Training-Free Predictive Spatio-Temporal Token Pruning for Few-Step Autoregressive Video Diffusion
cs.CV
Submitted: 2026-11-05
Updated: 2026-11-06
Terminology
Sources
- SkyReels-V2: Infinite-length Film Generative Model
- RegionE: Adaptive Region-Aware Generation for Efficient Image Editing
- Not All Tokens Need 40 Steps: Heterogeneous Step Allocation in Diffusion Transformers for Efficient Video Generation
- One-Forcing: Towards Stable One-Step Autoregressive Video Generation
- Towards Memory-Efficient Autoregressive Video Generation via Instance-Specific Parametric Absorption
- AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation
- Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models
- Train Short, Inference Long: Training-free Horizon Extension for Autoregressive Video Generation
- TriTD: Tri-Partite Trajectory-Distribution Distillation for Real-Time Autoregressive Video Generation
- Astraea: A Token-wise Acceleration Framework for Video Diffusion Transformers
- Flow caching for autoregressive video generation
- LIPAR: Training-free Latent Inter-Frame Pruning with Attention Recovery
- Movie Gen: A Cast of Media Foundation Models
- TokenTrim: Inference-Time Token Pruning for Autoregressive Long Video Generation
- MAGI-1: Autoregressive Video Generation at Scale
- Motion-Aware Caching for Efficient Autoregressive Video Generation
- LongLive: Real-time Interactive Long Video Generation
- Towards One-step Causal Video Generation via Adversarial Self-Distillation
- Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs