AI-generated image detection on Synthbuster (Model-Specific Performance) benchmark leaderboard · SOTA2 Research

AI-generated image detection on Synthbuster (Model-Specific Performance)

99.8DALL·E 2 ScoreDDA*

Evaluation Results

Method	Links
DDA*2026.05		99.8	99.9	99.9	99.5	99.9	99.9	99.9	99.9	99.9	99.8
STAL2026.05		99.6	99.9	99.8	98.6	99.9	99.9	99.9	99.9	99.9	99.7
Barni et al.2026.06		94.63	49.13	49.75	—	97.38	95.5	95.5	92.75	93.88	83.56
Li et al.2026.06		88.38	52.38	45.63	—	94.75	80.5	83.13	93.88	50.5	73.64
DDA2026.05		86.3	90	91.9	76.5	93.5	92.9	92.7	93.3	93.5	90.1
Khan et al.2026.06		84.63	41.13	85.38	—	58.63	71.38	72	74.25	79.5	70.86
UnivFD2026.05		83.5	47.4	89.9	53.3	52.5	70.4	69.9	75.7	68	67.8
DRCT2026.05		77.2	86.6	84.1	82.6	73.7	86.6	86.6	83.2	71.3	84.8
Color Transformer (Balanced)2026.06		61.75	61.75	53.38	—	88.25	94.88	95	76	82.13	76.64
He et al.2026.06		61.29	44.29	56.57	—	49.86	59.43	57.57	42	52	52.88
FatFormer2026.05		59.4	39.5	60.3	72.7	44.4	53.7	54	52.3	69.1	56.1
SAFE2026.05		58	9.9	10.3	52.2	56.7	59.4	59.1	53	59.5	46.5
C2P-CLIP2026.05		55.6	63.2	59.5	86.7	52.9	75.2	76.7	69.2	77.7	68.5
Tan et al.2026.06		53.55	53.55	50.85	—	81.35	97.1	97.65	76.8	98.4	76.16
NPR2026.05		51.1	49.3	46.5	48.5	52.8	51.4	51.8	46	52.8	50
AlignedForensics2026.05		50.2	48.9	51.7	53.5	98.7	98.8	98.8	98.6	97.3	77.4
Wang et al.2026.06		49.75	49.75	55.95	—	50.25	50.1	50.15	51.7	51.75	51.17
Gragnaniello et al.2026.06		49.4	49.4	68.2	—	52.7	57.05	57.55	56.95	61.9	56.64
Color Transformer (Robust)2026.06		48.38	48.38	45.25	—	76.5	92.13	92.5	56.13	92.13	68.92
Corvi et al.2026.06		48.1	48.1	57.15	—	97.15	98.1	98.1	96.65	95.3	79.83
Cozzolino et al.2026.06		47.8	47.8	77.95	—	97.5	97.8	97.8	97.35	96	82.5
Ojha et al.2026.06		46.3	46.3	88.8	—	51.4	69.4	68.85	74.65	66.9	64.07
AIDE2026.05		34.9	33.7	24.8	65	57.5	74.1	73.7	53.2	68.4	53.9