Canva launches homegrown image AI to rival competitors

Canva has launched its first in-house image generation model, with a notable distinction from most competitors: it was trained on design templates rather than photographs. The approach reflects Canva's design-first identity and sidesteps some of the copyright concerns that have surrounded photo-trained models. It positions the tool as purpose-built for graphic design output rather than photorealistic imagery.

Kyutai has released CADRE, a family of seven open causal diffusion video autoencoders designed specifically for interactive world models. The models achieve 20 frames per second on a single H100 GPU with a reaction latency under 230 milliseconds - a threshold considered important for interactive applications to feel responsive.