Google has a new image generation model to show you.
Credit: Google
AI is not showing any signal of the pace of progress to be sluggish. After upgrading the big image of the chat a few weeks ago, it is now the turn of Google to show new models to produce videos and pictures from the text prompt: We have been declared during Veo 3 (for video) and imagene 4 (for pictures), Google I/O 2025, and they come with some significant reforms.
Starting with Veo 3, this is the next step from the Veo 2 model that was recently pushed out to Mithun customers last month. Google says The Veo 3 brings with a remarkable improvement in real-world physics (some AI videos often conflict with) and details such as lip-syncing. In short: Your clip should look more realistic than ever.
Here is another important upgradation, and it is sound. Earlier, VO-made clips came without any audio attached, but AI is now smart enough to add to the appropriate environment sounds, including traffic noise, wildlife sounds, and even dialogues between characters.
Google has provided some examples videos to show new abilities, as you expect, including, Old sailorOf course, it is impressive that such a clip can be generated from a text prompt, and it is up to a higher standard in terms of realism-we are no longer getting six-fingered hands that we used to do with AI.
Nevertheless, the general identity of artificial intelligence is obvious: it is a normal sailor, on a normal sea, speaking a common dialogue about the sea. It is simultaneously mashing and average from every video of the sea and old sailors, which is trained on VO3, and may or may not be matched with the original prompt (which Google has not given).
The VEO 3 is available only to the brave to pay $ 250 per month for Google’s AI Ultra Plan, but VEO 2 is also getting some upgrade to people paying that tenth part for AI Pro. It is now better in control and stability, according to Google, with better camera movements and outpanning (expanding the view of a frame). It can now also go to connect and remove objects to the clip.
Moving on images: We have found imagene 4, 3 imagene 3 heirs. Here, we promise notable clarity in exact details such as complex details for high resolutions, water drops, and animal fur, support for higher resolutions (up to 2K) and more aspect ratio. You get top level results in both photorolic and abstract styles, according to Google.
What do you think so far?
Google’s AI World has large sheep as a tractor.
Credit: Google
Google has also faced one of one of the major problems with AI image generation, which is typography. Imagene 4 is clearly much better than models that came before making characters and words, which look united and accurate without any strange spelling or letters, which unknowingly dissolve in the painlip.
Imagene 4 is now available inside the Gemini app, for all users. Google has not mentioned any use limit, although if you do not have membership, you will kill these boundaries more quickly, as is with imagene 3 (there is no fixed quota for these boundaries, and it seems that they are dependent on general demand on Google’s AI Infrastructure).
Carefully curated samples have shown good by Google without any clear mistakes or impurities – just normal Ai Sheen. Imagene 4 is faster than the imagene 3, Google says, with a greater improvement in the way: a version on the model that is 10x faster than the imagene 3, is about to be launched soon.
There is another image and video tool to talk: FlowIt is an AI film production tool from Google that draws its lesson, video and image models together so that you can help you sew persistent scenes simultaneously with a characteristic of a similar characters and places together. If you are AI Pro or AI Ultra subscribers, you can use better models and flows with better models on a more expensive plan.
