Google announced its new PaliGemma 2 vision model

There are several models that give AI a set of eyes, and Google’s PaliGemma model is one of them. This is the company’s vision language model that’s able to identify objects and text in images. Google just launched the new PaliGemma 2 model, and it’s available to use now.

The first PaliGemma model is a useful tool. It’s useful for identifying objects in images along with captioning images. In fact, it can caption short videos as well. One of the more useful features of PaliGemma is the ability to answer questions about images. So, it’s an overall powerful model with a ton of use.

Google announced the new PaliGemma 2 model

Google officially introduced us to the PaliGemma model back in May. The company intends PaliGemma 2 to be a drop-in replacement for the first model. So, if you’re using PaliGemma version 1, then you’re encouraged to switch up.

Google offers PaliGemma 2 in several flavors. There are the 3 billion, 10 billion, and 28 billion parameter variants. Also, there are 224-pixel, 448-pixel, and 896-pixel resolutions.

As for other specs, PaliGemma 2 boasts long captioning. Also, Google says that it will go beyond just identifying objects. It seems that the model will be able to identify people and interpret their emotions. So, if a person is happy, sad, etc., then PaliGemma 2 will be able to pick up on that.

Also, it looks like the model will identify more of what’s going on in the scene in order to tell the whole narrative. Google says that PaliGemma is better when it comes to recognizing music scores, recognizing chemical formulas, identifying depth, and making chest X-ray reports.

Overall, it looks like Google delivered some pretty notable updates to PaliGemma. If you want to use it, then you can get the code for it in Hugging Face, Kaggle, and Ollama.

There’s another big model from Google

Google also released the private preview for its video generation model named Veo. The company announced it during Google I/O this year, and people have been waiting for it. If you’re using Google’s Vertex cloud platform, then you will be able to try it out. You can generate videos up to 1080p resolution.