A new AI image creator has appeared without a creator's name or explanation and it is already outperforming its competition.

I'm already sceptical about AI image generation. I'm concerned about the negative feedback loop that could result from replacing artists with AI image generation. It is important to exercise caution, especially after red_panda - a brand new image-generation tool - was spotted on a website dedicated to voting for AI's best.

According to TechCrunch's report, a website called Artificial Analysis contains a section called Arena that allows users vote on which of two images best fits a given prompt. The two images are generated by different AI image-generation models.

Red_panda is the top player in the leaderboard with a 72% winning rate. This means that it has only lost 28 percent of the battles in which its images have been involved.

Some commenters on Artificial Analysis' post claimed that this could be a version of Midjourney's AI tool or Baidu's AI. This could certainly be true, given the creator's identity is obscured. But only time will tell exactly what red_panda does. This new tool has a high ELO (generally regarded as a quality metric) and a quick generation time of seven seconds. However, it does not have a price per image as the technology is still relatively unknown.

The win rate was impressive, and the Artificial Analysis X account noted it. It had a 79% winning rate at the time.

The name red panda may seem in poor taste, considering the energy costs associated with AI models of this type. This is because it is a critically endangered animal due to deforestation as well as climate change. The images in the battles are not created instantly, but rather based on the generation time. Some models can take as long as a minute to create an image. This game does not seem to have a significant impact on AI image generation if you're not a fan.

I voted in the "Arena", and was surprised at my own personal leaderboards. Red_panda was only on my leaderboard twice out of 60 choices.

With 11 total selections, my win rate dropped to 61%. Two different versions FLUX models dominated my personal leaderboard. The same two models are ranked two and three on the global leaderboard.

It's one of the few AI models that can handle text competently. It's also good at removing the ill-proportioned glossy look that many AI images have. The site itself had problems when it ran the same image against itself. One time, I received three images, of which one was the wrong prompt (and couldn't be clicked), and the other two were all wrong. Red_panda also won against an image with static colours, when the model or website simply did not display correctly.

This is also worth noting, I came across duplicate images during that time. These figures could be technically weighted if a business wanted to do that.

This image-generation tool may seem impressive based on its selection statistics, but it is important to note that its competition has been pretty poor. Its win rate has also decreased over time.

We don't know what it does, how it works or where it scrapes images to train on. The model seemed to be at the top of the AI imagery generation tools, but I still found some issues.