Agnes open-sources 28.7B-parameter MoE video model
Agnes has open-sourced Agnes-Video-2.0-MoE, an audio-video generation model built by converting its roughly 15.3B-parameter dense pretrained model DaVinci-MagiHuman into a mixture-of-experts design: most of the structure is kept, while parts of 32 of its 40 layers are each expanded into three expert networks, with routing selecting one expert per step so activated parameters stay near 15.3B while total parameters reach about 28.7B. Agnes says the model generates a 5-second video in roughly 30 seconds on a single H200 GPU — about 20% faster inference than the dense version and more than twice as fast as comparable MoE video models. The team also applies EENP elastic expert neuron pruning, first scoring expert computation channels separately for video, audio and text and then merging the scores to cut channels; at 60% pruning, average activated parameters fall to about 11.7B and denoising time drops from 26.83s to 20.92s (about 22% less), though the VBench-2.0 score is about 92.4% of the unpruned MoE, with details degrading more in high-motion scenes as pruning deepens.
As video generation moves into ads, e-commerce and content pipelines, inference efficiency shapes serving capacity and per-clip cost, making dense-to-MoE conversion plus pruning a practical lever for batch production.
AI 视频越来越逼真，生成它的算力却并不便宜。复杂的动作、连贯的镜头和细腻的画面，背后都需要大量计算。一旦进入批量制作，反复生成、修改和重试的开销便会累积，成为创作者与视频平台必须考虑的成本。
Agnes 开源的音视频生成模型 Agnes-Video-2.0-MoE 正是一次针对推理效率的技术探索：单张 H200 GPU 约 30 秒生成 5 秒视频，推理速度较稠密（Dense）模型提升约 20%，较同类 MoE 模型快 2 倍以上。
其做法是把 Dense 模型改造成 MoE 将容量做大，再以专家剪枝把实际计算做小。这为视频生成走向规模化生产趟出了一条降本的新路。
（MoE 模型案例）
已有 Dense 模型的 MoE 化改造
要让视频表现更复杂的场景、动作和时序关系，模型需要更大的学习与表达容量。但如果只是把原有模型做大，通常也会增加每次生成视频所需的计算量。Agnes 选择从已有模型入手，通过 MoE 改造，探索扩大容量与控制计算规模的另一种路径。
#Agnes#Agnes-Video-2.0-MoE#DaVinci-MagiHuman#EENP#Agnes Video 2.5#Pavo#NVIDIA H200#VBench-2.0#Artificial Analysis