Latent diffusion models rely on a decoder to map denoised latents to pixels, adding complexity and inference cost. We introduce Decoding Without a Decoder (DWD), which lets a finetuned diffusion transformer produce final pixels directly through a linear head. Unlike previous methods, where the denoising output leads to final pixels, DWD separates the two roles: the original denoising head continues to drive sampling, then the new DWD head predicts the pixels at the final step. We extend pretrained latent models with DWD efficiently by finetuning only the last few transformer blocks, training the DWD head with pixel quality losses while anchoring the denoising head to the pretrained model to preserve the sample trajectory. Across latent image and video generators, DWD removes the decoder and reduces inference cost while maintaining or even improving pixel quality. On Wan video generation, it visibly reduces detail distortion and flickering, lowering warping error by 23%. Remarkably, the same separation also improves pixel diffusion models that do not even use a decoder: a final-step DWD head produces substantially better texture detail in text-to-image generation, and improves ImageNet FID from 1.55 to 1.43, a new state of the art among pixel flow models.
Figures & tables
Figure 1: Given the same seed, DWD preserves similar content to the original model while improving fine detail. (a) On latent-space video diffusion, DWD removes the decoder and keeps text rendering legible across frames, whereas the baseline distorts it. (b) On pixel-space image diffusion, DWD recovers photorealistic skin and hair texture that the baseline oversmooths.
Figure 2: Standard latent generation vs. DWD. (a) A standard generator samples a denoised latent and maps it to pixels with a decoder. (b) DWD retains the denoising head for latent denoising and adds a parallel DWD head that produces the final pixels only at the final step, bypassing this decoder. DWD finetunes the pretrained generator’s final blocks and heads (fire icon), while freezing all previous blocks (ice icon). Time t and conditioning c are omitted for clarity.
Figure 3: Last-layer tokens support more accurate linear pixel reconstruction. We fit patch-wise linear probes from denoised latents z^0 and last-layer tokens h to ground-truth pixels, keeping the pretrained models frozen. (a) Held-out relative RMSE, with percentages showing error reduction over the latent probe. (b, c) Probes of last-layer tokens preserve finer image detail and reduce video ghosting before finetuning.
Algorithm 1 DWD finetuning with typical t choice.
Method
Decoder
Params (M) (+ decoder)
GFLOPs (final step)
FID↓
IS↑
SiT-XL/2-REPA
SD-VAE
674.8 + 49.5
2 × 237.2 + 622.2
1.38
306.2
DWD
–
678.5
2 × 237.7
1.36 (−0.02)
312.5 (+6.3)
DWD-conv
–
678.9
2 × 239.2
1.35 (−0.03)
313.6 (+7.4)
SiT-XL/2-REPA-E
E2E-VA-VAE
674.9 + 41.4
2 × 237.3 + 252.2
1.12
302.9
DWD
–
678.5
2 × 237.7
1.06 (−0.06)
303.1 (+0.2)
Latent models
DWD-conv
–
679.0
2 × 239.2
1.10 (−0.02)
305.7 (+2.8)
Table 1: Results on ImageNet 256×256. Shaded rows are our DWD extensions, finetuned from the baseline above them. Italic rows are ablations: DWD-conv adds convolutional refinement, and w/o separate head fuses the DWD and denoising heads so that a single head receives both pixel losses and trajectory anchoring. DWD has slightly more parameters than the baseline denoiser mainly due to the duplicated timestep embedding MLP.
Figure 4: A nonzero final timestep is necessary for the DWD quality gain. (a) At tfinal=0 , where the DWD head receives a fully denoised input, FID matches or trails the baseline’s, so the gain vanishes; a small positive timestep recovers it. Larger circles mark the defaults used in Table 1 . (b) Qualitative results also confirm the trend: DWD at its default timestep recovers skin and fabric texture, while DWD at tfinal=0 is as smooth as the baseline. The large panels are zoomed crops, with the full image inset.
Method
Params (B) (+ decoder)
TFLOPs (final step)
COCO-10k
OneIG ↑
LongText ↑
FID ↓
pFID ↓
HPSv3 ↑
HPS v2.1 ↑
FLUX.2 klein Base 9B
9.08 + 0.05
2 × 75.51 + 10.47
15.13
18.69
10.54
0.284
0.556
0.909
DWD
9.15
2 × 75.54
15.05
18.38
10.54
0.284
0.556
0.908
w/o adversarial loss
9.15
2 × 75.54
15.10
18.97
10.57
0.284
0.555
0.907
FLUX.2 klein 9B (4-step)
9.08 + 0.05
2 × 75.51 + 10.47
21.50
25.52
12.23
0.307
0.535
0.867
Latent
DWD
9.15
2 × 75.54
21.92
25.33
12.15
0.307
0.537
0.871
Table 2: Results of text-to-image generation (1024×1024). Shaded rows are our DWD extensions, finetuned from the baseline above them.
Figure 5: Text-to-image comparisons. Each pair shows the baseline (left) and its DWD extension (right) for the same prompt and seed. (a) On the latent FLUX.2 klein models, DWD closely matches the baseline’s content and texture without the decoder (zoom in to view detail). (b) On the pixel AnisoFlow model, DWD preserves the composition while recovering photorealistic texture in bark, foliage, and clouds. The large panels are zoomed crops, with the full image inset.
Method
Params (B) (+ decoder)
PFLOPs (final step)
Warping error (×10 -3 )↓
Image metrics
VBench dimensions
HPSv3 ↑
HPS v2.1 ↑
Subject consist. ↑
Backgr. consist. ↑
Temporal flicker. ↑
Motion smooth. ↑
Dynamic degree ↑
Aesthetic quality ↑
Imaging quality ↑
Wan2.1 14B
14.29 + 0.07
2 × 1.68 + 0.28
1.52
9.06
0.266
0.951
0.981
0.990
0.983
0.736
0.682
0.677
DWD
14.49
2 × 1.68
1.18
9.04
0.266
0.952
0.982
0.993
0.984
0.711
0.685
0.671
finetuned decoder
14.29 + 0.07
2 × 1.68 + 0.28
1.39
9.03
0.266
0.951
0.982
0.992
0.983
0.731
0.681
0.676
Wan2.1 14B (8-bit)
14.29 + 0.07
2 × 1.68 + 0.28
1.69
8.99
0.265
0.950
0.979
0.990
0.982
0.728
0.675
0.676
DWD
14.49
2 × 1.68
1.20
9.01
0.267
0.951
0.982
0.993
0.984
0.703
0.683
0.668
Table 3: Results of text-to-video generation (832×480, 81 frames) on VBench. Shaded rows are our DWD extensions, finetuned from the baseline above them. Finetuned decoder keeps the baseline DiT and finetunes its decoder on the same data with the same pixel losses as DWD (Appendix B.1 ). The 8-bit baseline and its DWD extension both use FP8 W8A8 quantization with 8-bit SageAttention2 ( Zhang et al., 2025 ) .
Figure 6: Text-to-video comparisons (BF16). Three frames from the baseline Wan2.1 14B (left) and its DWD extension (right) for the same prompt and seed, with two zoomed crops per strip. The baseline decoder warps the wire mesh into wavy, uneven lines (top) and breaks the dense pipes into speckle (bottom); neither defect is static, both shifting from frame to frame. DWD keeps the mesh regular and the pipe structure intact, and holds them steady as the camera tracks across the scene.
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
Model
K
Iters
Batch
Adv. start
λ
ptraj
ppix
Shift
Final t
Pixel space
GPU hours
SiT-XL/2-REPA
16 / 28
100k
1024
52.5k
5
[0.04,1]
[0,0.1]
1
0.08
Oklab
170
SiT-XL/2-REPA-E
8 / 28
200k
1024
152.5k
10
[0.04,1]
[0,0.1]
1
0.08
Oklab
200
JiT-H/16
2 / 32
4k
1024
2.5k
50
[0.04,1]
[0,0.1]
1
sampler
RGB
8
AnisoFlow-H/16
2 / 32
4k
1024
2.5k
50
[0.04,1]
[0,0.1]
1
sampler
RGB
8
FLUX.2 klein Base 9B
4 / 32
26k
mixed
21k
10
[0.04,1]
[0,0.1]
dyn.
0.3
Oklab
350
FLUX.2 klein 9B (4-step)
4 / 32
26k
128
21k
10
[0.3,1]
[0,0.5]
7
sampler
Oklab
340
Appendix
Table 4: DWD finetuning settings per model. K is the number of finetuned blocks out of the total. Adv. start is the iteration at which the generator adversarial loss is switched on, after a discriminator warmup. λ is the trajectory anchoring weight in Eq. ( 2 ), and ptraj and ppix are the timestep ranges of the two losses before the timestep shift. Mixed-resolution runs use the batch sizes given in Appendix A.2 , and FLUX.2 klein Base uses the model’s resolution-dependent shift (dyn.). GPU hours are estimated for H100 GPUs.
Method
Warping error (×10 -3 )↓
Image metrics
VBench dimensions
HPSv3 ↑
HPS v2.1 ↑
Subject consist. ↑
Backgr. consist. ↑
Temporal flicker. ↑
Motion smooth. ↑
Dynamic degree ↑
Aesthetic quality ↑
Imaging quality ↑
Wan2.1 1.3B (4-step)
2.84
8.32
0.265
0.938
0.962
0.976
0.979
0.825
0.660
0.669
DWD
2.07
8.46
0.258
0.946
0.959
0.984
0.983
0.811
0.657
0.648
DWD (self-distillation)
2.45
8.45
0.261
0.942
0.956
0.982
0.981
0.822
0.656
0.631
Appendix
Table 5: Few-step video generation on 4-step Wan2.1 T2V 1.3B (BF16). All models use the same sampler timesteps.
Model
Latency (s) ↓
Peak VRAM (GB) ↓
FLUX.2 klein 9B (4-step)
0.776
37.31
DWD
0.695
35.54
Wan2.1 1.3B (4-step)
4.661
19.22
DWD
2.757
16.15
Appendix
Table 6: End-to-end latency and peak VRAM of 4-step models.
Method
FID ↓
pFID ↓
OneIG ↑
LongText ↑
Training cost (H100 GPU hours)
FLUX.2 klein Base 9B
15.13
18.69
0.556
0.909
–
AsymFlow
18.10
22.62
0.506
0.777
1100 (15k iters)
DWD
15.05
18.38
0.556
0.908
0 350 (26k iters)
Z-Image Turbo
–
–
0.528
0.917
–
L2P
–
–
0.499
0.787
–
Appendix
Table 7: Comparison with latent-to-pixel adaptation (1024×1024). Z-Image Turbo results are from its official report.
Figure 7: Additional same-seed comparisons on AnisoFlow-L/16-T2I (1/3). Feather barbs and wool weave are recovered by DWD; the baseline blurs both into flat regions.
Figure 8: Additional same-seed comparisons on AnisoFlow-L/16-T2I (2/3). Skin texture in the photo booth frames and the small stamp and envelope lettering on the letter are sharper under DWD. The letter prompt is from LongText-Bench.
Figure 9: Additional same-seed comparisons on AnisoFlow-L/16-T2I (3/3). COCO and OneIG prompts. DWD recovers facial detail, fabric texture, and the individual cobblestones that the baseline smooths.