Recently, a mysterious AI image generation model codenamed "red_panda" has achieved remarkable results in the benchmark tests on the crowdsourced analysis platform Artificial Analysis, significantly outperforming products from industry leaders such as Midjourney, Black Forest Labs, and OpenAI.
According to data from the Artificial Analysis platform, "red_panda" led the text-to-image tests with a score of 1244, surpassing the second-place Black Forest Labs' Flux1.1Pro by approximately 40 Elo points. Notably, this scoring system originates from the international chess player ranking system and is now used for AI model performance comparisons. In practical combat data, "red_panda" has a 9% higher win rate than Flux, demonstrating a clear technical advantage.
In terms of performance, "red_panda" is equally impressive. The model can generate an image in an average of just 7 seconds, which is over a hundred times more efficient than OpenAI's DALL-E3. Comparative tests reveal that images produced by "red_panda" exhibit a high degree of realism, almost completely free from the unnatural artifacts common in traditional AI-generated works, and even surpass the realism of the renowned Flux model. Additionally, the model's ability to understand and execute text prompts is also outstanding.
Currently, "red_panda" remains in a closed beta phase and is not yet available to the public. Various speculations have emerged about the development team behind it: some believe it might be a reincarnation of Midjourney-V7, while others speculate it could be a new offering from a Chinese tech company. However, based on technical characteristics, the model is more likely a finely tuned Lora model trained on SD3.5large turbo, rather than from mainstream Chinese tech companies.
As AI image generation technology rapidly advances, the emergence of "red_panda" has once again pushed the industry standards higher. The origins of this mysterious new player and how its leading technology will influence the industry's development trajectory are worth ongoing attention.
Artificial Analysis Address: https://artificialanalysis.ai/text-to-image/arena