Get started, Google’s answer to Dall-E: another artificial intelligence creates impressive images

Google this week unveiled its “text-to-image” computer model. I leftwhich generates hyper-realistic images by studying dozens of billions of voices.

Pathways Autoregressive Text-to-Image, or Parts, studies image sets, which Google calls “image token”using them to create new images, the search giant said on a search website.

Parts images become more realistic when you have more parameters (cards and other educational material) to review. The model studies 20 billion parameters before generating a final image.

Parts differs from Image, a text-to-image generator designed by Google to use spread learning. The process trains computer models adding “noise” to an image to make it darker, as if it were static on a television screen.

The model then learns to decode the static to recreate the original image. As the model improves, you can convert what appears a series of random dots in an image.

Google is not releasing Parts or Image to the public because AI datasets carry the risk of bias. Because datasets are created by humans, they can inadvertently lean on stereotypes or misrepresent certain groups. Google claims that both Parties and Google Image have a bias towards Western stereotypes.

The research giant has invested heavily in artificial intelligence as a way to improve its services and develop ambient computing, a form of technology so intuitive that it becomes part of the background.

At his I / O developer conference in May, CEO Sundar Pichai said AI is used to help Google Translate add languages, create 3D images in Maps and condenses documents into short summaries.

Parts and Image aren’t the only text-to-image templates available. The Dall-E, VQ-GAN + CLIP, and Latent Diffusion models are other non-Google text-to-image models that have recently made headlines. Dall-E Mini is a text-to-image artificial intelligence open source available to the public, but trained on smaller data sets.

Since E, the technology has gone viral

Since last week, social networks have been inundated with very particular images. Generated by artificial intelligence, some are very faithful to the title. Others seem deformed. The truth is that Dall-E has become a trend: it is a site that generates a representation of what the user writes. Even what we ask does not exist in the real world.

Dall-E is a technology created by a San Francisco company, OpenAI. The company aims to create “a safe and useful general artificial intelligence”, and is recognized in the IT environment to work on GPT-3, a tool that generates images from very basic indications such as, for example, a text. They also use Copilot, a tool that helps automate coding for software engineers.

What does all this mean? That Dall-E can generate images with words. From an algorithm that considers 12 billion parametersthe system creates images with letters only.

From this, the networks were flooded with this kind of images:

The Parti goes in an even more refined direction in terms of results.