AI-generated images are becoming increasingly difficult to recognize. Sure, there are common things, but those things are becoming less common with each new AI image model. For example, you can rely on AI-generated hands that include too many or too few fingers of people; Nowadays, this is not necessarily the case.
One particular area of ​​weakness for AI image models has been text generation. The image itself may look believable, but take a closer look at the words, and you’ll often see that they’re not real. In fact Correct. Some may be exact, while others are pretty close, but, in many cases, you’ll see a lot of inconsistencies: lots of repeated letters, letters that aren’t really letters, letters that blend in and out of each other. Many of these oddities remind me how Star Wars language looksAt least when the AI ​​is trying to replicate American English.
OpenAI’s Images 2.0 is capable of generating highly realistic AI images
But the latest AI models are getting much better with text generation. In fact, OpenAI’s latest model for ChatGPT, Images 2.0, Can render highly realistic textand a Very About it – to the point where I’m not sure many of us (or Any (of us) will be able to recognize it. According to OpenAI, Images 2.0 is the company’s first image model with thinking capabilities: this means the model can take its time breaking down each step of the request, generating more detailed or accurate images, as well as the ability to generate up to eight images from a single prompt (though this is only available to paid customers). Free users can still take advantage of Images 2.0 features, such as how it searches the web for information and double-checks its work. The company says that “the results seem less AI-generated and more intentionally designed”, which essentially means that images generated using ChatGPT will be more difficult to identify going forward.
The company seems quite confident about this latest model. This explains the number of different types of images it can create – not just photorealistic photos, but also screenshots of a computer’s UI, a magazine collage, a pile of rice (he is Lots to generate), a journal page, and a handwritten essay. They also mean handwritten notes down to the coffee stains on the paper. You can scroll through OpenAI’s official announcement post to see these examples How amazingly realistic they are. The post includes other examples, such as highly realistic photos, graphic novel pages, movie posters, and images with different aspect ratios – right up to the iPhone’s panorama view.
What do you think so far?
All of these developments are as frustrating as they are impressive, but for me, it’s the improvements to the text that really take it to another level. Many models are getting quite good at creating images that trick users into thinking they are real, but the level of detail in the text and writing on these examples is something I have yet to see. I asked ChatGPT to design a menu for an Italian restaurant, highlighting five dinner courses and two desserts — I left the specific recipes up to the AI’s discretion. As far as I can tell, it managed to produce something realistic with dinner and dessert entrees without mistakes. Then I asked him to prepare a newspaper entry announcing that the Red Sox and Yankees would be changing cities. He did the same, without any apparent mistake.
Credit: Lifehacker
I’m not saying these are pictures Excellent: They still have an AI “shine” that a trained eye or close observer will be able to notice. OpenAI says Images 2.0 struggles with some complex tasks, such as puzzles, as well as details found in hidden or awkwardly placed areas, such as inverted surfaces. But none of that really matters when the images produced by this model are impressive enough to fool most people who scroll through them. Infographics, photos, maps, comics, movie posters, you name it: people are going to use this tool, and you’re going to start seeing a lot more AI images in your life – often without even realizing it.
