FastVideo and TurboDiffusion
These are competitors offering different acceleration approaches for video diffusion models—FastVideo provides a unified inference and post-training framework, while TurboDiffusion focuses specifically on model acceleration through distillation and pruning techniques, so practitioners would typically choose one based on their priorities between inference speed, training efficiency, and implementation flexibility.
About FastVideo
hao-ai-lab/FastVideo
A unified inference and post-training framework for accelerated video generation.
Supports full model fine-tuning and LoRA adaptation for video diffusion transformers, alongside Distribution Matching Distillation and sparse attention techniques achieving >50x denoising speedup. Provides optimized inference through sequence parallelism and multiple attention backends (including Video Sparse Attention), with a Python API and CLI supporting H100/A100/4090 GPUs across Linux/Windows/macOS. Integrates with Hugging Face model hub and supports both autoregressive and bidirectional video generation architectures.
About TurboDiffusion
thu-ml/TurboDiffusion
TurboDiffusion: 100–200× Acceleration for Video Diffusion Models
Combines SageAttention and Sparse-Linear Attention (SLA) modules for efficient transformer computation, paired with rCM timestep distillation to reduce sampling steps from dozens to 1–4 without quality loss. Supports both text-to-video and image-to-video generation across Wan2.1 and Wan2.2 models at 480p–720p resolution, with quantized checkpoints optimized for consumer GPUs (RTX 5090) and unquantized variants for H100-class hardware.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work