About me
I’m a postdoc at Kyutai and UC Berkeley, working with Patrick Pérez and Alyosha Efros. I’m currently exploring the representation capacities of generative models.
I did my PhD at the IMAGINE (ENPC) and VISTA (École Polytechnique) labs, under the supervision of David Picard and Vicky Kalogeiton. My thesis focused on efficient conditional generative models, and I worked on GANs and diffusion models.
Research themes
Training with less
Image generation is a great place to make progress, but other fields rarely have unlimited data. I look for methods that need less data and compute, so that this progress carries over to those fields and stays within reach of academic labs.
Generative geolocation
I like how much one photo can reveal about where it was taken: the plants, the road markings, the light. Reading these clues takes broad knowledge of places and cultures, and since they are often ambiguous, generative models are a natural fit.
How generative models work
Beyond any single application, I'm curious about generative models themselves: why they work so well, and what they really learn about the data.
News
2026
-
We released Rethinking Neural Thickets, showing that a random bias on the output logits matches perturbing every weight.
-
We released CADRE, open video autoencoders for interactive world models.
-
Our work Diffusable Latents from Structure-Agnostic Distillation got accepted at the NeurIPS 2026 Workshop on Principles of Generative Modeling!
-
We released The FID Lottery, measuring how much FID varies across training and sampling seeds.
-
We released Surflo, a feed-forward model that reconstructs 3D surfaces from unposed views at any resolution.
-
I will be visiting UC Berkeley this summer to work with Alyosha Efros.
-
Started working at Kyutai as a PostDoc!
2025
-
Defended my PhD! Recording here. Huge thanks to my jury — Alexei A. Efros, Patrick Pérez, Diane Larlus, Giorgos Tolias, and Tero Karras — for their time and feedback.
-
Got an outstanding reviewer award at NeurIPS 2025 (Top 8%).
-
Got invited to the mainstream French tech podcast Underscore_ to talk about my work on Around the World in 80 Timesteps.
-
Attended CVPR 2025 to present our work Around the World in 80 Timesteps.
-
Our work Around the World in 80 Timesteps got accepted at CVPR 2025.
2024
-
Attended the ICVSS 2024 Summer School!
2023
-
Organized the first edition of the IMAGINE Hackathon.
2022
-
Pleased to get an outstanding reviewer award at ACCV 2022.
-
Started my internship at Meta London, in the Pytorch team.
2021
Publications
Recent Preprints
2026
Diffusable Latents from Structure-Agnostic Distillation
NeurIPS 2026 Workshop on Principles of Generative Modeling
2025
2024
E.T. the Exceptional Trajectories: Text-to-Camera-Trajectory Generation with Character Awareness.
ECCV 2024
2023
2022
Teaching
INF473V
Deep Learning in Computer Vision
Apr – Jun 2024
- Helped supervise the practical sessions of the course.
- Created a Kaggle competition for the students end of the course project. The goal was to create a classifier but we only provided them with the val and test data. They needed to create their own training data with pretrained diffusion models. The goal was to classify among 37 different cheeses.
INF473V
Deep Learning in Computer Vision
Feb – Jun 2023
- Helped supervise the practical sessions of the course.
- Built a new practical session on Transformers.
- Created a Kaggle competition for the students end of the course project. The goal was to classify synthetic images in a weakly supervised setting.
INF573
Image Analysis and Computer Vision
Nov 2022
- Helped supervise students projects.
Open Source
Contributions
pytorch
rl
TorchRL, the PyTorch library for reinforcement learning
Lightning-AI
metrics
TorchMetrics, machine learning metrics for PyTorch
huggingface
diffusers
Hugging Face library of pretrained diffusion models
Projects
kyutai-labs
structure-agnostic-distillation
Diffusable Latents from Structure-Agnostic Distillation
NeurIPS 2026 Workshop on Principles of Generative Modeling
Anttwo
Surflo
Surflo: Consistent 3D Surface Flow Model with Global State
NeurIPS 2026
kyutai-labs
ovie
One View Is Enough! Monocular Training for In-the-Wild Novel View Generation
NeurIPS 2026
kyutai-labs
logit-psv
Rethinking Neural Thickets Without Weight Perturbations
Arxiv Preprint
kyutai-labs
cadre-autoencoders-and-world-models
Rethinking Video Autoencoders for Interactive World Models
Arxiv Preprint
nicolas-dufour
miro
MIRO: MultI-Reward cOnditioned pretraining improves T2I quality and efficiency
ICML 2026
davidpicard
pom
PoM: A Linear-Time Replacement for Attention with the Polynomial Mixer
CVPR Findings 2026
eleftsonis
DIPSY
DIPSY: Training-Free Synthetic Data Generation with Dual IP-Adapter Guidance
BMVC 2025
nicolas-dufour
plonk
Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation
CVPR 2025
lucasdegeorge
T2I-ImageNet
How far can we go with ImageNet for Text-to-Image generation?
Arxiv Preprint
robincourant
DIRECTOR
E.T. the Exceptional Trajectories: Text-to-Camera-Trajectory Generation with Character Awareness.
ECCV 2024
nicolas-dufour
CAD
Don't drop your samples! Coherence-aware training benefits Conditional diffusion
CVPR 2024
gastruc
osv5m
OpenStreetView-5M: The Many Roads to Global Visual Geolocation
CVPR 2024
nicolas-dufour
SCAM
SCAM! Transferring humans between images with Semantic Cross Attention Modulation
ECCV 2022
Miscellaneous
Talks
2026
-
Oct 16
Talk at BAIR in Angjoo Kanazawa's group
-
Oct 9
Talk at the GdR IASIS day on diffusion and flow matchingThe FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation
-
Sep 8
Oral at the E.T. workshop at ECCV 2026The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation
-
Jul
Talk at Adobe Research
2025
-
Dec 18
Talk at MIT in Philip Isola's groupControllable and Efficient Generative Models
-
Dec 17
Talk at Cornell Tech in Andrew Owens's groupControllable and Efficient Generative Models
-
Dec 16
Talk at NYU in Saining Xie's groupControllable and Efficient Generative Models
-
Dec 10
Talk at BAIR in Alyosha Efros's groupControllable and Efficient Generative Models
-
Oct 30
Talk at ValeoAIControllable and Efficient Generative Models
-
Oct 23
Talk at KyutaiControllable and Efficient Generative Models
-
Jun 25
Podcast Underscore_Around the World in 80 Timesteps
2024
-
Sep 19
Talk at TUM in Zeynep Akata's groupConditional Generative Models
2023
-
Apr 12
IMAGINE SeminarLLM advances in 2023
-
Apr 12
IMAGINE SeminarRecent advances in diffusion models
2022
-
Apr 20
IMAGINE SeminarPresentation of the Perceiver papers
-
Feb 10
GeoVic SeminarSCAM! Transferring humans between images with Semantic Cross Attention Modulation
-
Jan 5
IMAGINE SeminarTutorial on GANs
2021
-
Oct 6
IMAGINE SeminarSCAM! Transferring humans between images with Semantic Cross Attention Modulation
Reviewing
Awards
-
Gold reviewer award at ICML 2026Jul 2026
-
Outstanding reviewer award at CVPR 2026Jun 2026
-
Outstanding reviewer award at NeurIPS 2025 (top 8%)Dec 2025
- Dec 2022
Others
-
I have co-organized the first edition of IMAGINE Hackaton. The topic was geolocationJul 2023
-
I'm co-organizing the IMAGINE reading group for 2023-2024Jun 2023