Super-resolution ×4 — SeedVR2 vs Real-ESRGAN (DIV2K valid, RTX 5090)

Timings & VRAM

Variant	CFG	Steps	Load	Avg gen / img	Peak VRAM	PSNR ↑	SSIM ↑	LPIPS ↓	DISTS ↓	NIQE ↓	MUSIQ ↑	CLIP-IQA ↑	Success
Bicubic x4 (baseline)	—	—	—	0.0s	—	27.38	0.7796	0.3869	0.1866	7.16	31.95	0.5129	14 / 14
Ground truth (original HR)	—	—	—	—	—	—	—	—	—	2.93°	66.67°	0.6447°	14 / 14
Real-ESRGAN x4 (fp16)	—	1	4.2s	0.2s	1.66 GB	24.84	0.7185	0.2281	0.1166	3.46	65.71	0.5842	14 / 14
SeedVR2 3B (bf16, 1-step)	1.0	1	7.6s	3.1s	12.69 GB	23.11	0.6495	0.2344	0.1185	2.73	70.52	0.6886	14 / 14
SeedVR2 7B (bf16, 1-step)	1.0	1	9.1s	3.3s	22.09 GB	24.46	0.6986	0.1749	0.0900	2.90	69.83	0.6854	14 / 14
SeedVR2 7B-sharp (bf16, 1-step)	1.0	1	9.1s	3.3s	22.09 GB	24.51	0.6998	0.1755	0.0870	2.89	69.86	0.6761	14 / 14

Shaded columns are no-reference — they judge “does this look like a real photograph?” with no ground truth. They have no winner: the ° ground-truth row is the calibration line, and out-scoring it means inventing texture prettier than the original. The unshaded columns are full-reference and measure distance to the ground truth. A model can win one axis and lose the other — that is the perception–distortion tradeoff, not a contradiction.

Drag the slider to compare A vs B; switch either dropdown to pick any variant. The number next to each dropdown is that variant's generation time for this example. Click the input thumbnail to view it full size.