Canva Launches Craft, Its First In-House Image Model

By
Canva Launches Craft, Its First In-House Image Model
Canva Launches Craft, Its First In-House Image Model
Affiliate disclosure: This post contains affiliate links. If you buy through them, I may earn a commission at no extra cost to you. Full disclosure

Canva released Canva Craft on Oct. 22, its first in-house image generation model, and every Canva user can run it now in the Dream Lab tab.

For an Ecommerce Paradise reader running a high-ticket Shopify store, the question is narrow. Craft was trained on Canva’s library of human-designed templates, not photographs, so it is a graphics engine for spec sheets, banners and ad overlays, not a replacement for manufacturer product photos. It also runs straight into a Google Merchant Center rule that can get listings disapproved.

Below: what Canva shipped, why it built the model, where it helps a store selling $2,000 to $20,000 products, and the five checks I’d run before a single Craft image touches a product feed. If you need the basics first, start with my breakdown of what high-ticket dropshipping is and how to start.

AI tools can make a store look established in a weekend, so make the business behind it official too: form your LLC with Northwest Registered Agent, the best LLC formation service for ecommerce owners. Northwest never sells your address and puts its own address on your public filings, which the cheap formation sites that upsell you can’t say. Form your LLC with Northwest →

Canva Craft Debuts in Dream Lab With 2K Output, 30 Aspect Ratios

Canva announced the model on Thursday, Oct. 22, 2026, according to TechCrunch. Canva built Craft with its own research team to compete with image models from Google, OpenAI, ByteDance and Meta.

Craft works in two stages, per TechCrunch. The first stage reads the prompt and builds a plan covering spatial relationships, relative scale, visual hierarchy and the text that has to appear. The second stage renders an image up to 2K resolution in any of 30 aspect ratios.

Canva says the approach improves object detail and makes text more legible. It also says Craft can edit one part of an image without disturbing the rest, according to TechCrunch.

The training data is the unusual part. Canva trained Craft on human-designed templates from its library, such as posters, social posts, presentations, infographics, menus and typography-heavy layouts, rather than on photographs. Images can then be converted into layered, editable designs with Canva Magic, the company’s image-to-design tool.

Availability is where the reports split. TechCrunch says all Canva users get Craft in Dream Lab and developers get it through the Canva API. Another report, from Traders Union, says API access is coming soon. Treat the API as unconfirmed until Canva’s developer docs say otherwise.

Canva also plans to put Craft inside third-party chatbots such as ChatGPT and Gemini through its MCP connector, per TechCrunch. Two cheaper variants, Craftfast and Craft-ultrafast, are in development and due “soon,” with no dates.

On quality, AI Weekly summarizes preliminary Design Arena results that rank Craft fourth on that independent leaderboard, ahead of entries from Google, Microsoft, Meta, xAI and ByteDance. The same summary calls the ranking preliminary and offers no per-task scores. It is one leaderboard, early, and I’d weigh it that way.

What nobody has published yet: pricing, usage limits, and whether Craft output is free or metered inside paid Canva plans. TechCrunch’s report does not say.

Canva Cut Its 2026 Growth Target to 20% Over AI Costs

The model makes more sense once you see the money. In early August, Canva cut its 2026 revenue growth expectation to about 20%, down from the 30% target it set at the start of the year, according to B&T. Quarterly revenue still rose 25% year over year to $921.9 million.

The cause was AI serving costs. Co-founder and CEO Melanie Perkins wrote in a letter to shareholders that “we were relying too heavily on frontier models” and that “several of our first-party models were not yet ready for release,” per B&T. Serving AI requests at scale was not economically sustainable, the report says.

Canva then spent about three months rebuilding its AI infrastructure and training more models of its own. B&T reports that the cost of processing a single AI task fell by almost 90%, and that Canva’s style-transfer, image and video models are 23x, 30x and 17x cheaper than comparable frontier models. Those are Canva’s figures as relayed by B&T, not independent measurements.

TechCrunch adds two pieces of background. Canva acquired Leonardo.ai in 2024, and Leonardo researchers contributed to Craft. External analysts reportedly put Canva’s valuation at $35 billion, down from $42 billion. TechCrunch frames Craft as a way to cut API bills as well as raise quality and speed control.

Here is the counterpoint. A model trained on templates is built for layouts, not for rendering a $9,000 leather sectional under real light. Canva’s own descriptions point at posters, menus and infographics. Nobody at Canva has claimed Craft replaces product photography, and nobody should assume it does. Vendors are also pushing AI deeper into ad creative everywhere, as I covered when Meta made advertisers liable for its AI ad edits.

What Canva Craft Means for a High-Ticket Store’s Product Images

This section is my analysis, not reporting.

My read is that Craft is a cost story first and a creative story second. Canva needs cheaper image generation to protect its margins, so you should expect a free or cheap tier to get generous and then get fenced in with credits as usage grows. Build a workflow that works on any tool, not one that only works on Craft.

For a high-ticket store, the buyer is often 50 or older, spending thousands, and reading everything. They trust a manufacturer photo with a model number they can look up. They do not trust a floating sofa that looks generated. Keep manufacturer images as your primary product images and put Craft where text and layout do the work.

That means size and fit guides, delivery timeline graphics, financing banners, comparison charts, collection headers for your Shopify store and ad overlays. Email headers for Klaviyo flows are another easy win.

It is the same work you already hand to a designer or build in Canva. If you are weighing outside help for it, 99designs is the paid-designer alternative to benchmark Craft’s output against.

The Google Merchant Center rule you can’t skip

Google requires that AI-generated images in Merchant Center carry the IPTC DigitalSourceType value trainedAlgorithmicMedia, and it prohibits removing that tag, according to Google’s Merchant Center help page. The rule covers image_link, additional_image_link and lifestyle_image_link. PPC Land reported when the rule arrived that noncompliance may lead to product disapprovals or other account-level actions.

I could not confirm from any report that Craft embeds that tag in its exports. Canva has not said either way. If it doesn’t, a Craft lifestyle image in your Shopping feed is a policy problem you created. If it does, an image optimizer or compression app that strips metadata can delete it for you.

Google keeps tightening image handling in feeds. I wrote about Google’s 2026 feed spec coming for your images in July, and PMax image extraction defaults on for new campaigns on Dec. 4, which I broke down in the PMax image extraction post. Every image on your product pages is now a potential ad asset.

Rough math, clearly hypothetical

Say you run 300 SKUs and want one spec or size graphic for each. That is a hypothetical, but a normal one. At a hypothetical $25 per graphic from a freelancer, you are at $7,500 and a few weeks of back and forth. If Craft gets you to a usable first draft in two minutes and your editing takes ten, the same 300 graphics cost about 60 hours of staff or VA time. At $8 an hour through OnlineJobs.ph, that is roughly $480. Those numbers are invented for illustration. Your rates and your reject rate will move them.

The number that matters is the reject rate. Use thresholds. If at least 8 of 10 test graphics come out with correct, readable text and no invented details, Craft earns a place in your infographic workflow. If 6 or 7 pass, use it for backgrounds and headers only, and keep a human on every number. If fewer than 6 pass, stay with your designer and recheck when Craftfast ships.

Wrong text is the expensive failure. A graphic that says a table is 72 inches wide when it is 60 turns into a return and, on a $3,000 order, possibly a dispute. Spec numbers, dimensions, warranty terms and shipping times should be typed by a person, not generated.

One more reason for caution: Craft is a version-one model. Canva itself says more models are coming, and the cheaper variants will probably trade some quality for speed. A prompt that works today may produce different output after the next release. Save your best prompts and the exported files in a shared folder so a model change never breaks a live listing.

Also decide who owns approval. In a small store the person generating graphics is usually the person publishing them, and that is how a wrong dimension ends up on a product page at 11 p.m. A second pair of eyes on every spec graphic costs minutes and prevents the returns that eat a high-ticket margin.

Where this fits the bigger AI shift

Image models are one layer. Shoppers are starting to use AI assistants to research and buy, which is why I tracked Shopify opening its catalog to personal AI agents. Walmart did the same on the retail side, as I covered in the Walmart AI agents post. Those agents read your product data and images. Clean, accurate, correctly labeled assets will matter more, not less.

Running all of this yourself is a lot: feeds, creative, ads, tagging rules, product pages. That is the work my team handles for turnkey done-for-you stores, and it is where most owners lose weeks they could spend on supplier relationships.

Would you rather have a team handle store creative, feeds and image rules while you focus on suppliers and sales? See the turnkey done-for-you service →

Test Canva Craft on 10 Products Before Black Friday Ads

Black Friday is Nov. 27, so you have about four weeks. Five moves, in order:

  1. Run a 10-product bake-off. Pick ten SKUs, write one prompt per product for a size graphic or comparison chart, and generate them in Dream Lab. Score each on text accuracy, layout and brand fit. Compare against what your designer or your own Canva templates produce today.
  2. Check the metadata before anything reaches Merchant Center. Export a Craft image, open it in a free metadata viewer, and look for trainedAlgorithmicMedia under the IPTC DigitalSourceType field. If it is missing, add it with an IPTC tool or keep that image out of your feed. Then confirm your image compression app does not strip it.
  3. Keep manufacturer photos as the primary image. Use AI graphics in additional image slots and on your own pages, not as the main image_link for a product. My guide on turning Google Shopping clicks into sales covers why the first image decides the click.
  4. Take specs out of the prompt. Generate the layout and background with Craft, then type dimensions, wattage, weight capacity and warranty terms yourself in a Canva text layer. Have one person check every number against the supplier sheet.
  5. Freeze your creative by Nov. 15. Lock the images you will run through Cyber Monday, then work through the high-ticket Black Friday stress test. Do not swap in untested AI assets mid-campaign.

If you want a second set of eyes on your creative and feed setup, book a call through my discovery page. Also set a reminder to read Canva’s pricing page when Craftfast ships, because that is when the real cost shows up.

Frequently Asked Questions

What is Canva Craft?
It is Canva’s first in-house image generation model, released Oct. 22, 2026, and available in the Dream Lab tab. It plans the layout first, then renders up to 2K resolution, per TechCrunch.

Can I use Craft images as my main product photos?
I wouldn’t. Craft was trained on design templates, not photography, and a high-ticket buyer wants to see the real product. Use it for graphics and extras, and read my brutal truth about high-ticket dropshipping for why trust beats polish at these price points.

Do AI images break Google Merchant Center rules?
Not by themselves. Google requires the IPTC trainedAlgorithmicMedia tag on AI-generated images and prohibits stripping it, per Google’s help page. Check each export for the tag.

How much does Canva Craft cost?
Unknown. Canva has not published pricing or usage limits, according to TechCrunch. Plan for it to be metered at some point, given Canva’s own AI cost pressure reported by B&T.

Should I use ChatGPT or Claude instead of Canva for store graphics?
Use whichever gets accurate text fastest in your test. ChatGPT and Claude are better for writing the spec copy and product descriptions you will then place in a design. I covered a lean AI setup in Claude for ecommerce.

What if I haven’t picked a niche yet?
Do that before you spend a minute on creative. Grab my free list of 1,000+ niches at ecommerceparadise.com/niches. Then read the high-ticket niches list for the reasoning behind the picks.

Is AI creative worth it for a brand-new store?
For cheap graphics, yes. For product trust, no. The beginner guide is the free starting point. The first-year budget breakdown shows where early money should go, and creative is rarely the first line item.

Want to work through AI creative and feed rules with other store owners and me inside the community? Join the Skool community →

That’s the Canva story. Subscribe to the YouTube channel for daily breakdowns. More breaking news coming through the day.

Related Articles

If this was useful, these go deeper:

Share on Facebook
Post on X
Save
Share