Google’s Imagen 2 AI Model Gets Upgraded With Video Generation Capabilities

Google has upgraded its native artificial intelligence (AI)-powered image generation model Imagen 2 with two new capabilities. These capabilities were announced at its annual Cloud Next conference in Las Vegas. The AI tool is an enterprise-focused product available within the Vertex AI developer platform and can help create logo elements and other visual resources for companies. Now, the tech giant has also added video-generation capabilities to the AI model and it can generate up to four seconds long videos.

The video generation feature is being called ‘Text-to-live image’ by the company. With this capability, Imagen 2 can create four-second-long videos from text prompts. These videos will have a range of camera angles and motion. According to a report by VentureBeat, the generated videos will have 24 frames per second and be delivered at a resolution of 360×640 pixels. Google reportedly has plans to improve these metrics in the future.

Google also posted a video on YouTube to showcase the capabilities of its AI model. Google Cloud CEO Thomas Kurian reportedly said in a press release that animated images can improve user engagement. The generation quality appears to be similar to Runway AI and Pika 1.0 — both are available for the general public as well as enterprises.

There is also a question of deepfakes. This is relevant since Gemini AI recently came under fire after it showed “inaccuracies in some historical image generation depictions”. Shortly after, Google removed the image generation feature from Gemini. But the company says it will be different with Imagen. The company’s spokesperson told TechCrunch, “The Imagen 2 model in Vertex AI has not experienced the same issues as the Gemini app. We continue to test extensively and engage with our customers.” As for added protection, images and videos generated by the AI model will be labelled by Google DeepMind’s SynthID technology.

Apart from video generation, Imagen 2 has also received inpainting and outpainting capabilities. This will allow users to edit a certain part of the image instead of regenerating the entire thing with a new prompt. Making granular changes to an image is an easier method to get to the desired result. The feature is also offered by Microsoft’s Copilot and OpenAI’s DALL-E 3.