Hey there,

Canva just built an image model that thinks like a designer.

It plans the layout before it draws anything, and I think that changes what AI images are good for. Here's what's inside this week:

I build a custom CRM for my team with Claude Code, from a free template to a live app that syncs both ways with Notion, without writing a line of code.

This week we shipped SVG animations in Limora AI, the AI toolkit I'm building for designers.

My co-founder and I built it together in person, in Lombok. He's a 50-minute flight from me, and for a big launch that trip is worth it.

Side by side, we moved much faster than we ever do on calls.

Now you can generate an SVG, animate it and keep it on brand, all inside Limora.

You can edit it on the canvas, then drop it into your website or app.

Because it's code, it stays light and fast.

And this is only the first version.

On Thursday, Canva launched Craft, the first image model it built from scratch.

Most image models learn from photos. Craft learned from Canva's own templates, like posters, social posts, menus and type-heavy layouts.

So it plans the layout first. It decides the scale, the hierarchy and where the text goes, and only then renders the image.

You get up to 2K resolution, more than 30 aspect ratios, and readable text in 55 languages.

It's live now in Dream Lab inside Canva. Developers can also use it through the Canva API.

You can also turn any result into a layered design and edit it like a normal Canva file.

That's the part I find interesting. Most image models were trained to make pretty pictures.

Craft was trained on layouts. That's the hard part of our job: hierarchy, spacing and type that actually reads.

Adobe has Firefly, and now Canva has Craft. Design tools are training AI on real design work.

For designers, that's good news. The models get better at what we do, and they show up in the tools we already use.

Try it: Open Dream Lab in Canva and ask Craft for a poster with a real headline.

Then turn it into a layered design and fix what you don't like.

Read the announcement →

Framer · MCP connector
It connects Claude Code, Codex and Cursor to your Framer project. Your agent can build on the canvas, manage CMS content and publish the site. So you can build a client's site with AI, then hand it over in Framer, where the client edits text and images without touching code.

Free · Open source
Give it a motion-design video and it rebuilds the animation as clean SVG and GSAP code that loops and stays sharp at any size. It compares its frames with the source side by side and fixes the gaps, so you get a file of a few dozen KB instead of a heavy video.

Free · Open source
It's a copy-paste React library built on Base UI and Tailwind, formerly known as Origin UI. This month it added chat components, an avatar group and a segmented control, and it ships an agent skill so your AI uses it the right way.

Linear's website is famous for its clean line illustrations that react to your mouse.

Now a free skill called Hairline lets Claude Code draw them for you.

Open Claude Code and run npx skills add lucasmarkes/hairline. It takes a few seconds, and the skill is free.

Type /hairline-create and describe what you want, like a sales funnel. Claude suggests a few concepts, you pick one, and it draws the whole illustration.

The illustration follows your cursor, and you can turn up the intensity or switch to dark mode. When you're happy, embed the single HTML file straight into your website.

Want to see the full tutorial? The whole workflow is in the video.

This week's build comes from Dudu, who also writes the Toolfolio newsletter.

Shotbase is a Mac app that captures, edits and shares your screenshots and screen recordings. It can grab a full web page at any breakpoint, and it names and tags every capture with on-device AI.

I use it for my own screenshots every week.

Built something with AI?
Reply to this email with a link and I'll feature it here for free. Yes, for free, in front of nearly 8,000 designers and builders.

One question for you:
Would you let an AI model lay out your next poster?

Hit reply and tell me. I read every answer.

Know someone who should read this? Forward it, or send them to logiaweb.net.

See you next Monday,
Adrien