Amazingly simple design. Have a question? Or looking for some design tips? We’re here to help: https://xitter.twitterapi.xyz/t.co/aiUZ91aNpW 😊
Sydney, Australia
Joined April 2009
- Tweets227K
- Following382
- Followers341K
- Likes62.1K
Canva retweeted
In this episode, Stefano Corazza, head of AI research at @canva, joins us to explore how his team is building custom models and agents around iterative design workflows. We discuss the trade-offs between training models from scratch and adapting open weights models, and how proprietary design data supports specialization. Stefano explains how Craft separates layout planning from pixel generation and how Brush uses design tools and visual feedback to create and revise compositions. We examine the challenges of evaluating design quality beyond a single generation, including the use of vision language models as judges alongside human evaluation. We also discuss lessons from simplifying agent harnesses, why faster inference enables more iterations, and how model ownership creates opportunities to optimize serving costs. Finally, we explore how APIs and MCP are expanding access to these capabilities and what changes when software companies build tools for agents as well as people.
🗒️Full show notes: twimlai.com/go/784.
📖 CHAPTERS
===============================
00:00 - Introduction
01:19 - Canva AI Research, Craft, and Brush
04:25 - Why Canva Builds Custom Models
09:43 - Canva Craft Layout-to-Pixels Architecture
12:27 - Canva Magic Layers
14:54 - Training Data and Detailed Scene Descriptions
21:45 - Synthetic Data and Diversity in Design Generation
22:53 - Evaluating Image Generation Models and Design Editing
25:00 - Vision-Language Models for Evaluating Design Quality
26:30 - Tuning Model Taste and Visual Style
31:13 - Canva Brush
35:49 - Simplifying Agent Harness
37:36 - Code Generation
39:44 - Optimizing Inference Stack
47:54 - Distilling Craft for Speed and Cost
50:24 - Canva Models Available Through MCP and APIs
54:23 - What’s Next for Canva Research
We didn't want just any old letters.
Craft on typography... may have a few people talking.
Experimenting? Drop us your #CanvaCraft creations 🤩
Canva retweeted
Training your own model when you could just plug in someone else's is a bold call....
Canva did it twice in two weeks. Craft for images, Brush for editing designs, both built in-house.
When 265M people use your product every month, owning the model starts to make a lot of sense.
Who's going to copy this playbook next?
🤝 Paid partnership
Canva retweeted
tested brush in canva on one of my messy decks lol
it looks at the layout and layers before touching anything, so the changes actually make sense. canva building its own models is a bigger deal than people think
Canva retweeted
Canva now has its own AI vision-language model capable of editing existing designs, called Brush.
Users can describe a change, and Brush then plans and executes the edits across the design. According to Canva, Brush is 69% faster overall on general design editing tasks and 46% cheaper per task than frontier models.
🤝 Paid partnership
Canva retweeted
SHIPPING AI TO 265M PEOPLE A MONTH MEANS ONE THING: A HUGE INFERENCE BILL.
Canva even had to slow down some AI rollouts this year because demand exploded.
So they built their own:
→ Craft for images
→ Brush for design editing, faster and cheaper than general-purpose models
→ Quality that matched or beat their previous setup in internal testing
Cheaper edits = AI for more people.
Canva retweeted
Canva Brush is here!
For years, we’ve been working to make AI feel like a true design partner inside Canva, not something that takes over your work.
The first time I used Brush, it finally clicked: super fast, understood what I wanted to change, and kept what I loved intact.
We can't wait to see what you'll create with it!
Canva retweeted
last week: Canva ships Craft, its first image model. It debuts #4 on Design Arena.
today: Canva ships Brush, a vision-language model built just for design editing.
One week between them.
How many people still think of Canva as a template app?
🤝 Paid partnership · 🤖 Made with AI
Canva retweeted
underrated detail about canva’s brush launch: earlier this year they had to slow down some AI rollouts because demand exploded
that’s the real reason brush exists. a model built only for design editing, faster and cheaper than general-purpose alternatives, matching or beating their previous setup in internal testing
at 265M monthly users, efficiency decides who gets AI and when
🤝 Paid partnership
Canva retweeted
the best AI design tool isn't the one that generates the most.
it's the one that understands what you already made and helps you refine it.
that's why canva built Brush, its own vision-language model: for editing, made to keep you in creative control while it handles the restyling, the new formats and the multi-page consistency.
Canva retweeted
🚨 THIS IS A BIGGER DEAL THAN IT LOOKS
Canva just launched its second in-house AI model in one week.
Last week: Craft, which debuted #4 on DesignArena.
Today: Brush, built specifically to edit designs.
→ Restyles a full design
→ Adapts it to new formats
→ Keeps changes consistent across pages
→ Plans edits across the design
Earlier this year, Canva had to slow down the rollout of some AI experiences as demand exploded.
Now it's building specialized models to make AI editing more efficient.
You think Canva is just a design app.
They're building their own AI stack.
Canva retweeted
Canva is building its own AI models for creating images and editing existing designs!
Last week, Canva's model Craft debuted at #4 on DesignArena's image-gen leaderboard.
Now comes Brush, a vision-language model focused on design editing.
Canva says Brush can restyle existing designs, adapt them to different formats and apply changes consistently across multiple pages.
🤝 Paid partnership
Canva retweeted
most AI design tools aren't bad at the first prompt. they're bad at the second one.
"now make it fit an instagram story"
"apply this style to all 12 pages"
"swap the color, keep everything else"
"wait no no, not like that"
"annnnd now it's gone"
that's what canva trained brush for. a vision-language model built only for design editing, running under canva AI.
general models get you the draft. specialized ones finish it
🤝 Paid partnership
Canva retweeted
🚨 Canva just dropped its SECOND in-house AI model in a week.
People who still think Canva is a template app are about about to be very wrong.
→ Craft: their image model, debuted #4 on Design Arena, behind only OpenAI
→ Brush: a vision-language model built only for editing designs
The first draft was NEVER the hard part. The 30 edits after it are.
🔖 Save this one.
🤝 Paid partnership
Great design happens in the edit. Brush is trained to support every step from draft to publish.
We started from a leading open-weights model, post-trained on real design and editing tasks using a combination of design expertise and anonymized insights from people who opted in to help improve Canva AI.
More about how we built Brush on our blog: canva.com/newsroom/news/canv…
A few ways you can start putting Brush to work for you:
🎨 Restyle an image
🪄 Make changes across a multi-page design
🔄 Remix and repurpose assets for different channels
Available for Canva Pro subscribers, with more availability coming soon.
Start working with Brush in Canva AI today: canva.com/ai
