EditP23: 3D Editing via Propagation of Image Prompts to Multi-View

▶️ Project Page 📄 arXiv 💻 GitHub

This is the official Gradio demo for EditP23, a method for fast, mask-free 3D editing that propagates 2D image edits to multi-view representations in a 3D-consistent manner. The edit is guided by an image pair, allowing users to leverage any preferred 2D editing tool, from manual painting to generative pipelines.

1. Input Images

See the "How to Use" tab for details on generating the Multi-View Image and creating your own Edited Condition.

Original Multi-View Image (src_mv.png)

Source Condition (src.png)

Target Condition (edited.png)

2. Parameters

T_steps

Total number of denoising steps.

1 100

n_max

Number of scheduler steps for edit-aware guidance. Increase for more significant edits.

1 50

Source CFG

Guidance scale for the source condition. Can typically remain constant.

1 10

Target CFG

Guidance scale for the target condition. Increase for more significant edits.

1 30

Seed

Random seed for reproducibility.

0 10000

3. Output Image

Edited Result

After generating, you can use the recon.py script to create a 3D model. See the "How to Use" tab for the full command.

Click an Example to Load

Example Edits

Original Multi-View Image (src_mv.png)	Source Condition (src.png)	Target Condition (edited.png)	T_steps	n_max	Source CFG	Target CFG

Pages: