Views
No views yet
models/ directory:huggingface-cli download yeates/aurora-weights --local-dir models| Path | Component | Notes |
|---|---|---|
aurora_editor.safetensors | Video editor | trained dit + mllm.context_projector + ref_vae_condition (~9.4 GB, bf16) |
aurora_agent_vlm/ | Agent planner adapter | PEFT LoRA (r=32, alpha=64) on Qwen/Qwen3-VL-8B-Instruct |
aurora_editor.safetensorsdit.* — the WAN2.2-TI2V-5B diffusion transformer (fine-tuned)mllm.context_projector.* — projects frozen Qwen3.5-4B hidden states into DiT widthref_vae_condition.* — multi-reference conditioning with per-reference index embeddingaurora_agent_vlm/Qwen/Qwen3-VL-8B-Instruct,
r=32, lora_alpha=64, on the attention + MLP projections. adapter_config.json
records base_model_name_or_path = Qwen/Qwen3-VL-8B-Instruct.models/ and installing the code repository:1# Editor
2from evaluation.pipeline_loader import load_v2_pipeline
3pipe = load_v2_pipeline("models/aurora_editor.safetensors", device="cuda:0", ref_max_items=8)
4
5# Agent planner (LoRA merged at load)
6import aurora.agent
7agent = aurora.agent.AgentVLM("models/Qwen3-VL-8B-Instruct", "models/aurora_agent_vlm", device="cuda:0")models/ (WAN2.2-TI2V-5B,
Wan2.2_VAE.pth, Qwen3.5-4B, Qwen3-VL-8B-Instruct) — see the code repository's
Model Zoo. The full inference recipe (3-pass CFG defaults, per-benchmark
commands) is in the repository README.1@article{yu2026aurora,
2 title={Aurora: Unified Video Editing with a Tool-Using Agent},
3 author={Yu, Yongsheng and Zeng, Ziyun and Xiao, Zhiyuan and Zhou, Zhenghong and Hua, Hang and Xiong, Wei and Luo, Jiebo},
4 journal={arXiv preprint arXiv:2605.18748},
5 year={2026}
6}