AI-Generated Image Detection on CommunityAI (In-the-wild) benchmark leaderboard · SOTA2 Research

AI-Generated Image Detection on CommunityAI (In-the-wild)

99.8Real AccuracyNPR

Evaluation Results

Method	Links
NPR2026.02		99.8	7.6	53.7
AIDE2026.02		99	9.3	54.2
LaDeDa2026.02		98.6	2.6	50.6
Gram-Net2026.02		98.5	6.1	52.3
SAFE2026.02		98.4	8	53.2
FreqNet2026.02		97.7	12.8	55.3
CNNSpot2026.02		96.9	11.2	54.1
DDA2026.02		96.8	72.5	84.7
PE-CLIP-Linear2026.02		96.6	97.5	97.1
DINOv3-Linear2026.02		94.9	94.6	94.8
MetaCLIP2-Linear2026.02		92.6	95.4	94
OMAT2026.02		85.3	41.4	63.4
SigLIP2-Linear2026.02		82.6	90.5	86.6
UnivFD2026.02		77.8	49.5	63.6
DINOv2-Linear2026.02		60.6	56.2	58.3
SigLIP-Linear2026.02		37	85.7	61.4
MetaCLIP-Linear2026.02		35.3	93.3	64.3
Effort2026.02		22.5	84	53.3