Diffusers
Safetensors
How to use from the
Use from the
Diffusers library
pip install -U diffusers transformers accelerate
import torch
from diffusers import DiffusionPipeline

# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("zihaosheng/NWM", dtype=torch.bfloat16, device_map="cuda")

prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Neutral World Model (NWM)

Breaking Survivorship Bias in Driving World Models

Paper | Project Page | GitHub

Model Description

The Neutral World Model (NWM) is a controllable video diffusion model for autonomous driving, built upon Stable Video Diffusion (SVD-XT) with a ControlNet adapter. Given an initial camera image, an ego trajectory, and optional agent bounding boxes, it generates 25 future frames at 512 ร— 320 resolution.

Training Data

NWM is trained on routine driving data from nuScenes and real accident videos from MM-AU. A two-stage training procedure and a bounding-box-guided objective improve the modeling of surrounding agents and collision dynamics.

Usage

Model weights are available on Hugging Face. Installation and inference instructions are provided in the GitHub repository.

License

The source code uses the MIT License. Pretrained SVD components are subject to their upstream licensing terms.

Downloads last month
-
Safetensors
Model size
2B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Paper for zihaosheng/NWM