New Image Generation Model Outperforms Competitors

According to TechCrunch, a new image generation model named 'red_panda' is outperforming models from Midjourney, Black Forest Labs, and OpenAI on the crowdsourced Artificial Analysis benchmark. The model is approximately 40 Elo points ahead of the next-best-ranking model, Black Forest Labs’ Flux1.1 Pro, on Artificial Analysis’ text-to-image leaderboard. The Elo ranking system, originally developed to calculate the relative skill level of chess players, is used by Artificial Analysis to compare the performance of various models it tests. Artificial Analysis ranks models through crowdsourcing, similar to the community AI benchmark Chatbot Arena. For image models, Artificial Analysis selects two models at random and feeds them a unique prompt. It then presents the prompt and resulting images to users, who choose which image better reflects the prompt. While there is some bias in this voting process, as Artificial Analysis’ voters are primarily AI enthusiasts, red_panda is also one of the better-performing models in terms of generation speed. The model takes a median of around 7 seconds to generate an image, which is over 100 times faster than OpenAI’s DALL-E 3. The origins of red_panda, including which company developed it and when it will be released, remain unknown. AI labs increasingly use community benchmarks to generate anticipation ahead of an announcement, so it might not be long before more information is revealed.