About

I'm a researcher in generative modeling with a focus on visual media, working on large-scale image, video and multimodal generation. At Black Forest Labs, I contributed to FLUX.3, BFL's multimodal foundation model for video, image and audio. My work spans generative frameworks, training and post-training, guidance and sampling, as well as large-scale data pipelines.

I completed my PhD (CIFRE) at Meta AI (FAIR) and Inria Grenoble (THOTH) / Université Grenoble Alpes (UGA) .
I defended my thesis, On the Efficiency and Controllability of Visual Generative Models, on October 6, 2026, advised by Karteek Alahari (Inria), Jakob Verbeek and Camille Couprie (Meta AI). I specialize in diffusion- and flow-based generative models for large-scale image and video generation.

Before my PhD, I completed the MVA master’s program in applied mathematics and machine learning, and an M.Sc.-level engineering degree (Diplôme d’ingénieur) with a major in Data and Information Sciences – Mathematics and Data Science (SDI–MDS) at CentraleSupélec .

My research focuses on developing training and sampling algorithms to obtain more efficient, flexible and controllable generative models, with applications on image, video and multimodal generation.

Experience

  • Member of Technical Staff · Black Forest Labs April 2026 – present · Paris, France & Freiburg, Germany

    Post-training of large-scale image and video generative models. Contributions to FLUX.3 across multimodal audio-video generation, with text, image, video, and audio conditioning. Work on guidance, gradient-based reward tilting, sampling optimization and data pipelines.

  • PhD Student (CIFRE) · Meta AI (FAIR) & Inria Grenoble / UGA March 2023 – October 2026 · Paris, France

    Developed training and sampling algorithms for diffusion and flow-based generative models, including Flowception (CVPR 2026). Built video training codebases and large-scale dataset filtering pipelines, contributed to EvalGIM and collaborated on OneFlow.

  • Research Engineer · Inria Grenoble October 2022 – March 2023 · Grenoble, France

    Improved instance segmentation models in collaboration with Meta AI.

  • Research Scientist Intern · Meta AI (FAIR) May 2022 – September 2022 · Paris, France

    Generative modeling for weakly supervised instance segmentation and image-to-image translation for watershed-based segmentation pipelines.

  • Machine Learning Research Intern · ScienceAtHome & Center for Hybrid Intelligence, Aarhus University March 2021 – September 2021 · Aarhus, Denmark

    Explored GAN latent spaces for image generation and embedding search, developed latent-space visualization methods and implemented expert opinion aggregation for weather prediction.

  • Deep Learning R&D Intern · Kinetix August 2020 – February 2021 · Paris, France

    Developed 3D human pose estimation models from monocular video, managed training and AWS deployment pipelines, and designed motion post-processing and representation-learning components.

Education

  • PhD (CIFRE), Applied Mathematics & Computer Science
    Meta AI (FAIR) & Inria Grenoble / Université Grenoble Alpes 2023 – 2026 · Defended October 6, 2026

    Thesis: On the Efficiency and Controllability of Visual Generative Models.

  • M.Sc. – Mathematics, Vision, Learning (MVA)
    ENS Paris-Saclay 2021 – 2022
  • M.Sc.-level engineering degree (DiplĂ´me d’ingĂ©nieur)
    CentraleSupélec – major in Data and Information Sciences – Mathematics and Data Science (SDI–MDS) 2018 – 2022

Publications

  • Flowception: Temporally Expansive Flow Matching for Video Generation FIRST AUTHOR
    Tariq Berrada Ifriqi, John Nguyen, Karteek Alahari, Jakob Verbeek, Ricky T. Q. Chen IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2026 arxiv | site
  • OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows John Nguyen, Marton Havasi, Tariq Berrada, Luke Zettlemoyer, Ricky T. Q. Chen Preprint. 2025 arxiv | site
  • Entropy Rectifying Guidance for Diffusion and Flow Models FIRST AUTHOR
    Tariq Berrada, Adriana Romero-Soriano, Michal Drozdzal, Jakob Verbeek, Karteek Alahari Advances in Neural Information Processing Systems (NeurIPS). 2025 arxiv
  • Boosting Latent Diffusion with Perceptual Objectives FIRST AUTHOR
    Tariq Berrada, Pietro Astolfi, Melissa Hall, Marton Havasi, Yohann Benchetrit, Adriana Romero-Soriano, Karteek Alahari, Michal Drozdzal, Jakob Verbeek International Conference on Learning Representations (ICLR). 2025 arxiv
  • EvalGIM: A Library for Evaluating Generative Image Models Melissa Hall, Oscar Mañas, Reyhane Askari-Hemmat, Mark Ibrahim, Candace Ross, Pietro Astolfi, Tariq Berrada Ifriqi, Marton Havasi, Yohann Benchetrit, Karen Ullrich, Carolina Braga, Abhishek Charnalia, Maeve Ryan, Mike Rabbat, Michal Drozdzal, Jakob Verbeek, Adriana Romero-Soriano Preprint. 2025 arxiv | code
  • On Improved Conditioning Mechanisms and Pre-training Strategies for Diffusion Models FIRST AUTHOR
    Tariq Berrada Ifriqi, Pietro Astolfi, Melissa Hall, Reyhane Askari-Hemmat, Yohann Benchetrit, Marton Havasi, Matthew Muckley, Karteek Alahari, Adriana Romero-Soriano, Jakob Verbeek, Michal Drozdzal Advances in Neural Information Processing Systems (NeurIPS). 2024 arxiv
  • Unlocking Pre-trained Image Backbones for Semantic Image Synthesis FIRST AUTHOR
    Tariq Berrada, Jakob Verbeek, Camille Couprie, Karteek Alahari IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2024 arxiv
  • Guided Distillation for Semi-supervised Instance Segmentation FIRST AUTHOR
    Tariq Berrada, Camille Couprie, Karteek Alahari, Jakob Verbeek IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). 2024 arxiv | code