Access the GR3EN model weights

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

GR3EN — License & Terms of Use

By requesting access to this repository you acknowledge and agree to the
following:

  1. License. The GR3EN fine-tuned weights are released under the
    Apache License 2.0. The base model (Wan2.2-TI2V-5B) and SAM 2 are
    redistributed under their respective licenses (Apache-2.0), which you
    also agree to comply with.
  2. Intended use. The model is intended for research and creative
    applications: relighting of videos and 3D scenes.
  3. Prohibited use. You will not use the model to produce content that
    is unlawful, deceptive (including undisclosed synthetic or edited
    media of real people or events), harassing, or that infringes the
    privacy or intellectual-property rights of others.
  4. Attribution. If you use the model in academic work, you agree to
    cite the GR3EN paper (SIGGRAPH 2026).
  5. No warranty. The model is provided "AS IS", without warranties or
    conditions of any kind, as stated in the Apache License 2.0.

Log in or Sign Up to review the conditions and access this model content.

GR3EN: Generative Relighting for 3D Environments

GR3EN (SIGGRAPH 2026) is a relighting model that finetunes Wan2.2-TI2V-5B for 3D-aware video generation with controllable lighting. Given input images and light control masks, GR3EN generates relit video sequences.

Repository contents

Path Description
gr3en_weights.pt GR3EN fine-tuned DiT + ControlNet weights
wan2.2-ti2v-5b/ Wan2.2-TI2V-5B base model, VAE and prompt embeddings
sam2/sam2.1_hiera_large.pt SAM 2 checkpoint (interactive demos only)

Usage

Download all files and arrange them under gr3en/inference/checkpoints/ as described in the GitHub README, then run the Studio demo or batch inference.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for xyxingx/GR3EN

Finetuned
(72)
this model

Space using xyxingx/GR3EN 1