> ## Content Index
> Fetch the complete content index at: https://bullquill.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Canva Craft AI image model is Canva's first built in-house
- URL: https://bullquill.com/canva-first-in-house-ai-image-model/
- Published: 2026-10-22T22:14:35.000Z
- Updated: 2026-10-24T07:13:27.000Z
- Description: Canva built and trained Craft itself for design work, not just pictures. It is available in Dream Lab and through the Canva API, with cheaper Craft-fast and Craft-ultrafast versions planned.
- Author: Rowan Ashby
- Tags: AI, Trending, Canva, Canva Craft, Generative AI, Image Generation, Leonardo.Ai, #lang-en, #tr-canva-first-in-house-ai-image-model

## Key takeaways

- Canva on 22 October 2026 released Canva Craft, the first AI image model it has built and trained itself, according to the company's announcement and reporting by TechCrunch.
- The Canva Craft AI image model works in two stages: it first plans the layout, scale and hierarchy of elements, then renders the image at up to 2K resolution.
- Canva says Craft was trained on human-made design templates such as posters, menus and social posts, not on photographs.
- The model is available in Canva's Dream Lab and through the Canva API; Canva says agent access through its MCP server is coming, without giving a date.
- Canva expects its own model to cut what it spends on outside image models, and plans cheaper "Craft-fast" and "Craft-ultrafast" versions.

Canva has launched the Canva Craft AI image model, the first image generator the Australian design company has developed in-house. Before it draws anything, Craft plans where text, objects and graphic elements should sit, then renders at up to 2K resolution.

## What is the Canva Craft AI image model?

Canva describes Craft as "the first foundation model built and trained from scratch by Canva Research", according to its announcement as quoted by Thurrott.com. TechCrunch reported that researchers from Leonardo.Ai, the image-generation startup Canva acquired in 2024, contributed to the work.

Most text-to-image models are trained largely on photographs and produce a single flat picture. Canva has taken a different route. Its company says it trained Craft on millions of templates made by human designers in its content library, including posters, social media posts, presentations, infographics, menus and layouts heavy on typography. The aim is a model that understands how a design is put together, not only what a scene looks like.

### How the two-stage process works

According to TechCrunch, Craft handles a prompt in two steps:

1. **Planning.** The model reads the user's instructions and builds a plan covering how elements relate to each other in space, their relative size, the visual hierarchy, and what text needs to appear and where.
2. **Rendering.** It then generates the image at resolutions up to 2K, with support for up to 30 aspect ratios suited to different platforms.

Canva says this approach produces accurate object detail and legible text, two areas where image generators have long struggled. The company also says users can make precise edits to one part of an image without disturbing the rest. These claims come from Canva and have not yet been tested independently at scale.

### From picture to editable design

Craft's output is still an image, but Canva connects it to its editing tools. TechCrunch reported that a picture generated with Craft can be turned into a layered, editable design using Canva Magic, the company's image-to-design transformer. Canva released its own layer-aware design-understanding model in 2025, according to the same report. In practice, that means a generated poster could be split back into separate text and graphic layers that a designer can adjust. Marketplaces are drawing lines on AI content: [Etsy's AI rules](https://bullquill.com/etsy-bans-ai-generated-patterns-recipes/) now ban AI-generated patterns and recipes.

## Canva Craft at a glance

| Feature             | What has been reported                                                               | Source                         |
| ------------------- | ------------------------------------------------------------------------------------ | ------------------------------ |
| Release date        | 22 October 2026                                                                      | TechCrunch; Canva              |
| Developer           | Canva Research, with contributions from Leonardo.Ai researchers                      | Canva via Thurrott; TechCrunch |
| Training data       | Human-designed templates (posters, social posts, presentations, infographics, menus) | TechCrunch; Thurrott           |
| Generation method   | Layout plan first, then image rendering                                              | TechCrunch                     |
| Maximum resolution  | 2K                                                                                   | TechCrunch; Thurrott           |
| Aspect ratios       | Up to 30                                                                             | TechCrunch                     |
| Where to use it     | Dream Lab in the Canva app; Canva API; MCP server "coming soon"                      | TechCrunch; Thurrott           |
| Planned variants    | Craft-fast and Craft-ultrafast (no release date)                                     | TechCrunch; Thurrott           |
| Early ranking claim | Fourth on the DesignArena image leaderboard, behind three OpenAI models, per Canva   | Thurrott                       |

Reports differ on who can use the model. TechCrunch said Craft is available to all Canva users through the Dream Lab tab, while Thurrott.com reported that access in the Canva app requires a paid subscription. Canva had not published pricing for Craft at the time of writing.

## Why Canva is building its own image model

Until now, Canva's image generation has rested on outside or acquired technology. When Canva launched Dream Lab in October 2024, it was built on Phoenix, the foundation model developed by Leonardo.Ai, which Canva had bought that July, according to Capital Brief and Gigazine. In April 2025, OpenAI said Canva was exploring its gpt-image-1 model for design generation and editing in Canva's AI tools.

TechCrunch reported that Canva currently uses several image models and expects Craft to lower its API costs, while giving it tighter control over quality and speed. The planned Craft-fast and Craft-ultrafast versions are meant to deliver images at "a fraction of the cost of leading frontier models", according to Canva's announcement as cited by Thurrott.com.

Cost matters for Canva's business. TechCrunch noted that the company was on track for about $4 billion in revenue last year, has cut its 2026 growth forecast, and has reportedly been valued by outside analysts at about $35 billion, down from $42 billion. Running its own model could cut a recurring expense on every image its users generate.

## How Craft compares with Adobe Firefly and other rivals

Canva's closest rival in design software, Adobe, takes a mixed approach. Alongside its own Firefly models, Adobe offers partner models inside Firefly and other apps, including image models from OpenAI and Google, according to Adobe's help documentation. Canva, by contrast, is now putting its own model at the centre of its image tools.

On benchmarks, Canva says Craft debuted fourth on DesignArena's image-generation leaderboard, behind three OpenAI models: GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT Image 2, as reported by Thurrott.com. Leaderboard positions change quickly and rely on user votes, so the ranking is an early signal rather than a settled result.

What sets Craft apart, if Canva's claims hold, is its focus. General-purpose models from OpenAI, Google or Meta aim to cover any kind of image. Craft is tuned for layouts with readable text and a clear structure, which is what most Canva users make: social graphics, flyers, slides and menus.

## What it means

For designers and small businesses, the practical test is whether Craft can produce a usable first draft that needs less fixing. Text that renders correctly and an image that can be split into editable layers would save time over flat images that must be rebuilt by hand.

For Canva, the launch is a strategic shift from licensing image technology to owning it. That could reduce costs and give the company more control over how fast it ships new features, but it also means Canva must keep pace with far larger AI labs on quality.

For the wider market, Craft adds another sign that software companies want their own models for their core tasks. Canva's plan to offer Craft through its API and, later, through its MCP server would put the model in reach of developers and AI assistants outside Canva's own app. No date has been given for the MCP release or the cheaper variants, and independent tests of Craft's text and editing claims are still to come.

Related reading on Bullquill: [how to build an MCP server in Python](https://bullquill.com/how-to-build-mcp-server-python/), [prompt engineering techniques that work](https://bullquill.com/prompt-engineering-techniques-guide/), [OpenAI models and Codex come to Google Cloud](https://bullquill.com/openai-models-codex-come-to-google-cloud/), and [DeepSeek's $15 billion funding round](https://bullquill.com/deepseek-15-billion-funding-round/).

## Frequently asked questions

### What is Canva Craft?

Canva Craft is the first AI image generation model built and trained by Canva's own research team. It was released on 22 October 2026 and is designed for layouts with text, such as posters, social posts and menus.

### How is Canva Craft different from other AI image generators?

According to Canva, Craft first plans the layout, relative size and hierarchy of elements and the placement of text, then renders the image at up to 2K. It was trained on human-made design templates rather than photographs, and its output can be converted into editable layers with Canva Magic.

### Where can I use Canva Craft?

Craft is available in the Dream Lab tab of the Canva app and through the Canva API. Reports differ on whether a paid plan is required. Canva says access through its MCP server for AI agents is coming but has not given a date.

### Does Canva still use Leonardo.Ai?

Canva acquired Leonardo.Ai in 2024 and built its Dream Lab tool on Leonardo's Phoenix model. TechCrunch reported that Leonardo researchers contributed to Craft, which Canva describes as its own model built from scratch.

## Sources

- [Canva Newsroom: Canva Craft announcement](https://www.canva.com/newsroom/news/canva-craft/?ref=bullquill.com)
- [TechCrunch: Canva releases its first image generation model (22 October 2026)](https://techcrunch.com/2026/10/22/canva-releases-its-first-image-generation-model/?ref=bullquill.com)
- [Thurrott.com: Canva releases its first AI model for images (22 October 2026)](https://www.thurrott.com/a-i/343158/canva-releases-its-first-ai-model-for-images?ref=bullquill.com)
- [Capital Brief: Canva unveils Leonardo.Ai-powered Dream Lab](https://www.capitalbrief.com/briefing/canva-unveils-leonardoai-powered-dream-lab-a011b92e-3236-418b-b630-113c071439f5/?ref=bullquill.com)
- [Gigazine: Canva launches Dream Lab based on Leonardo.Ai (October 2024)](https://gigazine.net/gsc%5Fnews/en/20241023-canva-leonardo-ai-dream-lab?ref=bullquill.com)
- [OpenAI: Introducing image generation in the API (April 2025)](https://openai.com/index/image-generation-api/?ref=bullquill.com)
- [Adobe Help: Partner models in Adobe products](https://helpx.adobe.com/creative-cloud/apps/generative-ai/non-adobe-models-in-adobe-products.html?ref=bullquill.com)