Yandex Launches YandexART 2.0 With Advanced Text Generation on Images

Yandex has introduced YandexART 2.0, a powerful upgrade to its visual neural network that adds text-generation capabilities and offers significant design tools for businesses and individual users.

Yandex presented a neural network prototype for generating images

YandexART 2.0 allows companies and creators to design logos, banners, and advertising materials with text generated directly on images. This release offers enhanced features, such as applying multiple styles to a single image, arranging elements naturally, and incorporating detailed prompts more effectively. Businesses can now create brand assets, product labels, and other marketing visuals with greater ease and customization.

The technology has already seen practical applications. Text.ru, a content service, uses YandexART to create visuals for web and blog materials. Another service, presentsimple.ai, leverages both YandexART and YandexGPT to automate slide presentation creation, combining content generation and visual design in a streamlined workflow. In Yandex Direct, 11% of advertisers are using YandexART-generated images, achieving 10-15% higher engagement in their campaigns.

Advanced Model Design

YandexART 2.0 combines convolutional and transformer neural network architectures. The convolutional component mimics human visual perception by recognizing edges, shapes, and textures in images, while the transformer model excels at handling more detailed text prompts. This hybrid architecture allows YandexART to understand and render complex image requests, such as a photorealistic can featuring an anime character.

The system was trained on hundreds of millions of images with detailed descriptions, improved with Yandex’s proprietary VLM model. It now also includes an extensive dataset of text-inclusive images, allowing YandexART 2.0 to generate Latin-text captions on images. The training focused on four parameters—relevance, aesthetics, defect minimization, and complexity. According to Yandex, YandexART 2.0 performed better than Midjourney V6.1 in complexity (66%) and aesthetics (58%) and closely matched it in relevance to user prompts.

Available on Yandex Cloud for Business Integration

YandexART 2.0 is now available through Yandex Cloud, with businesses able to access it via API integration. This provides companies a platform for faster, high-quality image generation, allowing for the creation of social media graphics, article illustrations, branding visuals, and more. Users can test the tool in demo mode or integrate it directly into their applications to fine-tune prompts.

User-Friendly Features for Everyday Creativity

Individual users can access YandexART 2.0 through the Pro option on Yandex’s virtual assistant, Alice. This option lets users generate customized images and refine them in real time by adjusting prompts in a conversational chat format. The feature is available across mobile and desktop platforms, providing an accessible tool for creating profile images, social media posts, app icons, and other personal or creative visuals.

YandexART’s quick image generation (within seconds) allows users to easily complete everyday tasks, such as designing social media content or creating unique prints for clothing.

About YandexART Technology

YandexART, a diffusion-based neural network, specializes in rendering images and animations based on text prompts. Trained on a dataset of 850 million image-description pairs, it understands Russian cultural context and uses a unique text-recognition algorithm. YandexART’s reach spans across Yandex services, such as Yandex Business, Direct, Browser, and Market, and is available to companies through Yandex Cloud.

YandexART 2.0 promises to be a powerful tool for both creative users and business-oriented applications, expanding possibilities for quick, high-quality, and customized visual content creation.