Google announced Veo AI for video creation and PaliGemma 2 for image and text processing

Google has introduced a new generative AI model called Veo, designed for video content creation. The tool, available through the Vertex AI platform, allows users to generate videos longer than a minute and in 1080p resolution.

No special skills are required to work with Veo – just ask a text query or upload an image. The AI ​​transforms the input data into a video sequence, taking into account the visual style and cinematic effects selected by the user.

The resulting videos can be edited, adjusted for individual elements, and added personalized details such as logos. Veo is aimed at creative professionals, marketers, and content creators, helping them quickly create visually appealing materials.

Google has introduced a new version of the PaliGemma 2 model, which is an evolution of the previous version and is designed to work with text and images. The announcement was made after a demonstration of Gemma 2’s capabilities at the I/O 2024 conference.

The PaliGemma 2 model expands the capabilities of its predecessor, which was focused on adding captions to images and videos, recognizing text, analyzing objects, and responding to visual queries. The new version has received a “long caption” function, which allows you to generate more detailed descriptions of visual content, including actions, emotions, and the general context of the scene.

Improved features:

• Long description generation: taking into account complex details and scene atmosphere.
• Analysis of complex structures: improved recognition of tables, chemical formulas, and musical scores.
• Spatial thinking: more accurate analysis of X-ray images and other medical data.

PaliGemma 2 is available in several versions with different numbers of parameters (3B, 10B, 28B), which allows it to be adapted to different tasks and data volumes. Performance is significantly improved, and the new model is compatible with the previous one, which simplifies its integration.

For the convenience of users, the PaliGemma 2 models and code are already hosted on platforms such as Kaggle, Hugging Face, and Ollama, making it accessible to developers and researchers.