5 days Ago
The Google vs OpenAI battle rages on. After OpenAI announced ChatGPT Search to compete with Google's Search business, it seems like Google has landed another punch in the AI text and video generation battle. Google on Monday announced updates to its advanced AI models - Veo 2 for video generation and Imagen 3 for image generation.
Google claims that both of these deliver state of the art results, and on paper, the odds are stacked in Google's favour. Google says Veo 2 has an improved understanding of real-world physics, and it can generate "state-of-the-art results against leading models." Google claims that Veo 2 can truly understand the nuances of cinematography, and therefore it can deliver varying results with very specific prompts like different genres, or even a particular camera lens like 18mm, 50mm, etc.
If you want full bokeh-style lights in the background while your subject is lit up, according to Google, Veo 2 will deliver exactly that. Google also claims that its Veo 2 suffers from fewer hallucinations, a problem that has plagued AI video generators since the beginning. As AI generation tools become increasingly realistic, Google's blog post also details out the security measures it is undertaking with Veo 2.
All Veo 2 outputs will feature an invisible SynthID watermark to identify them as AI generated, therefore reducing the risk of misinformation and misattribution. Veo 2 can create clips in resolutions up to 4K (4096 x 2160 pixels). That’s 4x the resolution that Sora can achieve, as its videos are capped at Full HD 1080p resolution.
Moreover, they are limited to 20 second clips, another area where Veo 2 has a theoretical advantage with a clip duration of over 2 minutes, over 6x the duration of Sora. Veo 2's access is currently limited to Google Labs' VideoFX only. New users can head to Google Labs and sign up for the waitlist to get access to Veo.
There are also plans to integrate Veo 2 with YouTube Shorts and other Google products in 2025, making it clear that Google wants AI-powered content creation in the future. Imagen 3 is the latest version of Google's commercial image generation model, and it too has received a similar treatment like Veo 2. It now produces "brighter, better composed" images.
It now has the capability to render images in styles like photorealism, impressionism, and anime, per Google DeepMind. Google's blog post detailed that Imagen 3 can now follow prompts more faithfully and can render "richer details and textures.".
