 |
| Illustration: a designer workspace with a blank canvas and image thumbnails. This is an AI-generated illustration, not a screenshot of Canva Craft. |
On 22 October 2026, Canva did something it had never done before:
it launched its own AI image generation model, called Canva Craft, built and trained entirely in-house by its own research team. For years, Canva's AI image features ran on models licensed from outside companies, most notably OpenAI and Google. Now the world's most popular design platform has stopped renting its AI and started building it. Craft is available to all Canva users worldwide through Dream Lab, and developers can also reach it through a public API.
This is a bigger story than one new feature. Canva Craft is the clearest sign yet that the design software industry is moving from licensing AI to owning it. Adobe already has Firefly. Figma, Notion and others have been reducing their dependence on outside AI labs. Canva is now the latest — and one of the largest — platforms to make that bet. And because Canva is the tool millions of small businesses, teachers, marketers and creators already open every day, a change to its AI engine affects far more people than a new model from a research lab ever could.
In this guide, we explain what Canva Craft is, how its unusual two-stage process works, what it can and cannot do, how to use it, what it costs, and how it compares with the other big AI generators that launched in October 2026 — Google's Nano Banana 2.1, Black Forest Labs' FLUX 3 Image, and Kling 4.0. Every fact here comes from the launch coverage and announcements, and where something is still unconfirmed, we say so plainly.
What Is Canva Craft? The Short Version
Canva Craft is Canva's first in-house AI image generation model, developed and trained from scratch by Canva Research, the company's AI lab. It was announced on Thursday, 22 October 2026, and reported widely the following day. According to the launch coverage, Canva built Craft to compete directly with image models from Google, OpenAI, ByteDance and Meta — the companies whose models it used to license.
That last point matters. Until now, Canva generated images with several outside models. The company bought the AI generation platform Leonardo.ai in 2024, and its researchers worked on Craft after that acquisition. Craft is the first time Canva has shipped a model it built and trained itself, rather than plugging in someone else's technology.
Here are the key facts about the launch, all confirmed in the launch reporting:
- Launch date: announced 22 October 2026, with coverage on 23 October 2026.
- Who built it: Canva Research, Canva's own AI lab, from scratch.
- Where to find it: inside Canva, in Dream Lab in the bottom-left panel, for all users worldwide. Developers can also use it through a public API.
- What it replaces: nothing yet, officially. Canva has not said whether Craft will fully replace the third-party models inside the platform or simply sit alongside them as another option.
- What it costs: Canva has not disclosed pricing details, and has not said whether Craft access will require a higher subscription tier.
- What it competes with: Google's Nano Banana image models, OpenAI's GPT Image models, ByteDance's image tools and Meta's image models.
One important detail for beginners: Craft is an image model, not a separate app. You do not download anything new or create a new account. If you already use Canva, Craft appears inside the tool you already know, in Dream Lab. That is exactly why this launch matters to so many people — the new AI arrives where the work already happens, instead of asking users to learn a new platform.
How the Two-Stage Process Actually Works
Most AI image generators work in a single pass: you type a prompt, and one model turns it directly into a finished picture. Canva Craft works differently — it uses a two-stage process, and that difference is the whole point of the model. Canva describes it like this: first, one model writes a plan of the composition. Then a second model turns that plan into the final image.
Think of it the way a real design studio works. Before anyone draws anything, someone sketches the layout: which elements appear, where they go, how big each one is, and what text they carry. Craft does the same thing, automatically. The planning model decides the composition — the placement of objects, the size of each element, the position of the text — and writes it down as a plan. Only then does the second model generate the actual image from that plan.
Why does this matter? Because the single-pass approach has a well-known weakness: it is often inconsistent on text placement, logo accuracy and layout. A single-pass model can produce a beautiful picture with a misspelled headline, a logo in the wrong place, or a layout that looks accidental. That is acceptable for a fantasy landscape, but it is a real problem for business design — a social post, an advertisement, a flyer — where text and layout need to look intentional.
Canva says the planning step gives it more control over precision and consistency. In practice, that means three concrete advantages:
- More accurate object placement: because the layout is planned before the image is drawn, elements land where they were meant to land, instead of drifting around the frame.
- More legible text: the planning model decides what the text says and where it goes before the image model renders it, which is why Canva claims Craft handles text better than single-pass models. Craft writes text in 55 languages.
- Single-element editing: because the image is built from a plan, users can change one element without touching the rest of the image. If the headline is wrong, you can fix the headline without regenerating the whole picture.
That last point is easy to underestimate. In most image generators, changing one small thing means rolling the dice again on the entire image. You ask for a different headline and the whole picture changes — the colours shift, the composition moves, the image you liked is gone. A plan-based model can, in principle, keep everything else fixed and change only the part you asked about. If Craft delivers on that promise in everyday use, it will save designers a large amount of time and frustration.
It is worth being honest about one thing, though: the two-stage approach is a design choice, not a guarantee of better pictures. A planning step helps with layout, text and consistency — the things business design cares about most. It does not automatically make Craft better at photorealism, artistic style or fine detail than the frontier models from Google and OpenAI. The real test will be independent comparisons of Craft's output against those models, which had not yet appeared at the time of writing. We return to that question later in this guide.
What Canva Craft Can Do: Resolution, Ratios, Languages and Magic Layers
So what can you actually make with Craft? The model outputs images up to 2K resolution, in more than 30 aspect ratios — or a custom ratio you set yourself. That range covers the formats small businesses actually use every day: Instagram posts, YouTube thumbnails, print flyers and presentation slides. The 2K ceiling is not the highest resolution in the market — Google's Nano Banana 2.1, for example, renders up to 4K — but 2K is more than enough for social media, web use and most print work at small sizes.
Here is what the launch reporting confirms Craft can do:
- Generate images up to 2K resolution across more than 30 aspect ratios, plus custom ratios.
- Prompt in any language — you can describe what you want in your own language, not only in English.
- Write text in 55 languages inside the generated image, which matters for businesses working in more than one language.
- Turn generated images into editable layers with Magic Layers, Canva's tool that separates the elements of an image so you can move, resize or replace them individually.
- Change a single element without regenerating the whole image, thanks to the plan-based approach described above.
Magic Layers deserves a closer look, because it is what turns a generated picture into a working design file. In most AI image tools, the output is a flat picture: one layer, take it or leave it. If you want to move the headline or change a colour, you have to open a separate editor — or generate the whole image again and hope for the best. Magic Layers separates the elements of a Craft image into layers you can edit inside Canva, just like the layers in a normal Canva design. That turns the AI output into something you can keep working on, rather than a finished picture you cannot touch.
The training data is the other distinctive part of Craft. According to the launch reporting, its training gives more weight to design than to photographs. It includes millions of human-made templates from Canva's own library — posters, social media posts, presentations, infographics and menus — and the company then fine-tuned it using ratings from its professional designers. That is a deliberate choice: Craft is trained to be good at the kind of images Canva users actually make, rather than trying to be equally good at everything.
This explains the trade-off Canva has made. A model trained on millions of real design templates should be strong on layout, typography, posters, ads and social graphics — the bread and butter of business design. It may be less strong on photorealistic portraits or fine art, where models trained mainly on photographs have the advantage. That is not a flaw; it is a choice about who the model is for.
There is one more number worth knowing. According to Canva, Craft debuted as the fourth-best image model on DesignArena, the largest AI design ranking based on user votes — and first for portraits of people (out of 87 models) and for landscapes and nature (out of 85). In the same ranking it places second for abstracts and patterns, and fourth for products. Those are striking debut numbers, especially the first-place finishes. But read them carefully: DesignArena is a ranking based on user votes, not a laboratory benchmark. It tells you what voters preferred in blind comparisons, which is useful, but it is not the same as a controlled test of image quality, and rankings based on votes can shift quickly as more people try a new model. Treat the DesignArena result as an encouraging first sign, not as proof that Craft is the fourth-best image model in the world.
How to Use Canva Craft: A Step-by-Step Guide
You do not need a new account or a new app to try Canva Craft — it lives inside Canva, in Dream Lab. According to the launch reporting, all Canva users worldwide can already use it from Dream Lab in the bottom-left panel of the Canva editor, and developers can reach it through its public API. If you have used Canva's AI image features before, the workflow will feel familiar. Here is how to use it, step by step.
Step 1: Open Canva and find Dream Lab. Open Canva in your browser or the Canva app, and open any design — or start a new one. Look for Dream Lab in the bottom-left panel. That is where Craft lives. If you cannot see it yet, remember that new features roll out gradually; Canva says Craft is available to all users worldwide, but if it has not appeared for you, check again after a day or two and make sure your app is up to date.
Step 2: Describe the image you want — in your own language. Type a description of the image you need, in any language — Craft accepts prompts in any language. Be specific about three things: the subject (what is in the picture), the style (a poster, a social post, a product photo, an illustration), and the text (exactly what words should appear, and where). The more precise your description, the better the planning stage can do its job, because the plan is only as good as the brief you give it. A vague prompt like "a nice poster" gives the planner almost nothing. A clear prompt like "a square Instagram post for a bakery, warm colours, a photo-style image of fresh bread on a wooden table, with the words 'Fresh Bread Daily' at the top in large letters" gives it a real brief.
Step 3: Choose your format before you generate. Craft supports more than 30 aspect ratios, or a custom ratio you set yourself. Pick the format that matches where the image will be used — square for an Instagram post, wide for a YouTube thumbnail, tall for a story or a flyer — before you generate, not after. Generating in the right ratio from the start is always better than cropping afterwards, because cropping throws away part of the image and can cut through text or important elements.
Step 4: Generate and review the plan in the result. Craft will produce the image from its two-stage process. When you look at the result, check the things the planning stage is supposed to get right: is the text spelled correctly, is it in the right place, are the objects where you expected them, does the layout look intentional? Judge Craft on layout and text first, and on artistic beauty second. If the text is wrong, try rephrasing it more simply: short words, one line, and no unusual punctuation tend to render more reliably.
Step 5: Fix one element instead of starting over. This is where Craft is supposed to be different from other generators. If one element is wrong — the headline, a colour, one object — change that element on its own rather than regenerating the whole image. Ask for the single change, keep everything else fixed, and check that the rest of the image has not moved. This is the workflow the two-stage design exists to support, and it is the habit that will save you the most time if it works as advertised.
Step 6: Turn the image into editable layers with Magic Layers. Once you have an image you like, use Magic Layers to separate its elements into layers you can edit inside Canva. Now the headline is real text you can retype, the shapes are shapes you can move, and the background is a background you can replace. This is the step that turns an AI picture into a real design file you can keep working on. If you skip it, you are left with a flat image — useful, but far less flexible.
Step 7: Drop it into your design and finish it in Canva. Place the image into your poster, post, presentation or flyer, and finish the job with Canva's normal tools: your brand fonts, colours, logo and template. The best results come from treating Craft as the starting point, not the finished product. Generate the image, then design around it. That combination — AI image plus human design sense — is where Craft is most useful.
A practical tip for first-time users: run the same prompt two or three times and compare the results before you settle on one. Every image model varies from run to run, and Craft is no exception. The first result is rarely the best one. Generate a few versions, pick the strongest layout, and then refine that one. And keep your prompts: when you find a prompt that works well for your kind of design, save it and reuse it as a template for the next one.
DesignArena Rankings: What They Really Mean
Canva's headline claim for Craft is its debut on DesignArena: fourth-best image model overall, first for portraits of people out of 87 models, first for landscapes and nature out of 85, second for abstracts and patterns, and fourth for products. Those numbers look impressive, and they are worth understanding properly — because what a ranking measures decides what it proves.
DesignArena is the largest AI design ranking based on user votes. In simple terms, people are shown images from different models, they vote for the ones they prefer, and the ranking reflects those votes. That makes it a real signal: it measures what ordinary users actually like when they look at pictures, not what a company's marketing department claims. A new model debuting at fourth overall, and first in two categories, is a genuinely strong start.
But a user-vote ranking is not a laboratory benchmark, and the difference matters. Here is what DesignArena does and does not tell you:
- It tells you what voters preferred in blind comparisons of finished images. That is useful, but it says nothing about how easy the model is to use, how well it follows complicated prompts, or how it handles edge cases.
- It does not test text accuracy systematically. A model can win votes with beautiful images and still misspell headlines half the time. Craft's main selling point is text and layout, so vote-based rankings are not the best way to judge its core strength.
- It does not measure consistency. A model that produces one stunning image and nine mediocre ones can still rank well if voters only see the stunning one. For business work, you need a model that is good nearly every time, not occasionally brilliant.
- Rankings move quickly. A debut ranking reflects early voters and early impressions. As more people try Craft — including people who push it harder and find its limits — the ranking can shift in either direction.
The honest way to read the DesignArena result is this: Craft made a strong first impression on voters, especially for portraits and landscapes. That is encouraging, and more than most new models manage. But it is not proof that Craft beats Google's or OpenAI's models at what businesses need — reliable text, consistent layout, repeatable quality. Those questions will only be answered by independent head-to-head comparisons and weeks of everyday use. Until then, treat the ranking as a promising sign, not a final verdict.
Why Canva Built Its Own Model (and Stopped Renting AI)
To understand why Craft matters, you have to understand the business behind it. For years, Canva's AI image features ran on models licensed from outside partners, most notably OpenAI and Google. Every time a user generated an image, Canva was, in effect, renting the AI from another company. That arrangement is common across the design and marketing software world — few companies build foundation models themselves, because doing so is expensive and slow. But it has a cost, and Canva has now decided to stop paying it.
The economics are straightforward: licensing fees for API calls add up fast at scale. Canva has hundreds of millions of users, and every image generation is a paid API call to someone else's model. Owning the model means Canva controls its own costs, can tune the model specifically for design use cases, and is not exposed to price changes or shutdowns from a third-party AI provider. Canva itself says that, beyond cost, having its own model gives it better control over quality and speed.
The strategy is also about independence. When your core feature depends on a competitor's technology, your roadmap is partly in their hands. If OpenAI or Google changes its pricing, limits access, or launches a competing design tool, Canva's product suffers. Building Craft removes that dependency. It also means Canva is no longer just a wrapper around someone else's technology — it is now a direct competitor to the AI labs it used to license from, at least in the narrow lane of design-ready image generation.
This fits a pattern playing out across the whole software industry. Adobe has its Firefly models. Figma, Notion and others have all moved to reduce their dependence on OpenAI, Google and Anthropic over the past two years. The companies that built their businesses on top of someone else's AI are now building their own. Canva is the latest and one of the largest to make that move, and its scale makes the shift especially significant.
There is also a product reason, not just a business one. A model Canva owns can be tuned for exactly what Canva users do. A general-purpose model from a lab has to be good at everything — photographs, art, diagrams, faces, landscapes. A Canva-owned model can be tuned for posters, social posts, presentations, infographics and menus, because that is what Canva's users actually make. The training data reflects that choice: millions of human-made templates from Canva's library, fine-tuned with ratings from professional designers. That specialisation is Craft's real advantage — and also its real limit, as we discuss later.
None of this means the licensed models disappear overnight. Canva has not said whether Craft will fully replace the third-party models or sit alongside them. For now, Craft is the start of a transition, not the end of one. Canva will watch how users respond, keep improving the model, and decide over time how much of its AI runs on its own technology. The launch is the beginning of that story, not the conclusion.
What Does Canva Craft Cost? What We Know — and What Canva Has Not Said
Here is the honest answer: Canva has not disclosed pricing details for Craft, and has not said whether it will require a higher subscription tier. That is not a small gap. Pricing decides whether Craft is a free upgrade for everyone who already uses Canva, or a premium feature only some users will ever touch. Until Canva announces it, nobody outside the company knows — and anyone who tells you the price is guessing.
What we do know is the context. Craft is available to all Canva users worldwide through Dream Lab, according to the launch reporting. That suggests it is not locked behind a closed beta or an invitation list. But "available to all users" and "free for all users" are two different things. Canva already has free and Pro tiers, and new AI features have historically landed in different places. The launch coverage is clear that the pricing question is unanswered, and that is the fact that matters most for anyone planning to build a workflow around Craft.
Here is what to watch for in the coming weeks:
- Whether Craft becomes a Pro-tier exclusive. If it does, that signals a revenue play: Canva using its own model to sell more subscriptions. If it stays available to free users, that signals a value-add play: Canva using its own model to keep users inside its ecosystem.
- Whether there are usage limits. Many AI image tools limit how many images you can generate per day or per month on lower tiers. Watch for a credit or generation limit, and what happens when you hit it.
- Whether the licensed models stay available. If Canva keeps OpenAI and Google models alongside Craft, users can choose the best tool for each job. If Craft replaces them, users lose that choice — and the ability to compare quality and pricing against competitors.
- How the API is priced. Developers can reach Craft through a public API. API pricing will decide whether small developers can afford to build on Craft, or whether it is only practical for larger companies.
The trade-off behind the pricing question is lock-in, and it is worth understanding before you commit to Craft. The more Canva's AI capabilities run on its own proprietary model, the harder it becomes to compare quality or pricing against competitors, and the more a business's workflow depends on staying inside Canva's ecosystem. If pricing shifts later, or a feature gets walled off behind a premium tier, switching costs are higher once a team has built habits and brand templates around the tool. That does not mean you should avoid Craft — it means you should keep your original design files portable, keep your brand assets outside any single tool, and test Craft against your current AI image workflow before assuming it is an automatic upgrade.
For now, the practical advice is simple: try Craft while it is available, learn what it does well, but do not rebuild your entire workflow around it until the pricing is public. A tool you cannot afford next month is not a tool you should depend on today.
What Is Coming Next: Craft-Fast, Craft-Ultrafast and the MCP Server
Canva has already said what comes next for Craft, and the roadmap is revealing. The company is preparing two faster variants, Craft-fast and Craft-ultrafast, designed to cost a fraction of what the most advanced models do. That tells you two things: first, that Canva is thinking about speed and cost as features in their own right, not just image quality; and second, that the current Craft model is the quality tier, with cheaper and faster tiers to follow.
Why would anyone want a faster, cheaper model? Because not every image needs maximum quality. When you are brainstorming ideas, testing ten different layouts, or generating rough drafts for a client to react to, speed matters more than perfection. A fast model that produces a good-enough image in seconds is more useful for iteration than a slow model that produces a perfect image in a minute. The fast tiers are for the draft stage; the full model is for the final image. That is a sensible product strategy, and it mirrors what other AI companies do — Kling, for example, launched a Flash tier of Kling 4.0 for fast iteration alongside its full model.
The second item on the roadmap is the MCP server. Craft will soon come to Canva's MCP server — the system that connects Canva to AI assistants and agents — so those assistants can create images ready to use in a design. According to the launch reporting, that opens the door to generating Canva images from other companies' assistants, such as ChatGPT or Gemini. In plain terms: instead of opening Canva, going to Dream Lab and typing a prompt, you could ask your AI assistant to make the image, and it would arrive in Canva ready to use.
That is a bigger change than it sounds. It means Craft stops being a feature inside Canva and starts being a capability other AI tools can call on. A marketing team could ask its AI assistant to draft a whole campaign — copy, images and layout — and the images would be generated by Craft and land in Canva automatically. For businesses that already work through AI assistants, that removes a step entirely.
Canva also says it will keep improving the model itself — for example with fine print and technical drawings, two areas where image models traditionally struggle. Small text and precise technical diagrams are hard for image generators because they require exactness, not just plausible-looking shapes. That Canva names these two areas specifically suggests it knows where Craft is weakest today, and is working on them first. Watch for updates on those two points: if Craft gets noticeably better at fine print, that will be the clearest sign the two-stage approach is paying off.
October 2026: The Busiest Month Ever for AI Generators
Canva Craft did not launch in a quiet month. October 2026 has been one of the busiest months in the history of AI generation, with major new image and video models arriving almost every week. Understanding that landscape matters, because Craft's real competition is the specific models that launched alongside it. Here is the month so far, in order:
| Date | Model | Company | What it is |
|---|
| 1 October 2026 | FLUX 3 Image | Black Forest Labs | Image generation and editing, up to 4K, ten references |
| 6 October 2026 | Nano Banana 2.1 | Google | Image generation and editing, up to 4K, 14 references |
| 15 October 2026 | FLUX 3 Image Klein | Black Forest Labs | Open-weights image model, native 4K, runs on 8GB VRAM |
| 16 October 2026 | Kling 4.0 | Kling AI (Kuaishou) | Video generation, 30 seconds in one pass, 10 keyframes |
| 22 October 2026 | ai& video | ai& (Japan) | Video generation from text, photos and clips |
| 22 October 2026 | Canva Craft | Canva | In-house image model, two-stage, up to 2K |
Five major launches in twenty-two days — and that is only counting the ones directly relevant to image and video generation. The same month also brought OpenAI's GPT Image 2.5, which shipped on 8 September 2026 and was still fresh in October, and Meta's Muse Spark Open, an 816-billion-parameter open-weight multimodal model announced on 15 October. The pace is the story: AI generation is no longer a race with one or two leaders and a long gap behind them. It is a crowded field where a new model can arrive every few days, and where yesterday's best model can be second-best by next week.
For readers, the practical consequence is simple: do not buy into any single tool on one launch announcement. Learn each major model's strengths, use the one that fits the job in front of you, and keep your files and prompts portable so you can switch when something better appears.
The New Models in Detail: Nano Banana 2.1, FLUX 3 Image and Kling 4.0
Google's Nano Banana 2.1 went generally available on 6 October 2026, as the successor to the Gemini 3.1 Flash Image model. It renders images up to 4K, supports up to 14 reference images for multi-image fusion, and maintains character consistency for up to 4 characters. Its context window is 131,072 tokens with a maximum of 32,768 output tokens. Pricing is usage-based and scales with resolution: a 1K image uses 1,120 tokens, a 4K image 3,780 tokens, at a listed rate of $0.0336 per 1K image. For brand work, the 14-reference limit is the headline feature — pass in several brand-reference images and generate variants that match your identity from the start, instead of generating hundreds of shots and filtering the ones that match.
Black Forest Labs' FLUX 3 Image arrived on 1 October 2026, with image generation and editing up to 4K and ten references, priced from $0.041 to $0.607 per image depending on the tier. Two weeks later, the company released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model capable of native 4K generation, pixel-precise editing and multi-reference synthesis — running on a consumer GPU with just 8GB of VRAM. That last detail is the important one: an open model that runs on an ordinary graphics card puts professional-grade image generation within reach of anyone with a decent gaming PC, with no per-image fee and no cloud account required.
Kling 4.0 is the video story of the month. Kling AI announced its global launch on 16 October 2026, and it is the first widely available video model that generates up to 30 seconds of video in a single unbroken pass. It supports up to 15 reference assets and 10 keyframes through its Omni Reference system, so creators can control characters, scene transitions and story progression frame by frame. It can even copy the camera work from a reference video while replacing the scene and characters. Ten-bit HDR output at 4K and 1080p, plus a 21:9 ultrawide format, have been announced as coming soon — they do not exist in anything you can generate today. A faster Flash tier opened in early access on 28 September 2026 for Ultra Yearly subscribers, generating up to 20 seconds per pass at up to 720p. Kling AI says it serves more than 100 million users and over 50,000 enterprise customers.
The Japanese service ai& video launched the same day as Craft, 22 October 2026. It turns text, photos and clips into new footage — creators can describe a scene, animate a photo, or continue a clip — generating video in as little as 12 seconds. On supported models it also generates audio. Pricing is pay-as-you-go with no subscription. It is a smaller launch, but it points to the same trend: video generation is becoming fast, cheap and local.
Put together, October's launches tell a single story: image models are getting better at references and consistency, video models are getting longer and more controllable, and open models are getting good enough to run at home. Canva Craft fits into that story as the design-focused entry — not the most powerful model, but the one built most specifically for the work its users actually do.
 |
| Illustration: generic AI image and video generator windows. This is an AI-generated illustration, not a screenshot of any real product. |
Canva Craft vs the Rivals: Which Tool for Which Job
So how does Craft actually compare with the other models that launched in October? No single model wins every job, and the honest answer is a table, not a slogan. Here is how the main options stack up on the facts we can verify:
| Model | Best for | Max resolution | Standout feature | Pricing known? |
|---|
| Canva Craft | Business design, social posts, ads | 2K | Two-stage planning; Magic Layers editing | No — not announced |
| Nano Banana 2.1 (Google) | Brand work, character consistency | 4K | 14 reference images; search grounding | Yes — $0.0336 per 1K image |
| FLUX 3 Image | High-end image generation and editing | 4K | Ten references; open Klein version | Yes — $0.041 to $0.607 per image |
| GPT Image 2.5 (OpenAI) | Precise edits to a locked image | Not stated here | Precision editing | Not stated here |
| Kling 4.0 | AI video, ads, short films | 1080p (4K HDR coming soon) | 30-second single takes; 10 keyframes | Partly — Flash tier pricing reported |
The practical rule is simple: choose the model by the job, not by the brand. Here is how that works in practice:
- Making a social media post, flyer or ad inside Canva? Use Craft. It is already where your design lives, it is tuned for exactly that work, and Magic Layers lets you keep editing afterwards. No other model is integrated into a design tool this deeply.
- Need a character or product to look identical across hundreds of images? Use Nano Banana 2.1. Its 14-reference conditioning is built for brand consistency at scale, and its 4K output gives you more detail to work with.
- Want the highest image quality and full control, or want to run a model on your own computer? Look at FLUX 3 Image — or the open-weights FLUX 3 Image Klein, which runs on a consumer GPU with 8GB of VRAM and charges no per-image fee.
- Need to edit one thing in an existing picture without changing anything else? GPT Image 2.5 is the tool the comparison guides recommend for precision edits to a locked still.
- Making a video, not a still image? That is Kling 4.0's job. Thirty seconds in one pass, with keyframes for planning, is a different capability from anything in the image models — and a different tool entirely. For a broader look at free video options, see our guide to the top free AI video generators with a credit system in 2026.
- Learning AI image generation from scratch? Start with our step-by-step guide to how to create AI images for free in 2026, then try Craft inside Canva once you know the basics.
One warning about comparisons: do not trust any ranking or review that tests only one prompt. Generative models vary from run to run, so a single image can flatter or insult a model. A fair comparison repeats the same prompt several times on each model, tests different kinds of jobs, and reports the range of results, not just the best one. Until independent comparisons like that appear for Craft, the table above is a guide to verified features, not a final score.
Who Should Use Canva Craft — and Who Should Skip It
Canva Craft is a strong fit for you if you already live inside Canva. Small business owners making social posts and ads, teachers making classroom materials, marketers making campaign graphics, freelancers making client work — if Canva is already where your designs happen, Craft removes the need to leave the app, learn a new tool, and import the result back. The two-stage planning, the 55-language text support and the Magic Layers editing are all aimed squarely at that user.
It is also a good fit if your work is text-heavy. Posters, flyers, menus, infographics, social graphics with headlines — anything where the words have to be right and the layout has to look deliberate. That is the job Craft was trained for.
Think twice before relying on Craft if you need top-tier photorealism. Businesses that need the most realistic product photography, cinematic portraits or fine-art images may still reach for a dedicated tool — Nano Banana 2.1, FLUX 3 Image or GPT Image 2.5 — while Craft proves itself on raw image fidelity. In-house models from design platforms have historically trailed the frontier labs on photorealism, even when they win on workflow. Craft's DesignArena debut suggests it may be closing that gap, but one ranking is not enough to settle the question.
Also think twice if you need full control over cost. Until Canva announces pricing, you cannot budget for Craft — and right now, Nano Banana 2.1 and FLUX 3 Image publish their prices while Craft does not. If you want to run a model on your own hardware with no per-image fee, FLUX 3 Image Klein is the only option in this group that does that.
Practical Tips: Getting Good Results from Canva Craft
Write prompts like a design brief, not like a search query. The planning model works from your description, so give it the same information you would give a human designer: the purpose of the image, the audience, the subject, the style, the colours, and the exact text. "A poster for a coffee shop" is a weak brief. "A warm, inviting poster for a neighbourhood coffee shop, photo-style image of a latte on a wooden table in soft morning light, with the words 'Morning Coffee, Fresh Daily' in large friendly letters at the top" is a strong one. Specific prompts produce specific plans, and specific plans produce better images.
Put the text in the prompt exactly as it should appear. Write the headline in full, with the exact spelling and capitalisation you want. Do not write "add some text about coffee" and hope the model invents good copy — it will invent something, and it may not be what you wanted. Short headlines work better than long sentences, and one line of text renders more reliably than a paragraph.
Use your brand assets as a starting point. If you have brand colours, a logo and a font, build the Craft image to match them — then finish the design in Canva with your real brand kit applied. The AI image is the raw material; your brand kit is what makes it yours. Never let an AI-generated image carry your brand on its own without checking it against your real brand assets.
Generate several versions, then pick the best layout. Do not settle for the first result. Generate three or four versions of the same prompt, compare them side by side, and choose the one with the strongest layout and the most accurate text. Then refine that one — fix the single element that needs fixing — instead of generating more versions and hoping for a better roll of the dice.
Keep a prompt library. When a prompt works well for your kind of design, save it — the exact wording, the ratio you used, and what the result looked like. Over a few weeks, you will build a small library of prompts that reliably produce good results for your business, and that library is worth more than any single image.
Check every image before you publish it. Read the text word by word — names, prices, dates, phone numbers. AI models can misspell words and garble small text, and a mistake in a published advertisement costs far more than the minute it takes to proofread. Treat every AI image as a draft until a human has checked it.
Common Mistakes and Limitations to Know About
Mistake 1: Treating the launch announcement as a review. An announcement tells you what a company says its model can do. It does not tell you what the model actually does in everyday use. The gap between those two things is where disappointment lives. Wait for independent testing before you judge Craft against its rivals.
Mistake 2: Assuming 2K is always enough. For social media and web use, 2K is plenty. For large print — a poster, a banner, a trade-show display — you may need higher resolution, and Nano Banana 2.1 or FLUX 3 Image at 4K will serve you better. Match the resolution to the output, not to the hype.
Mistake 3: Building a whole workflow around Craft before pricing is public. This is the biggest practical risk. If Craft becomes a Pro-only feature, or carries a per-image charge that does not fit your budget, a workflow built entirely on it will need rebuilding. Keep your files portable and your options open until the pricing is announced.
Mistake 4: Expecting the planning stage to fix bad prompts. The two-stage process can organise a good brief into a good layout. It cannot turn a vague brief into a good image. If your prompt does not say what the image is for, who it is for, or what it should say, no amount of planning will save it. The quality of the output is capped by the quality of the input.
Limitation 1: Fine print and technical drawings. Canva itself says it will keep improving the model in these areas — which is another way of saying they are weak today. If your work needs small, exact text or precise technical diagrams, test Craft carefully on those jobs before trusting it with them.
Limitation 2: Lock-in. The more your workflow depends on a proprietary model inside one company's tool, the harder it is to leave. Keep your brand assets, your templates and your prompt library somewhere you control, so that switching tools — if you ever need to — is a decision, not a crisis.
Limitation 3: The evidence is still thin. At the time of writing, Craft has been public for days, not months. There are no long-term reliability studies, no large independent benchmarks, and no track record of how it behaves under heavy, everyday use. That will change — but for now, every claim about Craft should carry a mental note: early days, verify for yourself.
Frequently Asked Questions About Canva Craft
What is Canva Craft? Canva Craft is Canva's first in-house AI image model, built and trained from scratch by Canva Research. Announced on 22 October 2026, it is available to all Canva users worldwide through Dream Lab, with a public API for developers.
How is Craft different from other AI image generators? Craft uses a two-stage process: one model plans the composition — elements, positions, sizes and text — and a second model turns that plan into the image. Most other generators do it in a single pass. Canva says the planning step gives better control over precision, consistency, text and layout.
Where do I find Craft in Canva? In Dream Lab, in the bottom-left panel of the Canva editor. You do not need a new account or a new app.
What resolution and formats does Craft support? Images up to 2K resolution, in more than 30 aspect ratios, or a custom ratio you set yourself. It can write text in 55 languages, and you can prompt it in any language.
How much does Canva Craft cost? Canva has not disclosed pricing, and has not said whether Craft will require a higher subscription tier. Watch for a pricing announcement before building a workflow around it.
Can I edit a Craft image after it is generated? Yes — Craft images can be turned into editable layers with Magic Layers, Canva's tool that separates the elements of an image so you can move, resize or replace them individually. You can also change a single element without regenerating the whole image.
Is Craft better than Nano Banana 2.1 or FLUX 3? There is no independent head-to-head comparison yet. Craft's DesignArena debut — fourth overall, first for portraits and landscapes — is encouraging, but it is a user-vote ranking, not a controlled benchmark. Choose by the job: Craft for design work inside Canva, Nano Banana 2.1 for brand consistency, FLUX 3 for quality or self-hosting.
Can Craft make videos? No — Craft is an image model. For video, the notable October 2026 launch is Kling 4.0, which generates up to 30 seconds of video in a single pass. See our guide to the top free AI video generators for the free options.
Conclusion: The Model Matters Less Than the Shift It Represents
Canva Craft is a good model with an interesting idea, launched into the busiest month AI generation has ever seen. The two-stage planning process is a genuine attempt to solve the problems business designers actually face — wrong text, accidental-looking layouts, and regenerating a whole image to fix one headline. The 2K output, 30-plus aspect ratios, 55-language text support and Magic Layers editing all point the same way: this is a model built for real design work, not for demos.
But the model itself is not the biggest story. The biggest story is the shift it represents. Canva — the tool hundreds of millions of people use to make things — has stopped renting its AI and started building it. Adobe did it with Firefly. Figma, Notion and others are doing it too. The design software industry is moving from licensing AI to owning it, and Craft is the largest example of that move so far. That shift will shape what these tools cost, what they can do, and how much control users have over their own work for years to come.
For readers, the practical takeaway is straightforward. Try Craft inside Canva, judge it on layout and text, and compare it honestly with the alternatives before you commit. Generate a few versions of a real job — a social post, a flyer, an ad — and see whether the planning stage saves you time. Keep your brand assets portable and your prompt library somewhere you control, and do not rebuild your workflow around Craft until the pricing is public. October 2026 gave you more good options than any month before it: Nano Banana 2.1 for brand consistency, FLUX 3 for quality and self-hosting, Kling 4.0 for video, and Craft for design work inside Canva. The best tool is the one that fits the job in front of you.
We will keep following Craft as independent comparisons appear, as Canva announces pricing, and as Craft-fast, Craft-ultrafast and the MCP server arrive. Check back for updates — and if you have tried Craft yourself, the most useful thing you can do is test it on your own real work and judge it by the results.
Sources
Key facts were verified against the following sources. Prices and specifications are as reported at the time of writing and may change; check the official pages before making a purchase decision.
Images in this article are AI-generated illustrations, not screenshots of Canva Craft or any other product.
Comments
Post a Comment