The arena concept doesn’t work for image models due to watermarks.
There are no watermarks in the arena.
There are no visible watermarks, but model makers can use steganographic codes to identify outputs from their own models.
Text-to-Image Models Leave Identifiable Signatures: Implications for Leaderboard Security
https://arxiv.org/pdf/2510.06525
This is true, however LMArena does employ some methods to mitigate attempts to manipulate the leaderboard, see https://openreview.net/forum?id=zf9zwCRKyP
They also control for style https://news.lmarena.ai/sentiment-control/