Physics-IQ benchmark reveals that generative video models exhibit limited physical understanding unrelated to their visual quality.
hub Canonical reference
Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps
Canonical reference. 80% of citing Pith papers cite this work as background.
22
Pith papers citing it
Background
80% of classified citations
abstract
Generative models have made significant impacts across various domains, largely due to their ability to scale during training by increasing data, computational resources, and model size, a phenomenon characterized by the scaling laws. Recent research has begun to explore inference-time scaling behavior in Large Language Models (LLMs), revealing how performance can further improve with additional computation during inference. Unlike LLMs, diffusion models inherently possess the flexibility to adjust inference-time computation via the number of denoising steps, although the performance gains typically flatten after a few dozen. In this work, we explore the inference-time scaling behavior of diffusion models beyond increasing denoising steps and investigate how the generation performance can further improve with increased computation. Specifically, we consider a search problem aimed at identifying better noises for the diffusion sampling process. We structure the design space along two axes: the verifiers used to provide feedback, and the algorithms used to find better noise candidates. Through extensive experiments on class-conditioned and text-conditioned image generation benchmarks, our findings reveal that increasing inference-time compute leads to substantial improvements in the quality of samples generated by diffusion models, and with the complicated nature of images, combinations of the components in the framework can be specifically chosen to conform with different application scenario.
hub tools
citation-role summary
background 4
baseline 1
citation-polarity summary
representative citing papers
TeDiO regularizes temporal diagonals in diffusion transformer attention maps to produce smoother video motion while keeping per-frame quality intact.
Oracle Noise optimizes diffusion model noise on a Riemannian hypersphere guided by key prompt words to preserve the Gaussian prior, eliminate norm inflation, and achieve faster semantic alignment than Euclidean methods.
PAMELA provides a multi-user rating dataset and personalized reward model that predicts individual image preferences more accurately than prior population-level aesthetic models.
PhaseFlow4D reconstructs and tracks 4D beam phase space from 2D projections via a latent diffusion model with built-in physics constraints, achieving 11000x speedup over full simulations while following time-varying conditions.
RF-Sampling enhances flow matching models by implicitly performing gradient ascent on text-image alignment scores via linear textual combinations and flow inversion.
Noise optimization during sampling recovers diversity in mode-collapsed diffusion models while preserving output fidelity.
IAFS is a training-free iterative inference-time scaling framework that uses adaptive frequency-aware particle fusion to resolve the perception-fidelity conflict in diffusion super-resolution models, outperforming prior scaling strategies.
dMLLM-TTS delivers up to 6x more efficient test-time scaling for diffusion MLLMs via O(N+T) hierarchical search and self-verified feedback, improving generation quality on GenEval across three models.
PG-DLM applies particle Gibbs sampling over full trajectories in diffusion language models to enable iterative refinement, yielding higher accuracy on reward-guided generation with theoretical convergence guarantees.
ICEdit achieves state-of-the-art instructional image editing in Diffusion Transformers via in-context generation, requiring only 0.1% of prior training data and 1% trainable parameters.
LatentBox is a latent-first storage system that cuts persistent storage for AI images by 78.7% while keeping mean and tail latency competitive with traditional pixel storage.
MIGA introduces two-stage alignment to close train-inference gaps and dual consistency enhancement via self-reflection and long-range guidance to achieve SOTA temporal consistency in infinite-frame video generation on VBench and NarrLV.
EPIC introduces predicate-guided inference-time search that lifts compositional T2I prompt accuracy from 34% to 71% on GenEval2 with 31-81% lower execution costs.
DeScore decouples CoT reasoning from reward scoring in video reward models using a two-stage training process to improve generalization and avoid optimization bottlenecks of coupled generative RMs.
FASTER models multi-candidate denoising as an MDP and trains a value function to filter actions early, delivering the performance of full sampling at lower cost in diffusion RL policies.
FRIGID scales a diffusion-based model for de novo molecular structure generation from mass spectra, reaching over 18% top-1 accuracy on MassSpecGym and tripling prior bests on NPLIB1 via large unlabeled training and inference-time fragmentation refinement with log-linear compute scaling.
VASR separates continuation and residual variance in reward-guided diffusion SMC, using optimal mass allocation and systematic resampling to achieve up to 26% better FID scores and faster runtimes than prior SMC and MCTS methods.
RSEdit adapts off-the-shelf text-to-image models into a collection of editing systems that follow text instructions while keeping geospatial structure intact in remote sensing images.
RAPO++ is a three-stage prompt optimization framework combining retrieval-augmented refinement, closed-loop test-time scaling, and LLM fine-tuning to enhance text-to-video generation quality.
CellFluxRL post-trains the CellFlux model with RL using seven biological reward functions to generate more biologically valid virtual cell images.
The serial scaling hypothesis formalizes inherently serial problems in complexity theory and demonstrates that diffusion models cannot solve them.
citing papers explorer
-
Do generative video models understand physical principles?
Physics-IQ benchmark reveals that generative video models exhibit limited physical understanding unrelated to their visual quality.
-
TeDiO: Temporal Diagonal Optimization for Training-Free Coherent Video Diffusion
TeDiO regularizes temporal diagonals in diffusion transformer attention maps to produce smoother video motion while keeping per-frame quality intact.
-
Oracle Noise: Faster Semantic Spherical Alignment for Interpretable Latent Optimization
Oracle Noise optimizes diffusion model noise on a Riemannian hypersphere guided by key prompt words to preserve the Gaussian prior, eliminate norm inflation, and achieve faster semantic alignment than Euclidean methods.
-
Personalizing Text-to-Image Generation to Individual Taste
PAMELA provides a multi-user rating dataset and personalized reward model that predicts individual image preferences more accurately than prior population-level aesthetic models.
-
PhaseFlow4D: Physically Constrained 4D Beam Reconstruction via Feedback-Guided Latent Diffusion
PhaseFlow4D reconstructs and tracks 4D beam phase space from 2D projections via a latent diffusion model with built-in physics constraints, achieving 11000x speedup over full simulations while following time-varying conditions.
-
Reflective Flow Sampling Enhancement
RF-Sampling enhances flow matching models by implicitly performing gradient ascent on text-image alignment scores via linear textual combinations and flow inversion.
-
It's Never Too Late: Noise Optimization for Collapse Recovery in Trained Diffusion Models
Noise optimization during sampling recovers diversity in mode-collapsed diffusion models while preserving output fidelity.
-
Iterative Inference-time Scaling with Adaptive Frequency Steering for Image Super-Resolution
IAFS is a training-free iterative inference-time scaling framework that uses adaptive frequency-aware particle fusion to resolve the perception-fidelity conflict in diffusion super-resolution models, outperforming prior scaling strategies.
-
dMLLM-TTS: Self-Verified and Efficient Test-Time Scaling for Diffusion Multi-Modal Large Language Models
dMLLM-TTS delivers up to 6x more efficient test-time scaling for diffusion MLLMs via O(N+T) hierarchical search and self-verified feedback, improving generation quality on GenEval across three models.
-
Inference-Time Scaling of Diffusion Language Models via Trajectory Refinement
PG-DLM applies particle Gibbs sampling over full trajectories in diffusion language models to enable iterative refinement, yielding higher accuracy on reward-guided generation with theoretical convergence guarantees.
-
In-Context Edit: Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer
ICEdit achieves state-of-the-art instructional image editing in Diffusion Transformers via in-context generation, requiring only 0.1% of prior training data and 1% trainable parameters.
-
LatentBox: Storing AI-Generated Images at Scale via a Latent-First Design
LatentBox is a latent-first storage system that cuts persistent storage for AI images by 78.7% while keeping mean and tail latency competitive with traditional pixel storage.
-
Enhancing Train-Free Infinite-Frame Generation for Consistent Long Videos
MIGA introduces two-stage alignment to close train-inference gaps and dual consistency enhancement via self-reflection and long-range guidance to achieve SOTA temporal consistency in infinite-frame video generation on VBench and NarrLV.
-
EPIC: Efficient Predicate-Guided Inference-Time Control for Compositional Text-to-Image Generation
EPIC introduces predicate-guided inference-time search that lifts compositional T2I prompt accuracy from 34% to 71% on GenEval2 with 31-81% lower execution costs.
-
Think, then Score: Decoupled Reasoning and Scoring for Video Reward Modeling
DeScore decouples CoT reasoning from reward scoring in video reward models using a two-stage training process to improve generalization and avoid optimization bottlenecks of coupled generative RMs.
-
FASTER: Value-Guided Sampling for Fast RL
FASTER models multi-candidate denoising as an MDP and trains a value function to filter actions early, delivering the performance of full sampling at lower cost in diffusion RL policies.
-
FRIGID: Scaling Diffusion-Based Molecular Generation from Mass Spectra at Training and Inference Time
FRIGID scales a diffusion-based model for de novo molecular structure generation from mass spectra, reaching over 18% top-1 accuracy on MassSpecGym and tripling prior bests on NPLIB1 via large unlabeled training and inference-time fragmentation refinement with log-linear compute scaling.
-
VASR: Variance-Aware Systematic Resampling for Reward-Guided Diffusion
VASR separates continuation and residual variance in reward-guided diffusion SMC, using optimal mass allocation and systematic resampling to achieve up to 26% better FID scores and faster runtimes than prior SMC and MCTS methods.
-
RSEdit: Text-Guided Image Editing for Remote Sensing
RSEdit adapts off-the-shelf text-to-image models into a collection of editing systems that follow text instructions while keeping geospatial structure intact in remote sensing images.
-
RAPO++: Cross-Stage Prompt Optimization for Text-to-Video Generation via Data Alignment and Test-Time Scaling
RAPO++ is a three-stage prompt optimization framework combining retrieval-augmented refinement, closed-loop test-time scaling, and LLM fine-tuning to enhance text-to-video generation quality.
-
CellFluxRL: Biologically-Constrained Virtual Cell Modeling via Reinforcement Learning
CellFluxRL post-trains the CellFlux model with RL using seven biological reward functions to generate more biologically valid virtual cell images.
-
The Serial Scaling Hypothesis
The serial scaling hypothesis formalizes inherently serial problems in complexity theory and demonstrates that diffusion models cannot solve them.