Canva Craft: How to Generate AI Images With Readable Text in Canva (New 2026 Model)

Introduction

If you make blog covers, social posts, posters or YouTube thumbnails, you already know the most annoying problem with AI image tools: the picture looks great and the text on it is gibberish. Misspelled headlines, extra words that nobody asked for, letters melting into each other. On 22 October 2026, Canva launched Craft, its first image generation model built entirely in-house, and its whole pitch is fixing exactly that problem. The direct answer, before the detail: Canva Craft is now available to all Canva users inside Dream Lab, and to developers through Canva's public API. It works in two stages. First it plans the layout of your design, where each object goes, how big it is, and where the text sits. Then it renders the final image at up to 2K resolution in more than 30 aspect ratios. Because it was trained on millions of human made design templates instead of photographs, it understands posters, social posts, infographics and menus the way a designer does. This article explains what Craft is, how to use it step by step, how it compares with the tools you already know, and the prompt habits that get clean, readable text on the first try.

What Canva Craft Is and Why It Is Different

Most AI image generators go straight from your prompt to pixels in one pass. You describe a poster, the model guesses the whole thing at once, and if the headline comes out wrong, you roll the dice again. Craft splits the job in two. Stage one reads your prompt and builds a composition plan: which elements appear, where each one sits, how large it is relative to the others, what the visual hierarchy is, and exactly where any text goes. Stage two renders the final image from that plan. That difference matters most for design work. A photograph does not need a headline in a specific corner at a specific size. A poster does. An Instagram post does. A menu does. Canva trained Craft on its own library of human made templates, so the model has seen millions of real layouts where text, images and spacing already work together. Researchers from Leonardo.ai, the media generation company Canva acquired in 2024, contributed to building the model, and Canva Research, a lab of more than 160 researchers and engineers, built it from scratch rather than fine tuning somebody else's model. The numbers Canva published at launch are worth knowing, with one honest caveat: they are company reported results, not independent tests. On DesignArena, Craft debuted fourth overall, behind three OpenAI GPT Image models but ahead of Google's Nano Banana 2.1. In individual categories it ranked first for People and Portraits out of 87 models and first for Landscapes and Nature out of 85 models, second for Abstract and Patterns and fourth for Products. Canva also says its own designers judged more than 25,000 pairs of images during development, and that when automated scores disagreed with a designer, the designer won. That is a useful signal about what the model was optimized for: designs a human would actually ship, not just images that score well on a benchmark. There is also a business reason behind the launch that is worth understanding. Every image Canva users generated with a partner model cost Canva an API fee. Building its own model gives Canva control over quality, speed and cost, which is exactly why the faster Craftfast and Craft-ultrafast versions are coming next. For users, the practical upside is a model tuned for the exact job Canva is used for, instead of a general purpose image model squeezed into a design tool.

How to Use Canva Craft in Dream Lab, Step by Step

You need nothing extra to try Craft. No new app, no separate signup, no plugin. If you have a Canva account, free or paid, you already have access, subject to the AI usage limits on your plan. Here is the full workflow. Step 1: Sign in to Canva and open Dream Lab. Dream Lab is Canva's AI image generation feature. On most accounts it sits in the left sidebar; on some accounts it lives under the More menu. You can also go straight to the Dream Lab page from Canva's homepage navigation. Step 2: Check the model selector. Under the prompt box, the model picker should show Canva Craft, listed as Canva Craft-1 2K on some accounts. If it shows a different model, switch it to Craft before you generate. This is the step people skip, and then they wonder why the results do not match what they read about. Step 3: Write a complete prompt. Name every element you want in the image, including the exact text that must appear. Put that text in quotation marks and spell it exactly as it should read. Say what should be biggest or most prominent. For example: a clean poster for a bakery sale, headline "Fresh Bread Every Morning" as the largest text, a loaf of bread on a wooden table, warm morning light, minimal style. Vague prompts give vague posters. Step 4: Choose the aspect ratio and style. Craft supports more than 30 aspect ratios, so pick the shape that fits where the image will live: square for feed posts, vertical for stories and posters, wide 16:9 for blog banners and video thumbnails. Picking the ratio before generating saves you a crop later. Step 5: Generate and pick from up to four results. Canva returns several options for one prompt. Open each one at full size and check the text letter by letter before you use it. Zoom in on numbers, symbols and any non English text, because small text errors are easy to miss at thumbnail size. Step 6: Refine instead of restarting. If one object needs to move or one word is wrong, use the edit prompt to request a single change at a time, or use Magic Layers to split the image into editable elements. Opening the result in the Canva editor lets you change fonts, colors and text directly, which is faster than generating ten versions hoping for a perfect one. One practical warning from early testers: the edit box is an instruction field, not a chat. If you type a question there, that question can end up painted onto the image as text. And if you do not tell Craft to use only the text you specified, it sometimes adds extra labels, badges or taglines nobody asked for. Add a line like "use only the text I specified, no extra labels or taglines" to prompts where clean text matters.

What Craft Means for Bloggers, Small Businesses and Creators

This is where Craft stops being tech news and starts being useful. If you run a blog, a small shop, or a social media page, you probably make images every week: post covers, sale announcements, event posters, product graphics, infographics. Hiring a designer for each one is expensive. Wrestling with a general image generator for each one eats time, because every poster needs three or four retries to get the headline right. Craft is built for that exact weekly grind. A small business owner can generate a sale poster with a readable discount headline, then open it in the editor and change the price next week without starting over. A blogger can generate a cover image in the exact ratio a post needs, with the title text placed where it belongs, and then compress it for fast page loading. A social media creator can test four versions of a post graphic in minutes and keep the one that reads best on a phone screen. That last point deserves emphasis, because it connects directly to website performance. AI images at 2K resolution are large files, and large images slow pages down, which hurts both readers and search rankings. Generate at the quality you need, then compress before you publish. ZaxisMedia's free Image Compressor is one quick way to trim a Craft image for the web without visible quality loss, and the Background Remover is handy when a generated product shot needs a clean transparent background for a shop page or a thumbnail. Craft also changes the cost calculation for Canva itself, which is good news for users in the long run. Partner models charge per image, and those costs eventually land on users through credit limits. An in-house model lets Canva offer more generations for the same plan, and the upcoming Craftfast and Craft-ultrafast versions are explicitly aimed at lower cost and faster output. For now, the honest limits are these: free accounts face AI usage limits, generation through the API takes roughly a minute per image according to Canva's developer forum notes, and reference image support plus deeper editing tools are still on the roadmap, not in your hands yet.

Comparison Table

Conclusion

Canva Craft is not the most powerful image model in the world, and Canva does not claim it is. It is the first model built specifically for the job millions of Canva users actually do every week: make a design that says something, clearly, in the right shape, without spending an afternoon fighting garbled text. The two stage planning process, the template based training, and the direct path into Canva's editor make it the most practical AI image launch of October 2026 for bloggers, small businesses and creators.

If you already use Canva, the advice is simple. Open Dream Lab this week, confirm Craft is selected as the model, and try one real job with it: your next post cover, sale poster or thumbnail. Write the headline in quotes, state what is biggest, add the no extra text guardrail, generate in the final ratio, and check the text at full size before you publish. Compress the result for the web, and keep the layered version in Canva so next week's version takes minutes instead of hours. The model will get faster and cheaper with Craftfast and Craft-ultrafast, but the workflow habit you build now is what will keep paying off.