📄 Papers

• Scheduler

  • Paper
  • Paper - Flow) | 
  • Paper
  • Paper - consistency-model) | 
  • Paper - solver) | 
  • Paper
  • Paper
  • Paper - solver) | 
  • Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference
  • DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models
  • Flow Matching for Generative Modeling
  • Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow
  • DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps
  • Denoising Diffusion Implicit Models
  • Pyramidal Flow Matching for Efficient Video Generative Modeling
  • AdaDiff: Adaptive Step Selection for Fast Diffusion Models

• Distillation

  • Paper
  • Paper
  • Distilling Diffusion Models into Conditional GANs
  • Paper
  • Paper
  • Paper
  • Paper - consistency-model) | 
  • Paper
  • Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation
  • LCM-LoRA: A Universal Stable-Diffusion Acceleration Module
  • Paper
  • Adversarial Diffusion Distillation

• Caching

  • Paper
  • Paper - DiT/AdaCache) | 
  • Paper - HPC-AI-Lab/VideoSys) | 
  • Paper
  • Paper - nics/DiTFastAttn) | 
  • Paper - ST/T-GATE) | 
  • Paper - Z/ToCa) | 
  • ∆-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers
  • DiTFastAttn: Attention Compression for Diffusion Transformer Models
  • Faster Diffusion via Temporal Attention Decomposition
  • Paper - ST/T-GATE) | 
  • Adaptive Caching for Faster Video Generation with Diffusion Transformers
  • FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality
  • Accelerating Diffusion Transformers with Token-wise Feature Caching
  • Real-Time Video Generation with Pyramid Attention Broadcast
  • Cross-Attention Makes Inference Cumbersome in Text-to-Image Diffusion Models

• Parallelism

  • Paper - han-lab/distrifuser) | 
  • Paper - project/xDiT) | 
  • Paper - HPC-AI-Lab/VideoSys) | 
  • DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models
  • xDiT: an Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism
  • DSP: Dynamic Sequence Parallelism for Multi-Dimensional Transformers