all the details that you just said, bounding box, relative position, transfer position, like all that. We have found that the more detail we can provide to a composition, really the better. And we have the ability to basically render all that detail in the decision model. Also, one thing that I want to mention is which is great for the creative industry overall, I think. There is this race to build the most capable LLM out there, right? And now we are already in the tens or hundreds of trillions of parameters.

[19:30](https://realalpha.app/v/heE80uxOmus/?t=1170) But for creative tasks, I think we have found that, let's say, 100 billion, it's already a pretty large number that can, And if properly tuned and post-trained and fed the right data, a model of that size can already achieve so much. And we are seeing also a lot of other models, also video models that leverage LLM as an input to create that structured data like the MiniMax. H3 is another model that has a similar approach, even though it's a video model. But we are seeing the size of the basically vision language model or LEM that is the input, you know, not growing 10x every three months anymore, right, but being kind of stable.

[20:18](https://realalpha.app/v/heE80uxOmus/?t=1218) In an industry where like everything grows, you know, 10x every month, like that number has shown that like, you know, you don't need more than 100 billion parameter to do most of the creative tasks, I would say. Probably video understanding maybe needs just a little bit more because it's probably the most complex type of task, but for everything else, below 100 billion parameters, you can already achieve wonders. And so to be clear on what you're describing there, in this two-stage model where you have an LLM like thing or an LLM period generating a scene description you're saying there's kind of diminishing returns beyond 100 billion and smaller models do just as well as larger models and kind of nailing that scene description and providing the right input to the diffusion model. That's correct and while being you know a lot faster and a lot cheaper right and so just the economics they don't work if you use like the latest frontier model to do that task, then, you know, it will be like a dollar per image generated or something like that. And so this, I think, together coupled with the ability of people to train those smaller models in very efficient ways, I think really unlocked a lot of value

[21:40](https://realalpha.app/v/heE80uxOmus/?t=1300) and a lot of usefulness of AI in the creative workforce. To what degree did you use synthetic data in the training process? Is that a big part of your recipe? Yeah, it is. I think like in every data set of this size, we also want to make it as diverse as possible. And so we have all kinds of data coming in from stock images and that kind of stuff to synthetic data, to user-generated data, to designs that we created. We have millions of designs that Canva created or licensed over the last decade. And so that was also, of course, a big part of it. It kind of helps also with the taste, right? Because we have seen AI slope and AI generation where everything looks very similar.

[22:28](https://realalpha.app/v/heE80uxOmus/?t=1348) And so I think we put so much effort on diversity, which also aligns with our company value. And so we are a company that wants to empower everybody to design. We are global. We serve every market, every language. And so we wanted to really push the diversity to make sure that the output also of the model is as diverse as possible. Talk a little bit about the way you think about benchmarks for image generation. You know, there are well-established benchmarks, but I always wonder, you know, how well they capture kind of the human perspective and human judgment and taste, as you just referenced.

[23:11](https://realalpha.app/v/heE80uxOmus/?t=1391) And I'm wondering, you know, someone who works on this constantly, like how do you think about, you know, the applicability of the popular benchmarks, how you supplement them, that kind of thing? That's a great question. I would say the benchmarks overall are lagging a little bit behind to where the power of AI in design is going to be because they're mostly focused on single shot, the first generation.

[23:36](https://realalpha.app/v/heE80uxOmus/?t=1416) **Stefano Corazza:** And you have benchmarks for image editing, which is great, but design editing is different than just image editing.

[23:44](https://realalpha.app/v/heE80uxOmus/?t=1424) And so as of today, There is no established benchmark for how good a model is editing a design, for example. There's some public benchmark for generating, let's say, a presentation, like there's a present bench. In one, you know, has like 200 examples of like presentation-related slides, related tasks. And you can test it, but again, it's new generation. And so there's not a lot of benchmarks that really can assess the effectiveness, editing and design, the speed and the cost also to go with that. So I hope there's going to be more. Of course, we have internally, we have a lot of those, But yeah, we haven't found an industry-established benchmark for the ability to actually edit a design because the industry has focused on, oh, AI is so magic, it's going to single-shot everything, which we know is not the case.

[24:38](https://realalpha.app/v/heE80uxOmus/?t=1478) And so I think now the understanding is finally landing and I hope we're going to see more benchmarks that really start from like a given set of pre-existing designs and defines rubrics for how those designs can be modified and then does the comparison, the benchmarking on that. That kind of implies that you can use an LLM or a VLM as a judge to look at an image and assess it visually, you know, from any number of perspectives. Are we there yet? Or, you know, how good are VLMs at kind of articulating the qualities that you might want to optimize an image for? I think they're pretty amazing. And we have also to keep in mind that if you structure the rubrics really well, they are really good at defining, I don't know, is this design readable?

[25:39](https://realalpha.app/v/heE80uxOmus/?t=1539) Like, what's the score of readability? What's the score of layout and composition? So we have very well articulated rubrics about design quality. And then we can use VLMs to judge that. And we get quite far along. Of course, we always do also human evaluation and some of the arenas are strongly, you know, leveraging that because it's the ultimate test. But I think, yeah, we came a long way. It forced us also to really define what a good design means, right? And so if humans are not able to articulate that, then VLMs are not going to be able to score it, right? And so I think we had to really go back and really understand what makes the design really good to be able to leverage VLMs to do it.

[26:25](https://realalpha.app/v/heE80uxOmus/?t=1585) But I think now we are getting to a place where it's quite reliable. When you think about this two model approach in craft, what are the qualities of an image come from having the two separate systems or models and what just come from having better training data? Yeah, that's an interesting question. We have seen just different behavior depending on different tasks, for example, which we are always trying to understand. Is the limitation the diffusion model or is it the, you know, the length side of things? And sometimes it's hard to know. we have seen for example the model perform really well on like portraits and people and also nature landscapes. We have seen the model doing not so great on the cartoon type of content for example. Is it not so great? I mean it's not ranking in the top three and so that's an area there.

[27:31](https://realalpha.app/v/heE80uxOmus/?t=1651) So like all that is together that is counterintuitive to me because um you know a portrait for example seems like it would kind of null the value of like the planning and the scene component like it's a portrait you know and a cartoon for example like you know you've got all these things in a cartoon you know that that would lean a lot on a a scene generating component. But the two results that you're seeing suggest that that model shouldn't be as strong as it is or that that's not the source of the strength of the model, right? Am I thinking about that correctly? Yeah, it's interesting. So the number one thing for people in portraits that you want is basically to be as photorealistic as possible, right?

[28:19](https://realalpha.app/v/heE80uxOmus/?t=1699) Like you don't want to have that AI skin effect and all that stuff. And we put so much effort in this because we were like, we got to nail it, it should not look like AI generated. I think that effort may have bled into the cartoon side where instead you weren't very stylized.

[28:38](https://realalpha.app/v/heE80uxOmus/?t=1718) **Stefano Corazza:** And so it turns out that the images, the cartoon images generated with craft initially were a little bit too realistic and too 3D looking.

[28:49](https://realalpha.app/v/heE80uxOmus/?t=1729) But again, this is something that, you know, we discover when we put it on the arena because probably they had like a different population of designers that would test it. But at the same time, it's something that is more about the taste of the model and that is a lot simpler, a lot easier, a lot faster to tune and correct. And so that we are in the process, basically, of doing exactly that. And, you know, tuning the taste of the model and the style of the model is more of a fine-tuning and RL type of task. You don't have to pre-train, right? In the pre-training, we, of course, had a lot of cartoon content.

[29:29](https://realalpha.app/v/heE80uxOmus/?t=1769) Just now, that capability didn't surface to the full extent. And so now we are like teaching the model, okay, this is a style that resonates more with humans and we want, you know, if this cartoon is to be flat-shaded and these lines and so on, I mean, it's fascinating, honestly. It's a journey of learning. As much as the model is learning, we are learning how to train the model and how the model behaves. Sometimes you'll find this maybe a little bit weird, but sometimes I have two small kids and sometimes I learn things about learning as a high-level concept that I learn by training the models.

[30:14](https://realalpha.app/v/heE80uxOmus/?t=1814) I'm like, oh, this I'm sure applies also to humans. you know and so it's kind of a fascinating especially on pre-training if you see pre-training is basically like teaching a human as many words as possible about vocabulary right but has nothing to do with the style of talking performance or like if that human becomes i don't know a stand-up comedian the the how many words it knows has nothing to do with like the style of the delivery Right. And so I think we, the concept is so true for pre-training and post-training and AI model. And it's also true for like humans where like in the early years, you want as much variety and as much exposure to different type of activities, sensory information, experiences, languages, and so on.

[31:06](https://realalpha.app/v/heE80uxOmus/?t=1866) Right. And so the parallel actually is fascinating. I think I'm sure this book's already written about it. Let's talk a little bit about Brush. Is Brush, what's the best way to think about it? Is it an editing model? Is it a design agent? A little bit of both of those things? Yeah, we call it like a vision language model, but it's basically an agent that can create design and can also edit design. So it has a deep understanding of what's going on in the layout of your design and can really take very fine-grained directives from humans and then operate on the design itself. So what is a design in this context?

[31:49](https://realalpha.app/v/heE80uxOmus/?t=1909) It's not an image, it's, think of it like a slide or a Canva document, multiple, a layered composition of things? Yeah, it could be multiple. Layers could be multiple elements, could also different type of elements. I can have some vector graphics and some image mixed up together and then some text as font. So it's basically this heterogeneous composition of different elements. They have particular features and particular type of details that humans can tweak and also the agent can tweak. And then they all come together to create that unique design that is the final output. And so it's a little bit like what craft generates in the first step, right?

[32:37](https://realalpha.app/v/heE80uxOmus/?t=1957) It's basically all the underlying structure of a design. Brush really understands that very well. It can really act on a design and then relay out, can move parts, can make it more readable, more professional, can take high-level inputs like, hey, make this thing pop. Or like, make it cool. I'm like, you know, make it Gen Z. Like these are, you know, directives that will be extremely hard for a designer that is not maybe a Gen Z designer to really understand, you know. So those are high-level directives instead can really be applied by Brash in a quite successful way. Like I'm kind of thinking about it like, so with craft, you've got this scene creator and you've got this generator and the generator, you know, generates one kind of one image and brush, you have the same scene layout thing, but maybe it has different tools and that, and one of those tools is generating images and it's got, you know, layout tools and the output is like a composition of sorts.

[33:47](https://realalpha.app/v/heE80uxOmus/?t=2027) Is that a reasonable way to think about it? Or where does that way of thinking about a breakdown from what's actually happening. Yeah, it's not that far. Like I would say right now, Brush can do more general things than just operating a layout description of the scene. But basically it operates a set of tools, like a large set of tools that are all design tools. It can make decisions on design. I think it goes beyond the image generation and editing, right? It can also create like, you know, 30 slides deck, can make sure that deck is cohesive with like branding. So I think you operate like a higher level of abstraction than craft, which is really about image and soon we'll be also doing more of image editing.

[34:35](https://realalpha.app/v/heE80uxOmus/?t=2075) And so, yeah, Brush is basically a full agent that you can ask for like more complex work and more like abstracted type of instructions. And then, you know, we are constantly iterating on, And I mean, it's a lot is like having basically an LLM in the loop using a bunch of tools and be able to see what it generates. And that loop is really what has changed everything because until, I don't know, a year ago, LLMs would just do like, we're blind and we'll just do a single shot and then be blind to the result, right? And instead now, not only that single shot becomes an iterative process with human in the loop, but LLMs now can see what they generate and can iterate

[35:18](https://realalpha.app/v/heE80uxOmus/?t=2118) **Stefano Corazza:** and the faster you make that iterative loop, the more effective everything is

[35:24](https://realalpha.app/v/heE80uxOmus/?t=2124) because now an ILM can do maybe 10 iterations before it even shows the result to a human and before every iteration would take a minute and so that was now feasible. But now that's why speed is not just a nice to have. Speed is like a complete game changer because it can really increase the quality, it can increase the user experience, it can really make it like an iterative process. Sounds like there's a simpler way to think about it. Craft is a model and brush is kind of an agent. And so in creating brush Like I'm imagining like part of the task here is like a custom harness. And I'm wondering what all that entailed.

[36:07](https://realalpha.app/v/heE80uxOmus/?t=2167) How did you approach that harness development? This is such a great question because I think people are like, usually they are all focused on the model and not the harness. But I think the importance of the harness cannot be overstated. And for us, this probably was the hardest part and the ones, the part where we get the most iterations. In fact, we already like our third generation of hardness for brush. And really like the high level goal is basically to give as much freedom to the LLM to really run free of constraint and really be able to like tap into the tools and have as much insight on what is able to generate.

[36:54](https://realalpha.app/v/heE80uxOmus/?t=2214) And so we have really focused on making the hardness are thinner and smaller and more agile and more simple, right? And then at the end, you know, a brush can write code to operate tools and make design operations.

[37:10](https://realalpha.app/v/heE80uxOmus/?t=2230) **Stefano Corazza:** And so we were at the same time trying to leverage how good these LLMs are at writing code and at the same time simplifying the harness in the most basic possible way

[37:20](https://realalpha.app/v/heE80uxOmus/?t=2240) to really unleash the full power of the LLM. And so that's where we did a lot of iteration. And actually, it was more about like removing the human, you know, knowledge and just letting the LLM like run free to some extent. And when you talk about the code generation, what kind of code are we talking about? I can imagine a lot of what Brush might want to do is running a Python code, but I could also imagine you creating some design DSL that more natively represents the things that you care about. Yeah, it can be that, or it can be just writing code also to operate API calls and layout operations and that kind of stuff.

[38:05](https://realalpha.app/v/heE80uxOmus/?t=2285) So is it primarily code or a little bit of both? It's definitely both, yeah. Like the capabilities are quite broad. What's an example of something that a user would do in brush that uses, you know, either code or this kind of some kind of DSL or abstraction? I mean, a lot of the editing operation that you can do on an editor, they can be expressed as like editing through an editing API, right? And the problem is there is there's like infinite ways of describing that design operation. and let's say I take an object, I move it to the left and I rotate and I make it bigger, right? Like these three operations can be encoded as a format change.

[38:51](https://realalpha.app/v/heE80uxOmus/?t=2331) You know, you can have a format that describes the whole state or you can describe them as diffs and they can be achieved through like an editing API. And so that is the complexity that we had to navigate trying to understand like what was more natural for an LLM, which actually has to do with what is the most available data out there that the LLMs are trained on, right? So this way, you kind of leverage a groove that has been already created by an enormous body of data instead of trying to teach something new or like that is underrepresented in the training set or is out of distribution in the training set of an LLM.

[39:36](https://realalpha.app/v/heE80uxOmus/?t=2376) And so I think that's where, you know, to align with the wind right and and the capabilities as opposed to trying to like fight it earlier i mentioned like kind of these two north stars you had quality and efficiency and i think we double clicked on quality but didn't really double click on efficiency recap your kind of vision for efficiency uh and you know talk through in the case of crap and brush like, what were the levers that you manipulated to drive that efficiency? Yeah, efficiency in inference specifically really enables affordability of AI, right? And the majority of our users are actually free users of Canva and we love that, like students, non-profits can use Canva for free.

[40:22](https://realalpha.app/v/heE80uxOmus/?t=2422) If we can bring the cost of AI inference low, we can also allow them to use more of AI. And so we really wanted to make sure AI was something that was within reach for everybody. And so the optimizations, what we have learned is we are overall relatively early in the journey of optimizing AI inference. If you can squeeze 10x performance improvement in a month, means you're very early, right? We are so far from the plateau right now and that's very exciting because it's also very satisfying for our AI engineers to be able to work on a problem for like two weeks and be like, oh, okay, it's five times faster than it was two weeks ago, right?

[41:08](https://realalpha.app/v/heE80uxOmus/?t=2468) And so we have put a lot of effort

[41:11](https://realalpha.app/v/heE80uxOmus/?t=2471) **Stefano Corazza:** and you can only make it really efficient if you own the full stack, right?

[41:18](https://realalpha.app/v/heE80uxOmus/?t=2478) Because there are optimizations you do at the GPU level, at the hardware level, the optimization you do at the kernel level, this optimization reducing the number of steps the model does, there's distillation. And so going from basically high level to low level hardware, and how you serve it, how you batch it, how you cache it, there's a pile of things that you can really do. And then if you tackle all of them, And you're going to squeeze like significant gains out of each step. And that's how we have seen, for example, the cost of generation of a second of 720p video in literally over a year go down two orders of magnitude.

[42:05](https://realalpha.app/v/heE80uxOmus/?t=2525) Right now you can generate some cent where like a year ago it was like half a dollar per second of generation. And so there's still a lot to do there. I'm not going to forecast that AI inference is going to be as cheap as storage but we have seen what happened with storage at the beginning we were paying by the gigabytes and it was so expensive and you had to do your math and I don't even know how many terabytes I have of storage in my Dropbox account I don't know I don't think I will ever have the problem to think about it and so I don't think AI inference will get to that level but it's getting so much better than it was a year ago.

[42:50](https://realalpha.app/v/heE80uxOmus/?t=2570) And I think given the first derivative is definitely still very healthy, I think we're going to see probably another order of magnitude going down. And when you get to like 0.1 cent per generated image or generated second of video, I think it's another breakthrough. And all of a sudden, the democratization and affordability go really beyond what we see today. In building that full stack and operating that full stack, are you operating the full stack? Do you have hardware in a data center or do you delegate part of that to the cloud but you're running everything on top of cloud providers, clusters?

[43:28](https://realalpha.app/v/heE80uxOmus/?t=2608) How far down do you go? Yeah, we are as far down as we can go other than owning the actual hardware ourselves. So we run those models on our cloud, we optimize them, we manage how we optimize the inference, how we serve it, all that. So the raw thing that you're getting from someone else's GPU instances and you're clustering them, you're loading them with whatever your inference. is you're tuning LLVM and all the bits and pieces to batch sizes and all the stuff you talked about. Yeah, exactly. Talk about how has your knowledge about that space evolved? Were there things that you found surprising once you started to work on these new models or did it kind of roll out as expected?

[44:24](https://realalpha.app/v/heE80uxOmus/?t=2664) Yeah, it's interesting because it was an industry that was completely dominated by AI researchers, if you want, where you had graphics engineers or cloud engineers coming in and being like, oh my God, there's so much more juice to squeeze here. And so it's been really phenomenal to see how, and here in San Francisco, I saw a whole wave of startups actually focusing on inference optimization. I mean, because there is so much money left on the table right now. And so there's been this like massive wave. I think I can probably name like a dozen of companies that do that because, you know, people that have a different background realize, okay, these models are really not run efficiently.

[45:12](https://realalpha.app/v/heE80uxOmus/?t=2712) We should rethink about how we feed all this data and these jobs into the GPU. So there's been like really like a flurry of interest because it's also a very simple value proposition to explain. This thing costs you now like a billion dollars and if I do this optimization, it's going to cost you one-tenth of that. Everybody gets it. It's very easy to explain and it's a very easy ROI to put in place. And we are seeing now, as AI adoption is still skyrocketing, we are seeing a lot of pressure on reduction of cost and making it scalable. And so it's been fascinating to see a lot of those startups and also in big companies bring in knowledge that is more about graphics. And graphics has been optimizing things for graphic cards for decades. And now that knowledge is brought to combine with cloud scalability knowledge and AI models. All those three components together are really delivering wonders.

[46:15](https://realalpha.app/v/heE80uxOmus/?t=2775) It's quite exciting. And are there prerequisites to being at the steep part of the optimization curve meaning do you have to have a certain degree of scale so that you can manipulate batch sizes or do you have to like do you have to build your own models such that you can take advantage of you know deep levels of control where do you think it makes sense to like own that full stack it was really like an interdisciplinary type of endeavor where we actually put together different teams that had different skill sets, right? And when we saw the combination of those come together, that's where like the best gains came.

[47:00](https://realalpha.app/v/heE80uxOmus/?t=2820) And yeah, absolutely. Like if you have to host, you know, you have to self-host your model and then you have to have access to the weights of the model. You know, if you have, you know, if you're using like an open weights model, that's, you know, that's good. But in some cases, you don't have access to that. So in that case, you build your own model. But when you're self-host and when you have access to the weights, there's a lot more that you can do and a lot more juice that you can squeeze out of those models. And again, we found that by bringing together a very interdisciplinary team with different backgrounds, we were able to really tackle the problem with a different perspective.

[47:42](https://realalpha.app/v/heE80uxOmus/?t=2862) And again, it goes back to diversity, right? in this case it was a very diverse mindset from the AI researcher that we brought in and I think it really delivered. Going back to Kraft and that kind of two-part model does There's... having a unique model like that get in the way of some types of optimizations like distillation? No, distillation is something that we do pretty regularly on every model that we develop. And so for Kraft specifically, we have announced that there's going to be a quick follow on Craft Fast and Craft Ultra Fast. And those are basically models where we are doing also the distillation among many other optimizations.

[48:30](https://realalpha.app/v/heE80uxOmus/?t=2910) And again, by having all the models in-house, we can also make sure that the distillation that we do on the LLM side and the one that we do on the diffusion model side are also tuned to work optimally together. So again, it goes back to that concept of making sure that the input and output are really tuned for the best quality. And so we're going to have more distilled the versions of Craft coming out in the next few weeks. Because again, we want to make it even more affordable and we want also to leave the choice to the user, whether it is a developer using an API or a designer actually using to the UI or to MCP. We launched it. I think now it's also available on MCP as of yesterday, Craft. And that's another place where if I'm building a prototype app, I don't want to run on the most expensive model, of course.

[49:29](https://realalpha.app/v/heE80uxOmus/?t=2969) And so the ultra-fast and faster version of the model I also designed. For that, they may become kind of the default, but then when you want quality, then you switch to the highest fidelity model. I'm sorry, and so the MCP is a Canva MCP or a Craft MCP? Through the Canva MCP, you will be able to also access Craft. Okay. Yeah, there's basically a bunch of like AI capabilities that we are exposing to MCP and the craft is being added. And then as we add more variants of craft, we're going to keep adding to MCP, which by the way, we are seeing like the usage of MCP on camera like really skyrocket.

[50:15](https://realalpha.app/v/heE80uxOmus/?t=3015) I think people are finding a lot of value. So we want to give more capabilities, you know, access to more capabilities to MCP going forward. Do I understand correctly that beyond the MCP access to Kraft, you are also making Kraft and some of the other models directly available to folks? I don't know the specifics, but I'm envisioning like via an open router or via Canva directly, but just per use access to the model? Yeah, absolutely. So we released it through our API called Canva Connect. And so now it's available. You can access through the API as well as MCP. So for developers that want to do that.

[51:00](https://realalpha.app/v/heE80uxOmus/?t=3060) That feels like a very different business, like running a super low latency, very narrow model inference service, as opposed to this is an API for our product. Is this the first model that you're hosting in this way? We have exposed APIs for various capabilities. I think also you can access Magic Layers and some of the image generation models that we had before, like Lucid Origin. But I think now the level of interest we are getting is a lot higher. And yes, you're right. It is a type of different business, if you want, right? But we are getting so much demand coming from mostly like two categories.

[51:42](https://realalpha.app/v/heE80uxOmus/?t=3102) So enterprise customer, our enterprise business is the fastest growing business that we have in the company. really triple-digit growth year-over-year. And we are getting a lot of demand also from those enterprise customers to access the capability. to, you know, an API because they may build their own custom workflow. Now that people can vibe code anything, right, there's going to be more demand for that. And then we are getting also a lot of demand from people using, I don't know, Cloud or ChargeGPT through MCP. They now want to access more capabilities, right? So the agents can go there and say, oh, you want to generate an image?

[52:22](https://realalpha.app/v/heE80uxOmus/?t=3142) I can use Canva. And then they can always then open in Canva for further editing. And so we are seeing these two type of use cases really growing. And sometimes those users are already using Canva and they just want to use it in a different way or they just want to build scalable applications using some of the Canva technology. And so we have seen a lot of demand. And yes, it's a new and exciting business and we're still learning a lot about how to best serve it. But I think this, we can say, is not unique to Canva. It's across the industry, right? There's a lot of companies that all of a sudden are like, oh, we are basically building not only for humans, but also for agents, right?

[53:04](https://realalpha.app/v/heE80uxOmus/?t=3184) Even though humans is our focus even for AI features, but there's going to be in the future just more agents using some of those capabilities. I think that's unavoidable. What do you think those agents need that's different from what humans need? I think they need more structure. They need documentation defined in a certain way so they can digest more things. They need the ability to try to preview things without having to subscribe. And so a lot of things that basically are designed to remove friction, finding information more easily, streamline workflows because ultimately many times it will be the agent that will decide what solution to use as opposed to the human.

[53:52](https://realalpha.app/v/heE80uxOmus/?t=3232) And so we have to basically make sure, for example, the agent is able to like spin up with all friction and not a lot of burden on like initial credentials or credit or purchase. And it can spin up something that the agent can test, verify it's working. And then, you know, go back to the human and be like, hey, I created this like a prototype app using this API. I tested it, it works. You and I give it a shot. Okay, now it's approved. Now we can go for the real scalable thing. So that's the kind of approach. Just to kind of wrap things up, where do you see Canva evolving in terms of research in particular?

[54:31](https://realalpha.app/v/heE80uxOmus/?t=3271) It's a world where typically you plan a year ahead and now it's completely impossible. It's like just a month ahead is already like a pretty foggy kind of thing. But at a high level, we have a very clear mission and vision, that one is not changing. So we want to make AI more accessible. We want to make it more part of the iterative creative process. And we want to remove any barrier. And we want to reimagine all those workflows. So that's the direction that we've been pursuing. And what we're actually doing, what models we work on, specifically at Canva, yeah, this is going to change every few months.

[55:19](https://realalpha.app/v/heE80uxOmus/?t=3319) And so I, you know, I told my team, basically, don't fight, you know, this ever-changing, constantly changing ecosystem, but just like embrace it. And so I think it's really like a mind shift of like everything is going to change, which means also like what is important today may not be important tomorrow, which opens up also for opportunities, right? Like, it's a double-edged sword to some extent, but at the same time, we see that as a big opportunity. And so we are constantly, you know, analyzing our business, see where there are opportunities, where there's bottlenecks, where there are. technical parts of our workflow that we can automate and then we prioritize those and we work on them.

[56:06](https://realalpha.app/v/heE80uxOmus/?t=3366) Like a year ago, none of the projects, the models that we ship now was underway, you know? And so it's logic to think that in one year we will probably release models that we haven't even started working on it today. And that's, you know, it could be terrifying or could be very exciting. So I think researchers are probably the most exposed to the speed at which AI is changing And so it takes really like some major adaptation and mindset shift, but at the same time, it's quite exciting. Well, Steph, thanks so much for taking the time to share a bit about what you and your team are up to. It's very cool to hear more about it.

[56:46](https://realalpha.app/v/heE80uxOmus/?t=3406) Thank you, Sam. It was a true pleasure and honor to be on your podcast. Thank you so much. Thank you. Thank you.