Fixed 39-image reconstruction and DiT-facing latent comparison. Benchmark JSON.
Computed over 39 images and 102,380 spatial tokens using matched 128-channel H/16 × W/16 DiT-facing tensors. DINAC and FLUX.2 are whitened by their exported contracts; KVAE is raw 2×2-folded.
| Pair | Token mean | Token std | P5 | Median | P95 | Whole-image mean |
|---|---|---|---|---|---|---|
| DINAC-AE D2 ↔ FLUX.2 | 0.00460 | 0.08687 | -0.13909 | 0.00539 | 0.14593 | 0.00384 |
| DINAC-AE D2 ↔ KVAE-2D 2.0 | 0.00457 | 0.08671 | -0.13866 | 0.00520 | 0.14588 | 0.00380 |
| FLUX.2 ↔ KVAE-2D 2.0 | 0.97147 | 0.01844 | 0.93681 | 0.97621 | 0.99042 | 0.97475 |
Original and posterior-mode reconstructions; PSNR is measured in [-1, 1].
p640x1536:94623
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94624
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94625
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94626
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94627
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94628
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216264
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216265
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216266
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216267
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216268
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216269
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216270
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216271
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94699
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94700
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94701
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94702
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94703
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94704
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94705
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94706
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15577
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15578
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15579
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15580
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15581
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15582
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15583
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15584
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144131
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144132
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144133
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144134
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144135
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144136
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144137
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144138
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
exports/dico_diffae_hf/VAE_accuracy_test_image.png
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
Absolute RGB error with one shared 99.5th-percentile scale per image.
p640x1536:94623
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94624
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94625
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94626
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94627
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94628
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216264
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216265
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216266
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216267
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216268
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216269
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216270
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216271
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94699
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94700
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94701
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94702
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94703
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94704
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94705
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94706
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15577
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15578
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15579
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15580
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15581
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15582
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15583
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15584
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144131
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144132
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144133
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144134
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144135
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144136
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144137
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144138
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
exports/dico_diffae_hf/VAE_accuracy_test_image.png
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
Independent top-3 per-image PCA maps. KVAE is raw H/8 posterior mode folded 2x2 to H/16 with no whitening; only RGB permutation/sign is aligned.
p640x1536:94623
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94624
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94625
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94626
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94627
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p640x1536:94628
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216264
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216265
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216266
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216267
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216268
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216269
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216270
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p960x1024:216271
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94699
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94700
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94701
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94702
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94703
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94704
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94705
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
p704x1472:94706
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15577
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15578
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15579
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15580
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15581
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15582
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15583
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p1344x704:15584
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144131
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144132
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144133
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144134
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144135
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144136
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144137
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
r256_p896x1152:144138
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |
exports/dico_diffae_hf/VAE_accuracy_test_image.png
Original![]() |
DINAC-AE D2![]() |
FLUX.2 VAE![]() |
KVAE-2D 2.0![]() |