---
title: Diffusion Once and Done (DOD)
url: https://www.emergentmind.com/topics/diffusion-once-and-done-dod
type: topic
---

# Diffusion Once and Done (DOD)

Diffusion Once and Done (DOD) denotes a one-step diffusion paradigm in which a model performs the task-relevant diffusion computation once and terminates, rather than executing the multi-step iterative denoising chain characteristic of standard diffusion models. The term appears explicitly in "Diffusion Once and Done: Degradation-Aware LoRA for Efficient All-in-One Image Restoration" [2508.03373], but closely related formulations include "You Forward Once" for text-to-image synthesis [2311.09257], "You Only Sample Once" for self-cooperative diffusion GANs [2403.12931], one-step diffusion by shortcutting probability flow paths [2512.11831], one-step diffusion scoring in inverse residual field space for unsupervised industrial anomaly detection [2604.18393], and Train-Once Plan-Anywhere kinodynamic planning with diffusion policies used as informed samplers inside sampling-based planners [2508.21001]. Across these works, DOD names a design objective: replacing repeated diffusion iterations with a single learned evaluation while preserving task-specific desiderata such as fidelity, restoration quality, anomaly separability, or planner-level guarantees.

## 1. Terminological scope and conceptual variants

The literature uses DOD in more than one operational sense. In image generation and image restoration, the most literal meaning is single-step sampling: a model maps a noisy latent directly to a clean latent or restored image in one forward pass [2311.09257, 2508.03373]. In anomaly detection, the same once-and-done principle is realized as single-step scoring rather than single-step generation: OSD-IRF computes an inverse residual field from a single call to the DDPM noise predictor and evaluates Gaussian typicality in that space [2604.18393]. In kinodynamic motion planning, the term is not used explicitly, but the paper states that DiTree embodies the intended sense of DOD because a diffusion-based policy is trained once in a single environment and then reused across unseen environments and robot systems without retraining, while the sampling-based planner preserves safety and probabilistic completeness [2508.21001].

A common misconception is that DOD is synonymous with distillation. The cited works show otherwise. The all-in-one restoration paper adopts distribution matching distillation to enable one-step restoration [2508.03373], but UFOGen uses a diffusion-GAN hybrid objective rather than ODE solvers or distillation [2311.09257], YOSO trains a one-step denoiser with self-cooperative adversarial learning [2403.12931], OSD-IRF avoids distillation and exploits an analytical residual-field construction [2604.18393], and "On the Design of One-step Diffusion via Shortcutting Flow Paths" reports a one-step model with FID50k of 2.85 on ImageNet-256x256 that requires no pre-training, distillation, or curriculum learning [2512.11831].

A second misconception is that DOD always concerns image synthesis. The surveyed papers place the paradigm in at least four domains: text-to-image generation [2311.09257, 2403.12931], all-in-one image restoration [2508.03373], unsupervised industrial anomaly detection [2604.18393], and kinodynamic motion planning [2508.21001]. This suggests that DOD is better understood as an inference-time principle than as a domain-specific model class.

## 2. Mathematical mechanisms for one-step diffusion

One family of DOD methods modifies diffusion training so that the learned model predicts a clean sample directly. UFOGen operates in the latent space of Stable Diffusion 1.5 and changes both the generator parameterization and the reconstruction target. Its objective applies an adversarial loss to the noisy marginal at $x_{t-1}$ while explicitly reconstructing clean data at $x_0$:
$$
\begin{aligned}
&\min_\theta\max_{D_\phi}\mathbb{E}_{q(x_0)q(x_{t-1}|x_0), p_{\theta}(x_{0}')p_{\theta}(x_{t-1}'|x_0')} \Bigl[ \log(D_{\phi}(x_{t-1}, t)) + \log(1-D_{\phi}(x'_{t-1}, t)) \\
&\quad + \lambda_{KL}\gamma_t\left\lVert x_0-x_0'\right\rVert^2 \Bigl].
\end{aligned}
$$
The paper argues that adversarial matching at $x_{t-1}$ implicitly matches the clean distributions because both marginals are convolutions with the same Gaussian kernel, while the direct $\|x_0-x_0'\|^2$ term makes clean alignment explicit and lower-variance [2311.09257].

YOSO also uses a diffusion-GAN hybrid, but it relocates adversarial matching to clean marginals $p_\theta^{(t)}(x_0)$ and smooths the target distribution by using the model’s own less-corrupted clean marginal $p_\theta^{(t-1)}(x_0)$ as the discriminator’s reference. This self-cooperative construction is designed to reduce discriminator overpowering and mode collapse relative to matching directly against $q(x_0)$, and the final objective adds both pointwise denoising and consistency regularization [2403.12931]. In this formulation, DOD emerges from a one-shot denoising generator trained under a smoothed adversarial divergence on clean marginals rather than from iterative reverse diffusion.

A second family of methods reaches one-step diffusion through shortcutting probability flow paths. The shortcut-model framework defines flow maps $X_{t,r}$ that jump from time $t$ to time $r$ directly and trains them by enforcing the consistency relation
$$
X_{s,r}(X_{t,s}(x_t)) = X_{t,r}(x_t).
$$
The paper provides Wasserstein-$2$ bounds of the form
$$
W_2(p_{\mathrm{data}}, p_\theta) \le C_1 L_{\mathrm{dtsc}}(\theta) + C_2 (t-s),
\qquad
W_2(p_{\mathrm{data}}, p_\theta) \le C_3 L_{\mathrm{ctsc}}(\theta),
$$
thereby giving a theoretical justification for one-step generation via shortcut objectives [2512.11831]. The continuous-time shortcut setting removes the explicit $(t-s)$ term and is reported to yield tighter control and better empirical fidelity.

A third mechanism is task-specific distillation. The restoration paper adopts DMD to distill multi-step latent denoising into a single reverse update and then recovers the clean latent by
$$
\hat{x}_0 = \frac{x_t - \sqrt{1 - \bar{\alpha}_{t^\star}}\, \epsilon_\theta(x_t, t^\star, c)}{\sqrt{\bar{\alpha}_{t^\star}}}.
$$
This one-step estimate is then decoded by a VAE decoder augmented with detail enhancement modules [2508.03373].

A fourth mechanism dispenses with generation entirely and performs one-step scoring. OSD-IRF defines the normalized residual field
$$
R_t := \frac{x_t - \sqrt{1-\beta_t}\,x_{t-1}}{\sqrt{\beta_t}} = \epsilon,
$$
so that $R_t \sim \mathcal{N}(0,I)$ for any neighboring time step. The inverse residual field is estimated by the DDPM noise predictor, either as $r_t=\epsilon_\theta(x_t,t)$ or, in the paper’s mean-path variant, as
$$
r_t = \epsilon_\theta(\sqrt{\bar{\alpha}_t}\,x_0, t).
$$
Anomaly scoring then reduces to Gaussian likelihood evaluation in IRF space [2604.18393].

## 3. One-step image synthesis

In large-scale text-to-image generation, UFOGen instantiates DOD as "You Forward Once." It uses a UNet generator and UNet discriminator in Stable Diffusion 1.5 latent space, with the frozen CLIP ViT-L/14 text encoder for conditioning and the SD VAE for decoding. Both generator and discriminator are initialized from the pre-trained SD 1.5 UNet. Inference is a single forward pass: sample $z_T \sim \mathcal{N}(0,I)$, compute $x_0' = G_\theta(z_T, t=T, c)$, and decode with the SD VAE [2311.09257]. On MSCOCO-2017 5k zero-shot, UFOGen (1 step) reports time 0.09 s, FID-5k 22.5, and CLIP 0.311; on MSCOCO-2014 30k, it reports time 0.09 s, 0.9B parameters, and FID-30k 12.78 [2311.09257].

YOSO frames the same objective as "You Only Sample Once." For text-to-image, it initializes the generator from PixArt-$\alpha$ or Stable Diffusion 1.5, uses a latent discriminator built from the pre-trained Stable Diffusion 1.5 UNet, and introduces a latent perceptual loss computed with SD HalfUNet features at $t=0$. It also addresses flawed terminal schedulers through Informative Prior Initialization and a quick adaptation stage that switches to $v$-prediction and zero terminal SNR [2403.12931]. On HPS v2.1 and AeS, YOSO-PixArt-$\alpha$ at one step reports HPS 28.21 and AeS 6.18; YOSO-SD 1.5 reports 28.11 and 6.25. With LoRA fine-tuning, YOSO-LoRA at one step reports HPS 23.34 and AeS 5.73 [2403.12931].

The shortcutting-flow-path literature places one-step image synthesis on a different footing: one-step generation is learned from scratch rather than obtained by finetuning a pre-trained diffusion backbone or by introducing an adversarial discriminator. The paper’s ESC configuration, built from continuous-time shortcutting, linear paths, variance reduction via a plug-in velocity, a gradual time sampler, and adaptive loss weighting, achieves 1-NFE FID50k = 2.85 on ImageNet-256x256 under classifier-free guidance, with no pre-training, distillation, or curriculum learning [2512.11831]. This result is significant because earlier one-step-from-scratch baselines in the same comparison are substantially weaker, including CT at FID50k $\approx 34.24$, IMM at $\approx 7.77$, and SCD at $\approx 10.60$ [2512.11831].

Taken together, these papers delineate three distinct synthesis routes to DOD: adversarial clean-latent matching with explicit reconstruction [2311.09257], self-cooperative adversarial learning over clean marginals [2403.12931], and shortcut consistency training over probability flow maps [2512.11831].

## 4. All-in-one image restoration

The paper that uses the name DOD explicitly defines it as an efficient all-in-one image restoration method that achieves superior restoration performance with only one-step sampling of Stable Diffusion models [2508.03373]. Its target setting is AiOIR, where a single model must handle heterogeneous degradations such as noise, rain, haze, blur, and low-light without relying on textual prompts.

The architecture retains the latent diffusion backbone of Stable Diffusion but adds three task-specific components. Multi-degradation Feature Modulation extracts degradation condition vectors $c$ from the intermediate $h$-space of a pretrained DDPM at timestep $t=0$, motivated by the empirical finding that these features are highly discriminative for different degradations [2508.03373]. Parameter-efficient Conditional LoRA then modulates the low-rank pathway by degradation-aware affine parameters:
$$
h = W_0x + B\,\phi(Ax \mid c), \qquad
\phi(Ax \mid c) = \gamma_c \odot (Ax) + \beta_c,
$$
with per-layer parameters produced from $[c,p_i]$ by a small mapper $M_\theta$ [2508.03373]. Finally, High-fidelity Detail Enhancement inserts RRDB-based Detail Enhancement Modules into the VAE decoder:
$$
F_m = F_d + \mathrm{RRDB}([ \mathrm{Conv}(F_e), F_d ]; \theta)\times w.
$$

Training is split into two stages. Stage 1 trains PLA/CLoRA while freezing the VAE decoder, combining a data term with DMD-based distribution matching. Stage 2 freezes PLA/CLoRA and trains only the decoder-side DEMs with reconstruction and SSIM losses [2508.03373]. Ablation identifies UNet-only conditional placement as preferable, and LoRA rank $r=8$ as the best trade-off; $r=4$ collapses training, while $r=16$ degrades slightly [2508.03373].

Empirically, the paper reports a three-task average over dehazing, deraining, and denoising at $\sigma=15,25,50$ of PSNR 31.87, SSIM 0.9122, LPIPS 0.0620, DISTS 0.0626, CLIPIQA 0.6660, NIQE 3.5212, MUSIQ 68.85, and MANIQA 0.7080 [2508.03373]. On five-task perceptual evaluation, it reports MUSIQ/MANIQA values of 67.99 / 0.7115 for dehazing, 70.49 / 0.7104 for deraining, 69.50 / 0.7219 for denoising at $\sigma=25$, 37.77 / 0.4286 for deblurring, and 73.75 / 0.6657 for low-light [2508.03373]. At 512×512 on RTX 3090, DOD reports 0.211 s inference time versus 17.664 s for DA-CLIP, 26.602 s for AutoDIR, 212.534 s for WeatherDiff, and 0.303 s for DiffUIR [2508.03373]. The paper summarizes this as approximately 84× faster than DA-CLIP while delivering superior perceptual quality [2508.03373].

The restoration setting also clarifies a central feature of DOD in low-level vision: one-step inference alone is usually insufficient. The method compensates by combining degradation-aware conditioning, parameter-efficient UNet adaptation, and decoder-side detail recovery [2508.03373].

## 5. Beyond image synthesis: anomaly detection and motion planning

OSD-IRF extends DOD to unsupervised industrial anomaly detection. The method trains an unconditional DDPM on normal data only, extracts features with EfficientNet-B4, and then evaluates the DDPM noise predictor once at a fixed time step to obtain the inverse residual field. Anomaly detection is performed by evaluating the probability density of the IRF under a Gaussian distribution and thresholding the score; with $\Sigma=I$, the image-level score reduces to $\frac{1}{2}\|r_t\|_2^2$ plus a constant [2604.18393]. The paper reports that this requires only single step diffusion because the IRF holds for any neighboring time step in the denoising process. On MVTec-AD it reports image AUROC 99.0, AP 99.6, F1-max 98.4; pixel AUROC 97.7, AP 54.3, F1-max 57.4, AU-PRO 93.5; mAD 85.7; FPS 133. On VisA it reports mAD 81.3 and FPS 105.9, and on MPDD mAD 83.8 and FPS 212.4 [2604.18393]. The text summarizes the speed gain as roughly a 2X inference speedup without distillation [2604.18393].

DiTree transfers the once-and-done intuition to kinodynamic planning. The method trains a diffusion policy once in a single environment, the D4RL AntMaze Large map, and then reuses it across unseen maps and two robot systems without retraining [2508.21001]. The policy samples short action sequences
$$
u_{1:N} \sim p_\theta(u_{1:N}\mid x_{\mathrm{near}}, x_{\mathrm{target}}, o),
$$
conditioned on the current tree state, a target encoded relative to that state, and a local occupancy observation; an RRT backbone then forward-propagates the sampled controls under the system dynamics, collision-checks the resulting segment, and inserts only feasible edges into the search tree [2508.21001]. The paper emphasizes that safety is enforced by collision checking and that returned solutions are collision-free by construction. It also states Theorem 1: for a Lipschitz-continuous system and a full-support action sampler, RRT-based DiTree is probabilistically complete, with solution probability at least $1-ae^{-bk}$ after $k$ samples for some constants $a,b>0$ [2508.21001].

The empirical evaluation trains once in a single environment and tests on 15 distinct unseen maps with 20 trials per scenario and a 120 s time limit per query. Aggregate CarMaze results are 73.3% success and 23.3 s runtime for DiTree, 45.0% and 67.3 s for RRT, 47.3% and 66.2 s for SST, and 46.0% and 20.7 s for DP-only. Aggregate AntMaze results are 35.7% success and 59.4 s runtime for DiTree, 7.3% and 54.5 s for DP-only, while classical SBPs fail entirely in that setup [2508.21001]. The paper summarizes DiTree as about 3× faster than classical SBPs on average and roughly 30% higher in success rate than all baselines across tasks [2508.21001].

These two papers broaden the meaning of DOD. In OSD-IRF, diffusion is executed once to produce a tractable diagnostic latent. In DiTree, the diffusion model is trained once and then repeatedly reused as an informed sampler inside a planner whose guarantees are inherited from the sampling-based planning backbone.

## 6. Guarantees, limitations, and open questions

The strongest formal guarantees in this literature appear in the shortcut-model and planning variants. Shortcutting-flow-path methods provide explicit Wasserstein-$2$ control for the learned one-step generator under discrete-time and continuous-time shortcut losses [2512.11831]. DiTree provides planner-level safety through collision checking, kinodynamic feasibility through forward propagation under $\dot{x}=f(x,u)$, and probabilistic completeness under Lipschitz dynamics and full-support action sampling; the paper further states that asymptotic optimality can be inherited by swapping the backbone to AO-RRT or SST [2508.21001]. By contrast, UFOGen and YOSO focus on training objectives and empirical quality rather than formal sampling guarantees [2311.09257, 2403.12931].

Training stability is a recurring issue. UFOGen introduces adversarial matching on a noisy marginal and explicit clean reconstruction because conventional one-step diffusion often sacrifices fidelity or requires multiple steps [2311.09257]. YOSO argues that directly matching generated clean marginals against real clean data destabilizes the discriminator and harms one-step efficacy; its ablations report FID 10.85 for direct matching to real data and 45.15 for matching over corrupted data, versus 3.82 for the self-cooperative design on CIFAR-10 [2403.12931]. The restoration paper addresses a different stability problem: one-step sampling tends to lose fine textures, so it supplements the denoiser with RRDB-based decoder enhancements [2508.03373].

The limitations are domain-specific. UFOGen notes missing objects, attribute leakage, and counting errors similar to SD-based models, and does not describe safety or content filtering measures [2311.09257]. YOSO identifies scheduler flaws as a source of artifacts in one-step generation and introduces Informative Prior Initialization plus quick adaptation to zero terminal SNR as remedies [2403.12931]. The restoration paper states that single-step sampling can under-restore extremely subtle textures or very challenging compound degradations without DEM, and that unusual artifacts not represented in the pretrained DDPM may weaken degradation conditioning [2508.03373]. OSD-IRF assumes anomalies are off-manifold and depends on accurate $\epsilon_\theta$; distribution shift in the normal-data training regime can degrade IRF typicality [2604.18393]. DiTree assumes full, static obstacle maps, can slow under severe distribution shifts or pathological local traps, and shifts the computational bottleneck from collision checking to diffusion inference [2508.21001].

A final conceptual clarification concerns what is being made "once and done." In synthesis papers, it is the reverse diffusion trajectory itself [2311.09257, 2403.12931, 2512.11831]. In restoration, it is the reverse update from degraded latent to restored latent [2508.03373]. In anomaly detection, it is the computation of a score-bearing latent variable rather than a reconstructed sample [2604.18393]. In planning, it is the training phase of the diffusion prior, not the planner’s search process, which still unfolds as an SBP tree expansion [2508.21001]. This suggests that DOD is not a single algorithmic recipe but a broader compression principle: the diffusion component is designed so that one evaluation, or one training investment followed by universal reuse, suffices for deployment.

Source: https://www.emergentmind.com/topics/diffusion-once-and-done-dod