Tim Dockhorn

I am a research scientist and cofounder at Black Forest Labs. I obtained my PhD and Master from the University of Waterloo, and a Bachelor of Science from the Technical University of Munich. My main research interest is generative modeling, with a focus on diffusion models.

Email  /  Google Scholar  /  Twitter  /  Github

profile photo
Research
clean-usnob Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation
Axel Sauer, Frederic Boessel, Tim Dockhorn, Andreas Blattmann, Patrick Esser, Robin Rombach
SIGGRAPH Asia, 2024
arXiv

We present a novel distillation method called Latent Adversarial Diffusion Distillation (LADD) and distill Stable Diffusin 3 down to one to four inference steps.

clean-usnob Scaling Rectified Flow Transformers for High-Resolution Image Synthesis
Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yannik Marek, Robin Rombach
ICML, 2024   (Best Paper Award)
arXiv

We present Stable Diffusion 3 an improved version of Stable Diffusion using a novel multimodal Diffusion Transformer architecture.

Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets
Andreas Blattmann*, Tim Dockhorn*, Sumith Kulal*,Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram Voleti & Adam Letts & Varun Jampani, Robin Rombach
arXiv, 2023
code / weights (svd) / weights (svd-xt)

We present Stable Video Diffusion,a latent video diffusion model for high-resolution, state-of-the-art text-to-video and image-to-video generation. We demonstrate the necessity of a well-curated pretraining dataset for generating high-quality videos and present a systematic curation process to train a strong base model, including captioning and filtering strategies.

clean-usnob SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis
Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, Robin Rombach
ICLR, 2024   (Spotlight Presentation)
arXiv / code / weights / refiner weights

We present SDXL an improved version of Stable Diffusion using several conditioning tricks and multi-resolution training. We also release a refiner model to even further improve visual fidelity.

clean-usnob Differentially Private Diffusion Models
Tim Dockhorn, Tianshi Cao, Arash Vahdat, Karsten Kreis
TMLR (2835-8856), 2023
arXiv / project page / code / twitter

We train diffusion models with strict differential privacy guarantees and outperform previous methods by large margins.

Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models
Andreas Blattmann*, Robin Rombach*, Huan Ling*, Tim Dockhorn*, Seung Wook Kim, Sanja Fidler, Karsten Kreis
CVPR, 2023
arXiv / project page / twitter

We take Stable Diffusion, insert additional temporal layers and fine-tune them on video data while keeping the spatial layers fixed.

Latent Space Diffusion Models of Cryo-EM Structures
Karsten Kreis*, Tim Dockhorn*, Zihao Li, Ellen Zhong
Machine Learning for Structural Biology Workshop, NeurIPS, 2022   (Oral Presentation)
arXiv / twitter

We train diffusion models over molecular confirmations from cryo-EM imaging data.

GENIE: Higher-Order Denoising Diffusion Solvers
Tim Dockhorn, Arash Vahdat, Karsten Kreis
NeurIPS, 2022
arXiv / project page / video / code / twitter

GENIE distills higher-order score terms into a small neural network and uses them for accelerated diffusion model sampling.

Score-Based Generative Modeling with Critically-Damped Langevin Diffusion
Tim Dockhorn, Arash Vahdat, Karsten Kreis
ICLR, 2022   (Spotlight Presentation)
arXiv / project page / video / code / twitter

We propose a novel diffusion using auxiliary velocity variables for more efficient denoising and higher quality generative models.

Demystifying and Generalizing BinaryConnect
Tim Dockhorn, Yaoliang Yu, Eyyüb Sari, Mahdi Zolnouri, Vahid Partovi Nia
NeurIPS, 2021
arXiv / video / twitter

We show that proximal maps can serve as a natural family of quantizers that is both easy to design and analyze, and we propose ProxConnect as a generalization of the widely-used BinaryConnect.

Density Deconvolution with Normalizing Flows
Tim Dockhorn*, James Ritchie*, Yaoliang Yu, Iain Murray
INNF+: Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models, ICML, 2020
arXiv / video / code / twitter

We denoise distributions using Normalzing Flows in conjuction with amortized variational inference.

clean-usnob Generative Modeling with Neural Ordinary Differential Equations
Tim Dockhorn
Master's Thesis, 2017
thesis

I give a comprehensive and self-contained introduction to continuous Normalizing Flows, and show that their training can be accelerated using tolerance schedulers.

clean-usnob A Discussion on Solving Partial Differential Equations using Neural Networks
Tim Dockhorn
arXiv, 2019
arXiv / code / twitter

I show that small neural networks (less than 500 parameters) can learn solutions of complex PDEs when optimized with BFGS.

clean-usnob Turbulence modeling for large eddy simulation using high-order discontinuous Galerkin methods
Tim Dockhorn
Bacherlor's Thesis, 2017
thesis

I test a variety of tubulence models within a high-order discontinuous Galerkin method.

Invited Talks
Accelerated Sampling and Improved Synthesis in Diffusion Models
IWR Colloquium, 11.01.2023
slides

Hosted by Fred Hamprecht.
Differentially Private Diffusion Models
Google Tech Talks: Differential Privacy for ML, 12.04.2023
slides

Hosted by Thomas Steinke.

Source code credit to Jon Barron