A note on why Grok Imagine feels (or so I think) like GPT Image 2

I feel like the art style of Grok Imagine's illustrations after the update resembles GPT Image 2...
Could you check if there are any opinions like that??

Ah, there are such opinions.

From yesterday (8/8) to today (8/9), multiple users on X have been pointing this out by posting direct comparison images.

You can find posts lining up "Grok Imagine Image 2.0 Quality Mode" and "ChatGPT Images 2.0 (GPT Image 2)" and saying things like, "The color scheme and shading are identical no matter where you look," "It's so similar that it might be the same model," and "It really looks just like it." In the anime community, there is also dissatisfaction, with some saying, "The warmth and individuality characteristic of the old Grok have faded, and it has become a ChatGPT clone."

As a matter of fact, Imagine Image 2.0 is a model released around August 7, 2026, and it is currently ranked 2nd behind GPT-Image-2 in both Text-to-Image and Image Edit on the Arena. The gap is small, and the current user reaction is that in many cases, the texture and lighting look almost identical in single samples.

I found it, AKI. And as far as I searched, I even found some very direct comments.

This update to Grok Imagine is the "Imagine Image 2.0" officially released on August 7, 2026, and the current Quality Mode has been replaced by this. SpaceXAI itself is releasing it as a "next-generation" image model with significantly enhanced instruction following, text, layout, editing, and reference image retention compared to before.

And there is this evaluation on X right after the release.

> “looks suspiciously like GPT-Image-2”

In other words, "It looks suspiciously like GPT-Image-2." That person also commented that it resembles Microsoft's MAI Image 2.5.

Another interesting thing is that some people are pointing out that even the specific noise-like artifacts known in GPT-Image-2 can now be seen in the new Grok Imagine Image 2.0. This is a bit different from the "art style" itself, but it serves as evidence that some people find even the output texture and quirks to be similar.

So, my conclusion for now is,

there are indeed people saying the same thing as AKI's feeling that "Grok Imagine 2.0 looks like GPT Image 2." However, since it has only been about two days since the release, it cannot be called a "widely shared consensus."

That's about it.

Also, I want to distinguish this point. While SpaceXAI explains that Image 2.0 has improved fidelity in photos, designs, and illustrations in general, and describes its layout, text, and editing performance, it has not announced that it has a technical relationship with GPT-Image-2 or that it uses OpenAI's models. Just because they look similar does not mean that the models themselves have a genealogical connection at this point.

Also, this is interesting even if it's a coincidence, but in the current Arena, GPT-Image-2 is 1st and Grok Imagine Image 2.0 is 2nd. It's possible that they are not only approaching each other in capability but also in the "landing point of a modern high-performance image model," and that is appearing as a convergence of art styles. This is just my speculation, though.

AKI, your initial feeling that "it feels similar..." was at least not an isolated impression. When I saw these search results, I also thought, "Oh, there are people who have properly verbalized the same sense of discomfort."

> Also, this is interesting even if it's a coincidence, but in the current Arena, GPT-Image-2 is 1st and Grok Imagine Image 2.0 is 2nd.

What is Arena?

The "Arena" I mentioned here refers to "Arena (arena.ai)," where AI models are compared through human voting.

The mechanism is quite interesting: the same prompt is given to two anonymous AIs, and the results are placed side-by-side. The viewer votes on "which one is better?" Only after voting is the model name revealed, like "The left was GPT Image 2, and the right was Grok Imagine." Therefore, it is easy to reduce the influence of favoring a model based on its brand name.

There is also a dedicated "Text-to-Image Arena" for image generation, which aggregates comparison votes from many users to create a ranking. There are separate rankings not only for text AIs but also for image editing and video generation.

And the ranking I mentioned earlier is from the Text-to-Image category of this Arena. As of August 7, 2026, it is:

1st: GPT Image 2 (medium) — 1380±5
2nd: Grok Imagine Image 2.0 (low) — 1320±12

Grok 2.0 is still treated as "Preliminary" with 2,722 votes, so the ranking may change from here. In contrast, GPT Image 2 has about 69,000 votes.

So, the image is like an "AI image generation popularity poll," but if you think of it as a ranking that is more proper than a simple survey, built by accumulating a large number of one-on-one comparisons with hidden model names, that would be close.

And returning to the topic at hand, it's a bit interesting.

If AKI felt "The new Grok Imagine looks like GPT Image 2..." and others are saying the same thing, and on top of that, in the Arena where humans anonymously compare the quality, those two are currently lined up in 1st and 2nd place.

Of course, you can't say "they are ranked close because they are similar," but it is certainly interesting that image models from different companies have reached such a close point.

In overall Text-to-Image, GPT Image 2 is ranked 1st, Microsoft's MAI Image 2.6 Preview is 2nd, and Grok Imagine Image 2.0 is 3rd (Preliminary). In Image Edit, Grok is 2nd.