You need to agree to share your Hugging Face information to access this model

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

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

ReDiverse: Restoring Diversity in Distilled Image and Video Generators by Probing Activations

Furkan Horoz1, Ozgur Kara1, Kaan Oktay2, Adil Kaan Akan2, Atakan Kara2, Anand Bhattad3,†, James Matthew Rehg1,†
1University of Illinois at Urbana-Champaign    2fal    3Johns Hopkins University
†Equal advising.

Project page arXiv 2611.18550 Code on GitHub All ReDiverse models

ReDiverse-FT adapter for Wan 2.1. Few-step distilled students keep their teacher's activation variance across seeds but concentrate it into far fewer directions at one block of the first denoising step, so different seeds give near-identical samples. This rank-32 LoRA on the student's blocks up to blocks.28 is trained to match the teacher's activations there and is active in the first sampling step only; the remaining steps run the unmodified student.

Teacher 4-step student This adapter
Wan2.1-T2V-14B lightx2v, distilled from the teacher ReDiverse-FT LoRA of the student, rediverse_ft.safetensors (445 MB)

Wan 2.1, a dog running happily: four seed-matched clips of the student and of ReDiverse

Usage

git clone https://github.com/H-furkan/rediverse && cd rediverse   # set up the environments of envs/README.md
hf download ReDiverse/ReDiverse-Wan2.1 --local-dir ReDiverse-Wan2.1
python -m wan.generate --arm rediverse --adapter ReDiverse-Wan2.1 --prompt "a corgi surfing a wave at sunset" --num-samples 4   # video env

--arm rediverse_ft uses the adapter alone; --arm rediverse (above) also adds the training-free ReDiverse-TF directions, which first runs the teacher on 25 noises for each new prompt (about 15 min on an 80 GB H100) and caches the result. Without --adapter, the code downloads the same file from ReDiverse/ReDiverse.

DINO diversity ↑ (VBench) Student ReDiverse-FT (this adapter) ReDiverse (FT + TF) Teacher (multi-step)
Wan 2.1 31.6 46.8 52.2 51.6

License

Apache License 2.0, like Wan 2.1, the lightx2v student and the Perflow LoRA used for the training targets; NOTICE credits them.

Citation

@article{horoz2026rediverse,
  title   = {Restoring Diversity in Distilled Image and Video Generators by Probing Activations},
  author  = {Horoz, Furkan and Kara, Ozgur and Oktay, Kaan and Akan, Adil Kaan and Kara, Atakan
             and Bhattad, Anand and Rehg, James Matthew},
  journal = {arXiv preprint arXiv:2611.18550},
  year    = {2026}
}
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for ReDiverse/ReDiverse-Wan2.1

Collection including ReDiverse/ReDiverse-Wan2.1

Paper for ReDiverse/ReDiverse-Wan2.1