Artificial intelligence programs that transform text into images aren’t brand new. Neural networks like DALL-E are astonishing in producing simple, realistic images by using brief but instructive sentences.
Last week, I enjoyed being introduced to Imagen (opening the image in an entirely open tab). Google’s Brain Team created Imagen; Imagen is an AI similar in its capabilities to LDM and DALL-E. LDM. However, the aim of the Brain Team using Imagen is to produce images with higher levels of accuracy and fidelity using the same descriptive and short phrase technique for producing images.
An excellent example is this: examples on Images on the Imagen website Imagen’s website “A photo of a fuzzy panda wearing a cowboy hat and black leather jacket riding a bike on top of a mountain.” It’s a vivid description. However, the text is written to ensure that the AI can identify every object according to its criteria.
The AI can then examine each part of the sentence to create an easily digestible piece of information and attempts to produce the image as with the text as feasible. Despite some oddity or inconvenience in the background, Imagen can do this in a surprisingly short amount of time and with accurate results.
This is the exact reason why Google Brain Team has no plans to release Imagen accessible to the general public, at least until it can come up with additional security measures that will stop using AI to conduct malicious actions. This means that the preview available on the site is limited to a few carefully selected factors.
In the end, it’s the most effective choice. There are instances where it has been demonstrated AI programs are being released to the public on the internet… however, with highly unfavorable results. You may have heard of the Microsoft Tay AI Twitter account. It was among the AI Twitter account brought to the social media platform approximately five years ago.
Tay is an intriguing research project created by Microsoft. The concept was to see what the response of an AI would be towards people who are in the media. In just several weeks, Tay went from a regular chatbot to one that could be an anti-Semitic source of talk points. Tay had gone through the process of having it “modeled, cleaned, and filtered,” according to Microsoft (thanks to The Verge).
With the precedents created by AI like Tay, what Imagen was limited in is clear. Even the most extensive filtering might not be enough.
It’s far from the perfect format.
Although I was surprised by Imagen and had fun playing with words to create various bizarre images, it’s not something that I would consider persuasive. For the moment.
In most cases, Imagen returned some frighteningly hilarious results. Animals, mainly, were typically seen in all kinds of weird dimensions. A typical scene was a raccoon’s huge head or arms that looked like human girths gripping the bicycle’s handlebars. While they’re amusing, these strange results, when paired with photorealism, often produce a bizarrely bizarre result.
The idea of making the oil paintings was appealing, and the artwork Imagen could produce would not be out of the realm of school work. This is beautifully done. In the end, the picture of the image of a Persian cat playing with an instrument could be more convincingly transformed into paintings than an actual photo.
As previously mentioned, it’s doubtful that we’ll get the official announcement for Imagen shortly. And certainly not anytime soon, in actuality. The dangers created by AI programs and neural networks capable of producing content that isn’t pleasant remain high. In the meantime, I’m content to have Imagen as a cute little item for those looking to create hilarious cowboy-themed animals that skate down the mountains.