CDM migrates distribution matching distillation to continuous time via dynamic random-length schedules and active off-trajectory latent alignment, yielding competitive few-step image fidelity on SD3 and Longcat-Image.
hub Mixed citations
SDXL-Lightning: Progressive Adversarial Diffusion Distillation
Mixed citation behavior. Most common role is background (50%).
37
Pith papers citing it
Background
50% of classified citations
abstract
We propose a diffusion distillation method that achieves new state-of-the-art in one-step/few-step 1024px text-to-image generation based on SDXL. Our method combines progressive and adversarial distillation to achieve a balance between quality and mode coverage. In this paper, we discuss the theoretical analysis, discriminator design, model formulation, and training techniques. We open-source our distilled SDXL-Lightning models both as LoRA and full UNet weights.
hub tools
citation-role summary
background 4
baseline 3
method 1
citation-polarity summary
representative citing papers
Introduces a Bridge latent interface that maps mismatched student latents into teacher space, enabling distillation from modern diffusion teachers to compact one-step students and raising SD 1.5 HPSv3 from 5.4 to 9.4 while keeping one-step speed.
Introduces a marginal-alignment regularizer for reflow distillation of diffusion models that aligns endpoint marginals, supported by a telescoping TV bound and benchmark experiments.
Asymmetric CUDA-stream pipelining, a compile-friendly LLLite reformulation, and periodic conditioning refresh sustain 27-30 fps video stylization on a consumer GPU with a 2.13B MLLM text encoder and 0.39B distilled U-Net.
OctoT2I uses a no-supervision PSEL loop to discover model capability frontiers and route T2I tasks, reaching 0.96 GenEval score with 90.3% speedup over Flow-GRPO.
JEPA guidance steers diffusion models toward low-density regions under an implicit density from a world model, producing minority samples with improved fidelity and semantic validity over generator-centric baselines.
AsymFlow uses rank-asymmetric velocity prediction to reach 1.57 FID on ImageNet 256x256 and enables finetuning of latent flow models into superior pixel-space text-to-image generators.
Defines Conditional Distribution Matching (CDM) as finding inputs whose induced conditional distributions match a target distribution and proposes the MLGD-F inference-time algorithm using pretrained diffusion models to solve it without retraining.
GeoEdit constructs local tangent frames from small perturbations to initial noise, enabling Jacobian-free on-manifold edits in diffusion models via alternating tangent steps and diffusion projections.
GDMD replaces raw-sample rewards with distillation-gradient rewards in RL-guided diffusion distillation, yielding 4-step models that surpass their multi-step teachers on GenEval and human preference metrics.
1.x-Distill achieves better quality and diversity than prior few-step distillation methods at 1.67 and 1.74 effective NFEs on SD3 models with up to 33x speedup.
Popular LLM activation steering methods are shown to act as proportional controllers; a PID steering framework is proposed that improves robustness and outperforms baselines in experiments across model families.
Diffusion trajectory distillation is reframed as operator merging, yielding an optimal variance-driven merging strategy via Pareto dynamic programming in the linear Gaussian case and unavoidable approximation errors from exponential mixture growth in the nonlinear case.
A three-stage distillation plus AR mask cache converts a bidirectional DiT editor into a real-time causal streaming editor that preserves non-edited regions at 12.66 FPS.
Per-position AR prediction entropy jointly drives speculative AR decoding and an anti-symmetric single-step drift decoder, yielding 3.8–5.5× faster hybrid visual generation without distillation.
BudCache optimizes step cache policies for a fixed inference budget in diffusion models via combinatorial search, outperforming threshold heuristics in quality on FLUX.1-dev and Wan2.1.
Z-Image Turbo++ narrows the quality gap to 8-step generation via three distillation techniques tailored for the 2-step regime.
Geometry-Aware Distillation restores initial noise sensitivity in text-to-image distillation by matching Jacobian-vector products to align local geometric structure, improving diversity while maintaining fidelity.
A multi-agent video world model using simplex rotary agent encoding and sparse hub attention achieves better fidelity, controllability, and consistency than baselines while generalizing from 2 to 4 players.
RTDMD unifies KL minimization to a reward-tilted teacher into distribution matching plus reward terms, using AC-DMD in stage one and hybrid GRPO-style gradients plus SubGRPO in stage two to reach new SOTA on preference, aesthetic, and compositional metrics with 4-step generation on SD3, SD3.5, and F
RAVEN aligns training and inference for causal autoregressive video diffusion via interleaved rollout repacking and introduces CM-GRPO for direct RL on consistency-model kernels, claiming better quality than recent baselines.
FlashClear delivers up to 122x faster object removal than prior diffusion models via adversarial step distillation and asymmetric attention caching while preserving visual quality.
Embedding Loss aligns feature distributions via MMD in random network embeddings to boost one-step diffusion distillation, reaching SOTA FID of 1.475 on CIFAR-10 unconditional generation.
By requiring and using highly discriminative LLM text features, the work enables the first effective one-step text-conditioned image generation with MeanFlow.
citing papers explorer
-
Continuous-Time Distribution Matching for Few-Step Diffusion Distillation
CDM migrates distribution matching distillation to continuous time via dynamic random-length schedules and active off-trajectory latent alignment, yielding competitive few-step image fidelity on SD3 and Longcat-Image.
-
Cross-Space Distillation: Teaching One-Step Students with Modern Diffusion Teachers
Introduces a Bridge latent interface that maps mismatched student latents into teacher space, enabling distillation from modern diffusion teachers to compact one-step students and raising SD 1.5 HPSv3 from 5.4 to 9.4 while keeping one-step speed.
-
Beyond Trajectory Matching: Reflow with Marginal Distribution Alignment
Introduces a marginal-alignment regularizer for reflow distillation of diffusion models that aligns endpoint marginals, supported by a telescoping TV bound and benchmark experiments.
-
Inverting the Streaming-Diffusion Bottleneck: Video-Rate MLLM-Conditioned Edit Diffusion on a Consumer GPU
Asymmetric CUDA-stream pipelining, a compile-friendly LLLite reformulation, and periodic conditioning refresh sustain 27-30 fps video stylization on a consumer GPU with a 2.13B MLLM text encoder and 0.39B distilled U-Net.
-
OctoT2I: A Self-Evolving Agentic Text-to-Image Router
OctoT2I uses a no-supervision PSEL loop to discover model capability frontiers and route T2I tasks, reaching 0.96 GenEval score with 90.3% speedup over Flow-GRPO.
-
Beyond Generative Priors: Minority Sampling with JEPA-Guided Diffusion
JEPA guidance steers diffusion models toward low-density regions under an implicit density from a world model, producing minority samples with improved fidelity and semantic validity over generator-centric baselines.
-
Asymmetric Flow Models
AsymFlow uses rank-asymmetric velocity prediction to reach 1.57 FID on ImageNet 256x256 and enables finetuning of latent flow models into superior pixel-space text-to-image generators.
-
Inverse Design for Conditional Distribution Matching
Defines Conditional Distribution Matching (CDM) as finding inputs whose induced conditional distributions match a target distribution and proposes the MLGD-F inference-time algorithm using pretrained diffusion models to solve it without retraining.
-
GeoEdit: Local Frames for Fast, Training-Free On-Manifold Editing in Diffusion Models
GeoEdit constructs local tangent frames from small perturbations to initial noise, enabling Jacobian-free on-manifold edits in diffusion models via alternating tangent steps and diffusion projections.
-
Guiding Distribution Matching Distillation with Gradient-Based Reinforcement Learning
GDMD replaces raw-sample rewards with distillation-gradient rewards in RL-guided diffusion distillation, yielding 4-step models that surpass their multi-step teachers on GenEval and human preference metrics.
-
1.x-Distill: Breaking the Diversity, Quality, and Efficiency Barrier in Distribution Matching Distillation
1.x-Distill achieves better quality and diversity than prior few-step distillation methods at 1.67 and 1.74 effective NFEs on SD3 models with up to 33x speedup.
-
Activation Steering with a Feedback Controller
Popular LLM activation steering methods are shown to act as proportional controllers; a PID steering framework is proposed that improves robustness and outperforms baselines in experiments across model families.
-
Toward Theoretical Insights into Diffusion Trajectory Distillation via Operator Merging
Diffusion trajectory distillation is reframed as operator merging, yielding an optimal variance-driven merging strategy via Pareto dynamic programming in the linear Gaussian case and unavoidable approximation errors from exponential mixture growth in the nonlinear case.
-
LiveEdit: Towards Real-Time Diffusion-Based Streaming Video Editing
A three-stage distillation plus AR mask cache converts a bidirectional DiT editor into a real-time causal streaming editor that preserves non-edited regions at 12.66 FPS.
-
Drift-AR: Single-Step Visual Autoregressive Generation via Anti-Symmetric Drifting
Per-position AR prediction entropy jointly drives speculative AR decoding and an anti-symmetric single-step drift decoder, yielding 3.8–5.5× faster hybrid visual generation without distillation.
-
Budget-Constrained Step-Level Diffusion Caching
BudCache optimizes step cache policies for a fixed inference budget in diffusion models via combinatorial search, outperforming threshold heuristics in quality on FLUX.1-dev and Wan2.1.
-
High-Fidelity Two-Step Image Generation via Teacher-Aligned End-to-End Distillation
Z-Image Turbo++ narrows the quality gap to 8-step generation via three distillation techniques tailored for the 2-step regime.
-
Restoring Initial Noise Sensitivity in Text-to-Image Distillation via Geometric Alignment
Geometry-Aware Distillation restores initial noise sensitivity in text-to-image distillation by matching Jacobian-vector products to align local geometric structure, improving diversity while maintaining fidelity.
-
Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players
A multi-agent video world model using simplex rotary agent encoding and sparse hub attention achieves better fidelity, controllability, and consistency than baselines while generalizing from 2 to 4 players.
-
Reinforcing Few-step Generators via Reward-Tilted Distribution Matching
RTDMD unifies KL minimization to a reward-tilted teacher into distribution matching plus reward terms, using AC-DMD in stage one and hybrid GRPO-style gradients plus SubGRPO in stage two to reach new SOTA on preference, aesthetic, and compositional metrics with 4-step generation on SD3, SD3.5, and F
-
RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO
RAVEN aligns training and inference for causal autoregressive video diffusion via interleaved rollout repacking and introduces CM-GRPO for direct RL on consistency-model kernels, claiming better quality than recent baselines.
-
FlashClear: Ultra-Fast Image Content Removal via Efficient Step Distillation and Feature Caching
FlashClear delivers up to 122x faster object removal than prior diffusion models via adversarial step distillation and asymmetric attention caching while preserving visual quality.
-
Efficient Diffusion Distillation via Embedding Loss
Embedding Loss aligns feature distributions via MMD in random network embeddings to boost one-step diffusion distillation, reaching SOTA FID of 1.475 on CIFAR-10 unconditional generation.
-
Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation
By requiring and using highly discriminative LLM text features, the work enables the first effective one-step text-conditioned image generation with MeanFlow.
-
BiasIG: Benchmarking Multi-dimensional Social Biases in Text-to-Image Models
BiasIG is a multi-dimensional benchmark for social biases in T2I models that shows debiasing interventions frequently cause confounding discrimination effects.
-
Continuous Adversarial Flow Models
Continuous adversarial flow models replace MSE in flow matching with adversarial training via a discriminator, improving guidance-free FID on ImageNet from 8.26 to 3.63 for SiT and similar gains for JiT and text-to-image benchmarks.
-
ExpressEdit: Fast Editing of Stylized Facial Expressions with Diffusion Models in Photoshop
ExpressEdit delivers fast, artifact-free stylized facial expression editing inside Photoshop via a diffusion model plugin and an accompanying expression database.
-
Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis
DP-DMD preserves sample diversity in few-step image synthesis by applying a teacher-derived target-prediction objective to the first distillation step and standard DMD loss to the rest.
-
WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modeling
A real-time video diffusion world model that uses dual action control, reframed position encodings, and context-aligned distillation to keep generated environments consistent over hundreds of frames.
-
Masked Language Prompting for Generative Data Augmentation in Few-shot Fashion Style Recognition
Masked Language Prompting masks selected words in reference captions and leverages LLMs to produce diverse, semantically coherent completions for style-consistent generative image augmentation without fine-tuning.
-
Curvature-Adaptive Consistency Flow Matching: Autonomous Trajectory Optimization via Reinforcement Learning
A reinforcement-learning agent that picks high-error trajectory segments during consistency distillation improves few-step text-to-image generation on FLUX and SDXL.
-
Generator-Refiner-Examiner: A Tri-Module Data Augmentation Framework for 3D Human Avatar Learning from Monocular Videos
TrioMan is a tri-module data augmentation framework using a Generator for pose/camera perturbations, a Refiner with one-step diffusion, and an Examiner with dual-branch attention to improve 3D avatar learning from monocular videos, claiming better results than prior methods on two benchmarks.
-
Reward-Aware Trajectory Shaping for Few-step Visual Generation
RATS lets few-step visual generators surpass multi-step teachers by shaping trajectories with reward-based adaptive guidance instead of strict imitation.
-
TurboTalk: Progressive Distillation for One-Step Audio-Driven Talking Avatar Generation
TurboTalk uses progressive distillation from 4 steps to 1 step with distribution matching and adversarial training to achieve 120x faster single-step audio-driven talking avatar video generation.
-
The Serial Scaling Hypothesis
The serial scaling hypothesis formalizes inherently serial problems in complexity theory and demonstrates that diffusion models cannot solve them.
-
Stabilizing, Scaling & Enhancing MeanFlow for Large-scale Diffusion Distillation
Stabilizes MeanFlow for large-scale diffusion distillation via discrete warm-up and trajectory alignment, reporting better results on FLUX.1-dev and HunyuanImage 3.0.
- Teacher-Feature Drifting: One-Step Diffusion Distillation with Pretrained Diffusion Representations