Artificial Analysis unveils image generation model leaderboard

Artificial Analysis has launched the Text to Image Leaderboard and Arena, which ranks image generation models based on human preferences.

This leaderboard calculates ELO scores based on over 45,000 human preference data points, comparing major proprietary and open source models such as Midjourney, DALL·E, Stable Diffusion, and Playground.

Key findings are as follows:

• Open source advances: While proprietary models like Midjourney and DALL·E 3 remain in the lead, open source models including Playground AI v2.5 are rapidly closing the gap, even surpassing DALL·E 3.
• Rapid technological progress: DALL·E 2, last year's leader, now sits near the bottom of the rankings, showing how quickly generational turnover is happening among models.
• Expected impact of Stable Diffusion 3: The soon-to-be-released Stable Diffusion 3 Medium is expected to make a significant contribution to improving open source community performance and the diverse fine-tuned model ecosystem.

Users can directly compare and vote on images from two models through the Image Arena, and after 30 votes, they can view their own personalized model rankings.