If you are a user of X (Twitter), you’re probably tired of closing tabs where the social network recommends you buy the premium subscription to access the platform’s new features. One of them is Grok, the company’s artificial intelligence which has now become free to use (but limited compared to Premium) and which also already includes the option to generate images directly on X.
Additionally, X claims that «Grok can generate high-quality images in several facets where other image-generation models often have difficulties» including representations of logos and registered trademarks, and even of people.
Grok, X’s AI, can now generate images. Here’s how you can do it
To give your its creativity free rein, you only need to follow these steps.
- Open X, either in the app or on the web.
- Tap the Grok icon on the toolbar (next to the magnifying glass on smartphones, and under «Messages» on desktop)
- Enter your query in the chat bar (Make an image of…).
As is common with this type of AIs, you can refine the image results the more details you include in your “prompt”. For example, indicating in your request the image style (“anime style”, or “realistic style”), whether you prefer it to be a portrait, a half-body image or even, if you know a bit about photography, specifying the particular aperture you want and other parameters.
Putting Grok to the test
In our tests, we’ve asked it something quite geeky, but very specific, to check how real that fidelity X boasts about actually is.
First we asked it to draw Ten Shin Han (a notable animated character from the Dragon Ball series) using a technique that has never been used in the work. Without giving much context, it created a realistic image of a fictional person. It didn’t achieve what we expected.
Next, we tried to guide it a bit more and, after a few retries in which the image failed to finish generating (were the servers overloaded?), it still did not hit the target (the character neither resembled the original nor used the requested technique), although the drawing style did resemble that of the series.
Aside from the disappointment, one of the most interesting details of this tool is that, being multimodal, you can send it an input image for Grok to reimagine according to your instructions. Thus, if you send a photo of your cat and tell it to make it in anime mode, it will return it in a cartoon style. Or if you want to see how you’d look with a different hair color, just attach a photo of yourself and ask it to make you blonde/dark/redhead. This feature will be available later.
However, the “realistic” images are certainly astonishing given the high level of detail they exhibit. It’s hard to distinguish them from real ones, since in many cases there are hardly any “traces” or inconsistencies that would allow us to detect they were generated.
.
