- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
ROAD trains one-step SDXL and SD3.5-M generators with a single non-saturating GAN loss against 500K LAION-Aesthetics images, with the teacher used only as initialization. The critic follows the projected-GAN using a frozen DINOv3 ViT-B, with three innovations: Lipschitz-controlled ViT heads with token-level cross-attention, random feature mixing re-drawn every 5k steps against "generator–critic co-adaptation", and Jacobian R1, which penalizes the full input–logit Jacobian. Validation: one-step COCO-10K FID of 11.03 (SDXL) and 13.73 (SD3.5-M), better than the multi-step CFG teachers. CLIP and preference scores, ablations of each component, and a time-to-FID comparison showing that ROAD reaches ODE-initialized DMD2's best FID with 1/9.25 of the GPU-hours.
Critical strengths
-
A very simple and effective training recipe. One loss, one stage, and no fake-score network, with a one-step performance comparable to the multi-step teacher model. This is a simple and cheaper method to one-step generators and would be appreciated by the community.
-
Alignment is preserved without a teacher target. A pure real-data GAN keeps CLIP at 0.338 on SDXL (teacher 0.331). This addresses the known weakness of adversarial distillation of poor prompt adherence.
Critical weaknesses
-
The FID comparison can be unfair The training target distribution is the LAION-Aesthetics data distribution (close to the coco dataset) (seen through DINOv3), so a lower COCO FID than a CFG teacher can be expected. In DMD2's own SDXL ablation, a GAN on the same 500K images reaches 13.77 FID (vs. 19.32) while CLIP drops to 0.307. On SD3.5-M, ROAD is worse than the teacher on HPSv2, ImageReward, PickScore and GenEval (Table 3).
-
The GAN objective is mode-seeking, yet diversity is never measured. The diversity of the generator can be critical for further post-training or test time scaling.
Questions for authors
- Each generator step decodes latents and backpropagates through the VAE decoder and DINOv3. What share of time and memory does decoding take? Would a latent-space backbone such as the teacher's features (as in LADD) work in your framework?
- Are Figure 3 and Figure 4 saying the two sides of the same thing?
- What are the SDXL and SD3.5-M teachers' FID and CLIP under your protocol for CFG ∈ {1.5, 2, 3, 4.5, 6}? Where does ROAD sit relative to that curve?
- What are the FIDs of Tab. 1 rows 6–8 (Text conditioning)? Does the ViTGAN head in row 4 use token cross-attention?
6 comments
Response to Reviewer YYJG (2/3): Weakness 2 and Question 1
W2. The GAN objective is mode-seeking, yet diversity is never measured. The diversity of the generator can be critical for further post-training or test time scaling.
Thank you for the insightful comment, and we agree that the concern is valid. We have now evaluated diversity. Experiments show that ROAD leads the distilled models on most diversity scores, and that it is more diverse than its teacher on all three main per-prompt scores.
Protocol. We follow DMD2 (App. B). Each method generates four images with seeds 0 to 3 for each of LADD's 128 PartiPrompts prompts. Per-prompt diversity is the mean pairwise distance among these images. The distances are LPIPS (DMD2), CLIP ViT-L/14 cosine distance (DP-DMD, ICML 2026), and the patch-level cosine distance of Harrington et al. (CVPR 2026). We compute the patch-level distance with PE-Spatial (Bolya et al., NeurIPS 2025) and DINOv3 in place of DINOv2. DINOv3 is for reference only, because ROAD's critic uses it. Recall (Kynkäänniemi et al., NeurIPS 2019) and coverage (Naeem et al., ICML 2020) use VGG-16 features with on COCO-10K. DMDX has no public checkpoint, so we omit it. Higher is better. Bold marks the best distilled model, and underline marks the second-best. Teacher rows are not ranked.
| │ | │ | |||||||
|---|---|---|---|---|---|---|---|---|
| Method | │ | LPIPS | CLIP | PE-Spatial | │ | Recall | Coverage | |
| SDXL | │ | │ | ||||||
| SDXL base (cfg 6) | │ | 0.633 | 0.168 | 0.535 | │ | 0.641 | 0.665 | |
| Hyper-SDXL | │ | 0.586 | 0.090 | 0.444 | │ | 0.311 | ||
| DMD2 | │ | 0.614 | 0.127 | 0.496 | │ | 0.570 | ||
| SiDa | │ | 0.184 | │ | 0.572 | 0.767 | |||
| SDXL-Turbo | │ | 0.605 | 0.097 | 0.485 | │ | 0.551 | 0.741 | |
| SDXL-Lightning | │ | 0.627 | 0.167 | 0.501 | │ | 0.572 | 0.639 | |
| NitroSD-Realism | │ | 0.639 | 0.115 | 0.501 | │ | 0.626 | 0.687 | |
| ROAD (ours) | │ | 0.674 | 0.556 | │ | 0.661 | 0.705 | ||
| SD3.5-M | │ | │ | ||||||
| SD3.5-M (cfg 4.5) | │ | 0.610 | 0.128 | 0.482 | │ | 0.721 | 0.670 | |
| SD3.5-M-Turbo | │ | 0.565 | 0.442 | │ | 0.561 | 0.611 | ||
| SenseFlow | │ | 0.641 | 0.121 | 0.491 | │ | 0.618 | 0.411 | |
| SwD | │ | 0.521 | 0.093 | 0.404 | │ | 0.675 | 0.477 | |
| SiD-DiT | │ | 0.136 | │ | 0.603 | ||||
| ROAD (ours) | │ | 0.650 | 0.157 | 0.536 | │ | 0.803 |
Results. The five main scores on SDXL and SD3.5-M give ten comparisons. Among the distilled models, ROAD has the highest value in 7 of the 10 comparisons. It has the second-highest value in 2 further comparisons. No other method has the highest value in more than 2 comparisons. The exception is SDXL coverage, where ROAD ranks fourth after SiDa, DMD2 and SDXL-Turbo. On both models, ROAD is also more diverse than its teacher on all three per-prompt scores.
Q1. What share of time and memory does decoding take? Would a latent-space backbone such as the teacher's features (as in LADD) work in your framework?
ROAD's reported cost already includes decoding, and we believe that its critic can use a latent-space backbone, because its components do not depend on the backbone.
On decoding, its cost is already included in the training GPU-hours of Fig. 1. Per generator step, VAE decoding takes 21.61% of the time and 6.69% of the peak memory.
On a latent-space backbone, ROAD's head, refresh and JR1 act on any grid of feature tokens, so the backbone can be exchanged. ROAD already works with three different frozen backbones, PE-Spatial, SigLIP2 and DINOv3, with close FID (Table 4b, L478–482). The teacher's features, as in LADD, fit ROAD naturally when the teacher is a DiT whose tokens form one uniform grid, as LADD shows with SD3. For a UNet teacher such as SDXL, the features form a multi-resolution pyramid, so ROAD will have different token numbers for different levels. In this latent-space setting, the real samples will be either generated by teachers as LADD or obtained by the VAE encoder. We thank you for this insightful question, and we think that extending ROAD to a latent-space critic is worth future work.
Response to Reviewer YYJG (3/3): Questions 2–4
Q2. Are Figure 3 and Figure 4 saying the two sides of the same thing?
Yes, they are the two arms of one controlled comparison. The runs share seeds, data and the diagnostic set, and they differ only in whether the mixing is fixed or re-drawn every 5k steps.
Q3. What are the SDXL and SD3.5-M teachers' FID and CLIP under your protocol for CFG ∈ {1.5, 2, 3, 4.5, 6}? Where does ROAD sit relative to that curve?
Thank you for this question. Our original teacher baselines use the established CFG settings of 6 for SDXL and 4.5 for SD3.5-M, consistent with official model guidance and previous studies such as DMD2. Following your suggestion, we have conducted these experiments, and the following tables give the results. Bold marks the best value and underlining the second-best value in each column.
Each table has two ROAD rows, and they differ only in the data used to select the checkpoint. The row "ROAD, 1 step (val set)" is the checkpoint of the submitted paper, which we selected on COCO-10K. This selection follows our closest predecessors, DMD2 and Projected GAN, and established GAN papers such as StyleGAN2 and StyleGAN3. To address a concern of Reviewer aA4p, we drew 10,000 COCO-2014 validation images and captions at random, disjoint from DMD2's COCO-10K, and we selected the checkpoint on this disjoint set. The row "ROAD, 1 step (disjoint set)" reports this checkpoint, which the revised Tables 2 and 3 now use. We show the val-set rows in gray for reference, and we do not rank them. Kindly refer to our response to Reviewer aA4p (W1 and Q1) for details on the checkpoint selection issue.
| SDXL base, cfg | FID ↓ | CLIP ↑ | HPSv2 ↑ | Pick ↑ | Aes ↑ |
|---|---|---|---|---|---|
| 1.5 | 33.57 | 0.275 | 25.71 | 20.69 | 6.08 |
| 2 | 24.07 | 0.296 | 26.46 | 21.16 | 6.13 |
| 3 | 19.14 | 0.315 | 27.32 | 21.65 | 6.16 |
| 4.5 | 0.326 | 27.97 | 21.96 | 6.18 | |
| 6 (default) | 19.29 | ||||
| ROAD, 1 step (disjoint set) | 11.25 | 0.338 | 28.61 | 22.14 | 6.62 |
| SD3.5-M, cfg | FID ↓ | CLIP ↑ | HPSv2 ↑ | IR ↑ | Pick ↑ | Aes ↑ | GE ↑ |
|---|---|---|---|---|---|---|---|
| 1.5 | 0.325 | 27.42 | 0.39 | 21.29 | 5.85 | 0.52 | |
| 2 | 15.46 | 0.335 | 28.06 | 0.67 | 21.73 | 5.97 | 0.61 |
| 3 | 16.23 | 0.341 | 28.70 | 0.88 | 22.13 | 6.06 | 0.65 |
| 4.5 (default) | 18.01 | 0.341 | 29.19 | 1.00 | 22.38 | 0.69 | |
| 6 | 19.12 | 0.344 | 6.09 | ||||
| ROAD, 1 step (disjoint set) | 13.89 | 28.90 | 0.93 | 21.85 | 6.17 | 0.66 |
On SDXL, ROAD is better than the teacher at every guidance scale on all five metrics. Even the teacher's best value in each column, such as FID 18.49 at cfg 4.5 or CLIP 0.331 at cfg 6, is behind ROAD. This shows that no guidance scale lets the teacher reach ROAD.
On SD3.5-M, ROAD has the best FID and aesthetic score, and it is better than the teacher's best-FID setting (i.e., cfg 1.5) on every metric. A higher guidance scale raises the teacher's preference scores but worsens its FID, so that at cfg 4.5 and 6 the teacher leads ROAD on most preference metrics while its FID is more than 4 points worse. These conclusions also hold for both ROAD checkpoints.
Q4. What are the FIDs of Tab. 1 rows 6–8 (Text conditioning)? Does the ViTGAN head in row 4 use token cross-attention?
The table below gives the FID and CLIP score of Table 1, rows 6–8, with row 4 for reference.
| Table 1 row | Critic head | Text conditioning | FID ↓ | CLIP ↑ |
|---|---|---|---|---|
| 4 | ViTGAN | Token-level cross-attention | 15.40 | 0.319 |
| 6 | ViT (ours) | Unconditional | 14.31 | 0.320 |
| 7 | ViT (ours) | Miyato-style pooled-text projection | 14.27 | 0.322 |
| 8 | ViT (ours) | Token-level cross-attention | 14.54 | 0.331 |
Rows 1 and 2 use Miyato-style pooled-text projection, and row 4 (ViTGAN) uses token-level cross-attention. We note that ViTGAN improves FID significantly over the convolutional head but improves CLIP score only marginally. We attribute this to its sub-optimal Lipschitz control strategies. Namely, it uses L2 attention with tied query and key projections, and applies spectral normalization to linear weights only at initialization instead of throughout training.
Response to Reviewer YYJG (1/3): Weakness 1
We sincerely thank you for your thorough and insightful review. We are grateful that you recognize "three innovations" in ROAD's critic, call ROAD "A very simple and effective training recipe" that "would be appreciated by the community", and note that it "addresses the known weakness of adversarial distillation of poor prompt adherence". Below, we respond to each of your points in turn; we hope these responses address your concerns, and we would be glad to clarify anything that remains unclear.
We thank you for this concern. We believe that ROAD's FID gains over the teachers are fair, because ROAD's training images are not close to COCO, and the gains come from the ability of GANs to synthesize higher-fidelity images. Importantly, unlike the DMD2 ablation you cite, ROAD does not achieve its FID gains at the expense of text alignment, while maintaining competitive human-preference scores.
1. ROAD's FID gains over the teachers are fair.
On the data distribution, to the best of our knowledge, no study reports that the image distribution of LAION-Aesthetics is close to that of COCO, although both datasets consist of real images. We therefore evaluated their closeness directly. We computed the FID of 10,000 random images (mean of five draws) from our training dataset, which is constructed from LAION-Aesthetics. The FID is 46.7 against real COCO images, whereas ROAD's generated images achieve 11.25 FID against real COCO-10K images. This substantial difference suggests that ROAD does not obtain its low FID simply by imitating a training distribution already close to COCO.
On the source of the FID improvement, in our view, the compelling FID comes from the ability of GANs to synthesize higher-fidelity images. In the ablations of ADD and DMD2, removing the adversarial loss gives oversmoothed or oversaturated images with a much worse FID (ADD, Table 1d; DMD2, Table 4). Moreover, these ablations show that relying more on the adversarial objective can improve FID with the same real images. In your own example, which concerns four-step models, DMD2 with the GAN term alone reaches 13.77, against 19.32 for DMD2 with both score distillation and the GAN term (DMD2, Table 4). ROAD's purely adversarial objective makes the fullest use of this ability.
2. ROAD achieves a strong balance between FID and other quality metrics.
Importantly, ROAD overcomes the CLIP score degradation observed in the DMD2 ablation you cite. While that GAN-only model achieves a lower FID at the cost of a CLIP score of 0.307, ROAD achieves both a substantially lower FID and a CLIP score of 0.338. On SDXL, ROAD also surpasses its teacher on both human-preference metrics, HPSv2 and PickScore, as well as aesthetic score (Table 2). As for SD3.5-M, we agree that ROAD is behind the teacher on preference metrics. However, it is worth mentioning that, on SD3.5-M, there is no other one-step model to our knowledge, so we are competing with four-step baselines. ROAD, the first one-step method on SD3.5-M to our knowledge, surpasses the teacher on FID, CLIP score and aesthetic score, and its gaps on the preference metrics are small. At the teacher's best-FID guidance scale, ROAD is ahead on every reported metric (see Q3). These results suggest that ROAD achieves strong FID without a general degradation in other quality metrics.