|
Zongze Wu I'm a research scientist/engineer at Adobe Research in San Francisco. At Adobe, I work with the Firefly team on projects including Structure Reference and real-time generative AI features presented at MAX Sneak 2023, MAX Sneak 2024, and Adobe Summit Sneak 2025. I got my M.S. & Ph.D at Hebrew University of Jerusalem (2016–2022), under the supervision of Prof. Dani Lischinski and Eli Shechtman from Adobe Research. I got my bachelor degree at Tongji University (2011–2016). |
|
|
My research focuses on generative modeling, spanning diffusion models, GANs, and autoregressive models. I am particularly interested in training efficiency, simpler training frameworks, and efficient architecture design — recent work includes one-step generation via mean flows, representation alignment for diffusion, separable causal inference in video diffusion, and improved representation autoencoders. |
|
Improved Baselines with Representation Autoencoders
Jaskirat Singh, Boyang Zheng, Zongze Wu, Richard Zhang, Eli Shechtman, Saining Xie arXiv 2026 [project page] [arXiv] RAEv2 achieves state-of-the-art gFID of 1.06 on ImageNet-256 in just 80 training epochs, significantly improving training efficiency over the original RAE. |
|
End-to-End Training for Unified Tokenization and Latent Denoising
Shivam Duggal, Xingjian Bai, Zongze Wu, Richard Zhang, Eli Shechtman, Antonio Torralba, Phillip Isola arXiv 2026 [project page] [arXiv] A unified framework that jointly trains tokenization and latent denoising end-to-end, improving generation quality by removing the disconnect between the tokenizer and generative model. |
|
Causality in Video Diffusers is Separable from Denoising
Xingjian Bai, Guande He, Zhengqi Li, Eli Shechtman, Xun Huang, Zongze Wu arXiv 2026 [project page] [arXiv] We show that causal reasoning in video diffusion models is separable from denoising, introducing Separable Causal Diffusion (SCD) that significantly improves throughput and per-frame latency. |
|
What Matters for Representation Alignment: Global Information or Spatial Structure?
Jaskirat Singh, Xingjian Leng, Zongze Wu, Liang Zheng, Richard Zhang, Eli Shechtman, Saining Xie arXiv 2025 [project page] [arXiv] Spatial structure, rather than global semantic performance, drives the generation quality of representation alignment in diffusion models. |
|
Improved Mean Flows: On the Challenges of Fastforward Generative Models
Zhengyang Geng, Yiyang Lu, Zongze Wu, Eli Shechtman, J. Zico Kolter, Kaiming He arXiv 2025 [project page] [arXiv] Improved MeanFlow (iMF) achieves 1.72 FID with a single function evaluation on ImageNet 256x256, closing the gap with multi-step methods using no distillation. |
|
VLM-Guided Adaptive Negative Prompting for Creative Generation
Shelly Golan, Yotam Nitzan, Zongze Wu, Or Patashnik arXiv 2025 [project page] [arXiv] A training-free, inference-time method that uses VLMs to adaptively steer diffusion away from conventional concepts, encouraging novel and creative image generation. |
|
TurboFill: Adapting Few-step Text-to-image Model for Fast Image Inpainting
Liangbin Xie, Daniil Pakhomov, Zhonghao Wang, Zongze Wu, Ziyan Chen, Yuqian Zhou, Haitian Zheng, Zhifei Zhang, Zhe Lin, Jiantao Zhou, Chao Dong CVPR 2025 [project page] [arXiv] A fast image inpainting method that achieves high-quality results in just 4 diffusion steps, outperforming multi-step methods that require over 50 steps. |
|
SliderSpace: Decomposing the Visual Capabilities of Diffusion Models
Rohit Gandikota, Zongze Wu, Richard Zhang, David Bau, Eli Shechtman, Nicholas Kolkin ICCV 2025 [project page] [arXiv] Automatically decomposes a diffusion model's visual capabilities into controllable, interpretable directions using low-rank adaptors for concept decomposition and style exploration. |
|
TurboEdit: Instant text-based image editing
Zongze Wu, Nicholas Kolkin, Jonathan Brandt, Richard Zhang, Eli Shechtman ECCV 2024 [project page] [arXiv] [Video] Users can upload an image and edit it with natural language; each edit takes half a second. Featured at the 2024 Research Mid-Year All Hands, Q3 CTO All Hands 2024 (presented by CEO), and GenAI Summit II 2024. |
|
Lazy diffusion transformer for interactive image editing
Yotam Nitzan, Zongze Wu, Richard Zhang, Eli Shechtman, Daniel Cohen-Or, Taesung Park, Michaël Gharbi ECCV 2024 [project page] [arXiv] Instead of generating the entire image, we only generate the mask region to facilitate fast inpaint task. |
|
Third time's the charm? Image and video editing with StyleGAN3
Yuval Alaluf, Or Patashnik, Zongze Wu, Asif Zamir, Eli Shechtman, Dani Lischinski, Daniel Cohen-Or AIM ECCVW 2022 [project page] [arXiv] We show StyleGAN3 can be trained with unaligned images, and its w/w+ spaces are more entangled than StyleGAN2. |
|
StyleAlign: Analysis and applications of aligned StyleGAN models
Zongze Wu, Yotam Nitzan, Eli Shechtman, Dani Lischinski ICLR 2022 (Oral Presentation) [project page] [arXiv] The child model's latent spaces are semantically aligned with its parent's, inheriting incredibly rich semantics. |
|
StyleCLIP: Text-driven manipulation of StyleGAN imagery
Or Patashnik*, Zongze Wu*, Eli Shechtman, Daniel Cohen-Or, Dani Lischinski ICCV 2021 (Oral Presentation) [project page] [arXiv] [ICCV Video] [Demo Video] Text-based image editing through mapping CLIP space to StyleGAN latent space. |
|
StyleSpace analysis: Disentangled controls for StyleGAN image generation
Zongze Wu*, Dani Lischinski, Eli Shechtman CVPR 2021 (Oral Presentation) [project page] [arXiv] [Video] The space of channel-wise style parameters is significantly more disentangled than the other intermediate latent spaces in StyleGAN. |
|
Fine-grained foreground retrieval via teacher-student learning
Zongze Wu*, Dani Lischinski, Eli Shechtman WACV 2021 [arXiv] [Video] Retrieve foreground images that are semantically compatible with the background. |
|
|
|
