Video catalogue · Verified local path

NeoVerse local guide

Apache-licensed 4D world model for turning a monocular video or image into novel-trajectory MP4 video.

Choose an app

Start with the recommended app, or choose another supported option.

Compare video models
Open model filesRecommended verified starting pointOpen in UnslothLaunches Unsloth Desktop on this modelOpen Gradio setupLocal browser interfaceCheck my machineSee whether this model fits your saved hardwareOpen official sourcePublisher repository and primary evidence

Desktop app links require the app to be installed. If nothing opens, LocalClaw will show app-download and model-file fallbacks.

What it does

Video To VideoImage To VideoNovel View SynthesisCamera Controlled Video4d World Model

CASIA and CreateAI publish Apache-2.0 code plus a Hugging Face checkpoint with sharded safetensors, Wan2.1 VAE assets, UMT5 text encoder files, a WorldMirror-style reconstructor checkpoint and a LightX2V Wan2.1 LoRA. The official README documents CUDA 12.1/PyTorch 2.3.1 and CUDA 12.8/PyTorch 2.7.1 setups, hf download commands, CLI inference with output_path defaulting to outputs/inference.mp4, a four-step Gradio workflow, and optional Depth Anything 3 reconstruction. The default path reports about 47 GB allocated and 74 GB peak GPU memory, while --low_vram offloads models between phases and reduces peak GPU memory to about 38 GB, so LocalClaw records 128 GB RAM and 40 GB NVIDIA VRAM as the conservative local floor.

Hardware figures are practical entry floors, not performance guarantees. Resolution, duration, precision, offloading and runtime versions can materially change memory use.

Strengths

  • Official code, checkpoints and Gradio demo
  • Video or image input with predefined or custom camera trajectories
  • Documented low-VRAM offload mode reduces peak GPU memory to about 38 GB

Limits to know

  • Still workstation-class even in low-VRAM mode
  • Depends on Wan 2.1, LightX2V and reconstructor components
  • Specialized novel-view and 4D reconstruction workflow rather than a general video editor