A mysterious new image generation model has emerged

A mysterious new image generation model beats models from Midjourney, Black Forest Labs and OpenAI in a crowdsourced test of artificial analysis.

The model, called “red_panda,” is about 40 Elo points ahead of the next highest-ranked model, Black Forest Labs’ Flux1.1 Pro, on the Artificial Analysis text-to-image leaderboard. The artificial analysis uses Elo, a ranking system originally developed to calculate the relative skill levels of chess players, to compare the performance of different models being tested.

Similar to the Chatbot Arena community AI test, Artificial Analysis ranks models through crowdsourcing. For image models, artificial analysis selects two models at random and gives them a unique hint. It then presents the clue and the resulting images, and users choose which one they think best represents the clue.

Of course, there is some bias in this voting process. Artificial Intelligence voters are largely AI enthusiasts, and their choices may not reflect the preferences of the broader generative AI user community.

But red_panda is also one of the best performing models on the leaderboard in terms of generation speed. It takes an average of about 7 seconds to create an image of a model—more than 100 times faster than OpenAI DALL-E 3.

So where did red_panda come from? What company did this? And when can we expect its release? All good questions. However, AI labs are increasingly using community tests to build anticipation ahead of an announcement, so it may be a little while before we find out.