LongLive-Plug: Once-for-All Distillation for Video Generation
A reusable LoRA distillation layer is meant to spare video models from repeating the same speed and long-context training work.
LongLive-Plug distills capabilities once on a base video model, then applies them training-free to compatible downstream models. The paper says those adapters cover single-pass classifier-free guidance, few-step sampling, and long-context error correction for autoregressive generation. The authors report deployment across 54 downstream models, spanning three backbone families and eight task categories. Code and models are listed as available. HF Daily Papers' note
LongLive-Plug distills capabilities once on a base video model, then applies them training-free to compatible downstream models. The paper says those adapters cover single-pass classifier-free guidance, few-step sampling, and long-context error correction for autoregressive generation. The authors report deployment across 54 downstream models, spanning three backbone families and eight task categories. Code and models are listed as available. HF Daily Papers' note
score 4