Genex World Initializer

🚀 GenEx-World-Initializer 🧭🌍

GenEx World Initializer is a panorama generation pipeline built on the FluxFillPipeline. It can transform a single view image into a 360° panoramic image using vision-conditioned inpainting.

• 🖼️ Input: One image (any size, will be center-cropped to square)
• 🧠 Prompt: Optional text to guide panoramic generation
• 🎯 Output: 2048 × 1024 equirectangular image
• 🧩 Mask: Uses a fixed panoramic mask

🚀 Quick Start

The following is a simple example to show you how to use this pipeline.

✨ Features

• Transforms a single view image into a 360° panoramic image.
• Supports optional text prompts to guide panoramic generation.
• Uses a fixed panoramic mask for inpainting.

📦 Installation

Make sure you have installed the following dependencies:

diffusers>=0.33.1
transformers
numpy
pillow
sentencepiece 

💻 Usage Examples

Basic Usage

from diffusers import DiffusionPipeline
from PIL import Image
import torch

pipe = DiffusionPipeline.from_pretrained(
    "genex-world/World-Initializer-image-to-panorama",
    custom_pipeline="genex_world_initializer_pipeline",  
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
).to("cuda")

# Load your image (any resolution)
image = Image.open("example_input.jpg")

# Run inference
front_view, output = pipe(image=image)
output.images[0]

🏁 Mask

The following mask is used to train the inpainting diffuser and used to inference automatically.

🔧 Technical Details

The model uses vision-conditioned inpainting to transform a single view image into a 360° panoramic image. It is built on top of the FluxFillPipeline.

✨ BibTex

@misc{lu2025genexgeneratingexplorableworld,
      title={GenEx: Generating an Explorable World}, 
      author={Taiming Lu and Tianmin Shu and Junfei Xiao and Luoxin Ye and Jiahao Wang and Cheng Peng and Chen Wei and Daniel Khashabi and Rama Chellappa and Alan Yuille and Jieneng Chen},
      year={2025},
      eprint={2412.09624},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2412.09624}, 
}

📄 License

This project is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).