Instructions to use zhaors00/Stroke3D with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use zhaors00/Stroke3D with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("zhaors00/Stroke3D", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
Stroke3D
This repository contains the released checkpoints and data for Stroke3D: Lifting 2D Strokes into Rigged 3D Model via Latent Diffusion Models (ICLR 2026).
[OpenReview] [Code] [Project page]
Repository layout
Checkpoints
ckpt/Sk-VAE/: skeleton variational autoencoder checkpoint.ckpt/Sk-DiT/: skeleton latent diffusion transformer checkpoint.ckpt/SKDream-SKA-DPO/: final SKDream ControlNet fine-tuned with SKA-DPO.
Data
data/Skeleton-Data/: shared skeleton-coordinate training data for Sk-VAE and Sk-DiT; rendered PNG images are excluded.data/DPO-Data/: 2,000 SKA-DPO preference pairs.data/TextuRig/: 6,633 captioned skeleton/GLB pairs; rendered PNG images are excluded.
Detailed environments, commands, and data formats are documented in the
component READMEs in the Stroke3D code repository:
SkDiff/README.md,
SKDream_DPO/README.md, and
TextuRig/README.md.
Usage notes
The SKDream/SKA-DPO checkpoint is not a standalone Diffusers pipeline. Use
repository ID zhaors00/Stroke3D and subfolder ckpt/SKDream-SKA-DPO with the
Stroke3D code.
Download the DPO archive with:
hf download zhaors00/Stroke3D data/DPO-Data/stroke3d_ska_dpo_margin_0.10.tar --local-dir stroke3d-assets
Download all TextuRig metadata and shards with:
hf download zhaors00/Stroke3D --include "data/TextuRig/*" --local-dir stroke3d-assets
Licensing and provenance
No blanket license is asserted over upstream checkpoints or source-derived assets. Review the component cards and all applicable upstream terms before redistribution or commercial use. Camera metadata in the historical DPO archive uses Python pickle and must only be loaded from a trusted release.
Citation
@inproceedings{zhao2026stroke3d,
title = {Stroke3D: Lifting 2D Strokes into Rigged 3D Model via Latent Diffusion Models},
author = {Zhao, Ruisi and Zheng, Haoren and Yang, Zongxin and Fan, Hehe and Yang, Yi},
booktitle = {The Fourteenth International Conference on Learning Representations},
year = {2026},
url = {https://openreview.net/forum?id=VgOWxor3LV}
}
- Downloads last month
- -