You need to agree to share your contact information to access this model

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

Summary

This repo contains merged model weights created by applying task arithmetic to three upstream checkpoints:

[ \text{Merged} = \text{ATI} + \lambda \cdot (\text{AniSora} - \text{Wan2.1-Base}) ]

The merge used (\lambda = 1.0) in the source script.

Files in this folder

Diffusion weights (sharded): diffusion_pytorch_model-00001-of-00007.safetensors … diffusion_pytorch_model-00007-of-00007.safetensors
Index: diffusion_pytorch_model.safetensors.index.json
Model config: config.json
Aux weights / tokenizers:
- Wan2.1_VAE.pth
- models_t5_umt5-xxl-enc-bf16.pth
- models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth
- tokenizers under google/umt5-xxl/ and xlm-roberta-large/

Upstream sources (reference)

Wan2.1 base: Wan-AI/Wan2.1-I2V-14B-480P
ATI: bytedance-research/ATI
AniSora V3.1: IndexTeam/Index-anisora (folder V3.1)

How the merge was produced (reference)

In this workspace, the merge code lives at:

Index-anisora/anisoraV3/Merge-ATI-AniSora/merge_models.py
It merges per key using safetensors and outputs to this Merged-Model/ directory.

Important behavior: the script only applies arithmetic when a key exists in both Base and AniSora; otherwise it keeps the ATI tensor for that key.

Downloads last month: -

Safetensors

Model size

16B params

Tensor type

F32

Model tree for KuOnoda/merged-ati-anisora

Base model

IndexTeam/Index-anisora

Finetuned

(4)

this model