Models compared: DALLE 3, DALLE 2, DALLE 3 HD, Stable Diffusion 3 Large, Stable Diffusion 2.1, Stable Diffusion 3 Large Turbo, Stable Diffusion XL 1.0, SDXL Lightning, Stable Diffusion 1.6, Midjourney v6, Midjourney v6.1, Playground v2.5, Stable Diffusion 1.5, Amazon Titan G1 (Standard), Stable Diffusion 3 Medium, Amazon Titan G1 v2 (Standard), FLUX.1 [pro], Playground v3 (beta), Ideogram v2, FLUX.1 [dev], FLUX.1 [schnell], Ideogram v2 Turbo, Ideogram v1, FLUX1.1 [pro], Stable Diffusion 3.5 Large, Recraft 20B, Stable Diffusion 3.5 Large Turbo, red_panda

Number of inference steps: for open source models with inference/DDIM step setting, creator defaults preferred. Stable Diffusion XL 1.0 (SDXL): 30. SDXL-Lightning (4-Step): 4. Stable Diffusion 1.6: 50. Stable Diffusion 2.1: 50. Playground v2.5: 50.

Other notable links: See also imgsys.org for another arena which also includes fine-tuned open-source models.

Full methodology will be published alongside leaderboard.