Canva has spent years adding artificial intelligence to generate images, edit photographs, and turn instructions into designs. Canva Craft, however, represents a different change: it is the first foundation image generation model developed and trained internally by Canva Research.
The distinction matters because Canva no longer has to depend exclusively on external models for a central function of its platform. With Craft, it can exercise greater control over composition, speed, quality, cost, and the connection between a generated image and Canva’s editing tools.
The announcement also raises an important question for communicators, designers, and marketing teams: what changes when artificial intelligence does not merely create an attractive image but attempts to understand its structure as a design?
What is Canva Craft?
Canva Craft is a visual generation model built from scratch by Canva Research, the company’s internal artificial intelligence lab. Its purpose is not merely to produce synthetic photographs. It aims to generate compositions that follow design principles such as hierarchy, alignment, scale, element distribution, and typography.
According to Canva, the model can create images at up to 2K and supports more than 30 aspect ratios within the platform. It is also available to developers through Canva’s public image generation API.
The most important clarification is that Craft is not Canva’s first visual generation feature. The platform already offered Dream Lab, Magic Media, and access to technology from Leonardo.Ai, OpenAI, and Google. The new development is that Canva now owns a model built specifically to perform that task.
Canva is not adding image generation for the first time. It is beginning to directly control the model responsible for producing some of those images.
Why Canva built its own model
Integrating external models allows a company to add capabilities quickly, but it also creates dependencies. The provider controls pricing, response times, updates, restrictions, and part of the technical experience.
An in-house model gives Canva greater freedom to adapt generation to its own product. The company can optimize it for the formats its community uses, connect it with other tools, and develop faster or more economical versions.
TechCrunch notes that owning the model may also reduce the cost of relying on third-party APIs. Canva has announced two future variants, Craft-fast and Craft-ultrafast, designed to reduce generation cost and latency.
Canva’s acquisition of Leonardo.Ai in 2024 also contributed to this process. Researchers from that company participated in Craft’s development, according to Canva.
An architecture that plans before generating
Craft’s most distinctive technical feature is its two-stage architecture. Instead of translating a prompt directly into pixels, the system first develops a plan for the image.
Stage one: visual planning
The model interprets the request and organizes elements such as:
- The objects that should appear.
- The spatial relationship between elements.
- Their relative scale.
- The visual hierarchy.
- The placement and content of text.
- The emphasis each component should receive.
Stage two: rendering
A second model translates that plan into the final image. Separating planning from rendering is intended to reduce common failures such as misplaced objects, inconsistent scale, illegible words, or compositions that do not follow the original intention.
This approach is closer to assembling a visual piece than producing a flat image from a broad description. It does not guarantee that every result will be correct, but it allows the system to consider the composition before it renders.
Why training on designs matters
Canva says Craft’s training placed substantial emphasis on millions of human-made templates from its content library. The collection includes posters, social posts, presentations, infographics, menus, and typography-rich layouts.
A photograph teaches a model what the world looks like. A template can also teach it how visual communication is organized: where to place a headline, how content blocks relate to one another, which element deserves priority, and how much space each component needs.
This orientation helps explain why Canva describes Craft as a model built for design. Its purpose is not only to compete in photorealism but to create images that can enter a communication workflow with less friction.
Legible text and compositional control
Generating text has been a recurring weakness of image models. Distorted letters, incomplete words, and invented phrases restrict their usefulness in advertising, posters, or informational content.
Canva says the planning stage helps Craft define text before rendering it. This may improve legibility and placement, especially in compositions where typography is part of the generated result.
An improvement is not a guarantee. Brand names, prices, dates, figures, calls to action, and legal text require human review. For professional publishing, editing the final wording as live text inside Canva remains safer than relying entirely on lettering embedded in a flat image.
From a flat image to an editable design
Craft connects with Magic Layers, a feature that transforms images into editable components. Structural information created during planning can help identify objects, backgrounds, and blocks so they can be separated into layers.
This capability may matter more than the initial generation. Many models produce an attractive image, but any correction requires regenerating the entire piece and may change elements that were already approved. We also explore this need for creative control in our article on PhotoCraft and code-based image editing.
When the result can be converted into layers, the user gains a workflow closer to traditional design: move an object, replace a background, adjust a position, or correct text without rebuilding the whole composition.
Still, “editable” should not be interpreted as a perfectly organized professional file. Separation quality will depend on the image, its complexity, and Magic Layers’ ability to recognize the components correctly.
Craft, Canva Design Model, and Dream Lab are not the same
| Tool or model | Main function | What it produces |
|---|---|---|
| Canva Craft | Generate images using Canva’s first-party model. | Images at up to 2K prepared for the design workflow. |
| Canva Design Model | Understand the structure, hierarchy, and logic of a complete piece. | Designs organized as editable elements. |
| Dream Lab | Give users an interface for image generation. | Visual options based on prompts and references. |
| Magic Layers | Interpret an image and convert its components into layers. | Elements that can be adjusted within the editor. |
Dream Lab is the access point; Craft may be the model generating the image; Magic Layers helps separate it; and the Canva Design Model works with the broader logic of the design. Distinguishing these components prevents attributing the entire process to one tool.
What its DesignArena results mean
Canva announced that Craft debuted in fourth place overall on DesignArena and achieved strong results in categories including people, portraits, landscapes, and nature. The company also used this ranking to describe Canva Research as one of the most advanced image generation labs.
DesignArena uses pairwise comparisons and community votes to produce its rankings. This method offers a view of human preferences, but it is not a definitive measurement of every model capability.
A position can change as new models appear, voting increases, or categories evolve. Performance in portraits or landscapes also does not automatically prove precision in text, products, brand identity, or complex instructions.
Rankings should therefore be treated as a snapshot from a particular evaluation rather than permanent proof of superiority.
How communicators and creators can use it
Canva Craft may provide value in tasks where speed and format adaptation matter:
- Exploring visual concepts before a campaign.
- Creating backgrounds and scenes for social posts.
- Proposing compositions for advertisements and covers.
- Generating images in horizontal, square, or vertical formats.
- Preparing variations for presentations, stories, and thumbnails, a workflow also related to our guide on creating Instagram carousels with Edits.
- Building a visual base that can later be edited in layers.
- Testing creative directions before producing final photography or illustration.
Its best use may not involve publishing the first result. Craft can serve as an exploration, visualization, and production stage before a professional applies design, brand, and content corrections.
A responsible workflow for Canva Craft
- Define the objective. Before writing the prompt, establish the audience, format, channel, and purpose of the image.
- Describe the hierarchy. Specify what viewers should notice first, where elements belong, and how much space the text needs.
- Generate alternatives. Comparing options prevents an initial output from becoming an automatic decision.
- Inspect the details. Hands, faces, objects, shadows, text, numbers, and spatial relationships may contain errors.
- Separate and edit. Use Magic Layers when appropriate and manually rebuild critical elements.
- Apply the brand identity. Colors, fonts, logos, and proportions should be checked against official assets.
- Validate rights and context. Before publishing, review licenses, unwanted similarities, and potentially misleading interpretations.
- Require human approval. Responsibility for the final message belongs to the professional or organization publishing it.
Availability and usage limits
Canva announced that Craft is available globally through Dream Lab. However, “available to everyone” does not necessarily mean unlimited use.
Image generation tools draw from Canva’s monthly AI allowance. The number of uses, speed, and available capabilities may depend on the subscription, prompt complexity, and Canva’s current policies.
For developers, Craft appears in the API as Canva’s first-party model. API documentation and capabilities can change, so teams should verify supported formats, permissions, and conditions before building a service around it.
Does Craft replace the designer?
Craft can reduce the time required to generate concepts, adapt formats, and create a first proposal. It does not independently decide which message an organization needs, how an audience will interpret an image, or when a visual solution contradicts the strategy.
Automation increases the number of options. Professional work shifts toward selecting, guiding, correcting, and creating meaning. The easier it becomes to produce technically attractive images, the more important it is to distinguish between a visually competent piece and genuinely relevant communication.
Judgment does not disappear when the model learns design principles. Hierarchy, alignment, and balance can be partially evaluated; intention, cultural sensitivity, originality, and brand coherence still require human decisions.
Frequently asked questions about Canva Craft
Is Canva Craft Canva’s first image generator?
No. Canva already offered visual generation tools. Craft is its first foundation image model developed and trained internally.
Where can Canva Craft be used?
Canva says it is available through Dream Lab and its public API. Effective access and generation limits may depend on the plan and monthly AI allowance.
What resolution does it generate?
The model can produce images at up to 2K. Resolution and available formats may vary according to the tool or integration being used.
Can it generate text correctly?
Canva says its architecture improves text planning and legibility. Names, figures, prices, and important messages should still be reviewed and, when possible, edited as independent text.
Can the images be edited?
Craft is designed to connect with Magic Layers, which can transform an image into editable elements. The result depends on the composition’s complexity and how accurately its components are identified.
Conclusion
Canva Craft matters for more than the quality of the images it can generate. Its significance lies in Canva beginning to control a technology layer that previously depended more heavily on external providers.
The planning architecture, design-oriented training, and connection with Magic Layers point toward a workflow where generation and editing are no longer separate stages. For communicators and creators, this may reduce friction and accelerate the production of variations.
A model that understands composition more effectively does not automatically understand strategy. The tool can organize elements, produce alternatives, and simplify corrections. Purpose, verification, and responsibility for the message still belong to people.
Sources and official documentation
- Canva: Introducing Canva Craft.
- TechCrunch: Canva releases its first image generation model.
- Canva Help Center: Generate images with Dream Lab.
- Canva Help Center: Understanding your AI usage.
- Canva Developers: Create image generation job.
- DesignArena: About its community evaluation and ranking methodology.