Canva Unveils Canva Craft, a Generative AI for All Creatives

Key Takeaways

1Canva launches Canva Craft, its first proprietary AI image model aimed at designers

2The model is available to all users and is based on a two-step architecture

3Canva Craft ranks fourth on DesignArena, performing well in portraits but less so in marketing

4The debate on creative AI continues, with concerns about training and authenticity

Why it matters : Canva Craft marks a milestone in Canva's internal AI strategy and illustrates the ongoing tensions surrounding generative AI in creative professions.

Canva unveils Canva Craft, a generative image model presented as tailored for designers and available to all users. The company highlights a two-step architecture, advanced editing capabilities, and an internalized AI strategy. Initial public rankings are favorable for certain uses, amidst lively debates in creative professions.

Initial Rankings and Creatives' Concerns

The first benchmarking results place Canva Craft fourth overall on DesignArena, behind OpenAI's ChatGPT Images and ahead of Google's Nano Banana and Microsoft's MAI-Image models. The model performs well for generating portraits and product-focused images but is less effective in the marketing subcategory. Meanwhile, the controversy surrounding creative AI continues. Photographers, illustrators, and artists express concerns about the alleged training of AI models on their works without copyright respect, the creation of abusive content, and the portrayal of these tools as replacements for artists. Companies like Adobe and Canva defend AI as a tool for creators. Eric Wittman, CEO of VSCO, responded to a LinkedIn post by Melanie Perkins about Canva Craft, stating that human photographers will remain number one for portraits. Professionals must also increasingly verify that their works are indeed created by humans, as digital verification methods remain sporadic, complicating the situation for creators on platforms, including VSCO.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Fine Editing of Outputs with Magic Layers and Craft

Canva places particular importance on the ability to modify images generated by AI. Its Magic Layers feature allows users to break down an AI-generated design into fundamental elements for adjustment, a process described as technologically complex. Canva Craft builds on this foundation and allows for the modification of generated results to refine a visual when the initial output does not exactly match the intent.

Two-Step Architecture and Training Data

Canva describes a two-step architecture for Canva Craft, associated with prompt-writing tools capable of reasoning before generating a design. According to Stefano Corazza, AI can conceive the overall structure of the image, including layout, by mimicking the approach of designers who position elements to build a narrative. The model's training relies on Canva's library, public and licensed datasets, as well as information provided by users who have consented to participate.

Internalization Strategy and Global Availability

Canva launches Canva Craft, its first proprietary AI image model, designed for designers and accessible to all users worldwide. The company has integrated AI tools and third-party models for several years, following an initial step into creative AI with Magic Media in 2023. According to Stefano Corazza, head of AI research, the acquisition of Leonardo.Ai and the strengthening of internal research have supported the creation of this model. He notes that keeping AI technology in-house could reduce cloud costs and accelerate inference time.