Sam Motamed

Research

I'm broadly interested in video generation and video-language models, with a current focus on improving their generation and understanding of physically plausible scenes. I also work on enabling user-intuitive control over generative models and adapting large vision and language models to solve personalized tasks using limited data. Relevant work is highlighted here.

Research Projects

• Physics Plausibility in Video Models: Improving how video generation and video-language models understand and produce physically plausible content.
• Generative Vision for Video Synthesis: Developing user-intuitive controls for generative video models to enable interactive and personalized content creation.
• Adaptation of Large Models: Adapting large-scale vision and language models to solve personalized tasks efficiently using limited data.

Publications

TRAVL: A Recipe for Making Video-Language Models Better Judges of Physics Implausibility

Saman Motamed, Minghao Chen, Luc Van Gool, Iro Laina

arXiv, 2025

code  /  arxiv

A recipe to make video language models better understand physics, and a rigorous benchmark to test VLMs on physics understanding.

Do generative video models learn physical principles from watching videos?

Saman Motamed, Laura Culp, Kevin Swersky, Priyank Jaini, Robert Geirhos

WACV, 2026

code  /  arxiv

A benchmark of real videos for testing physics understanding of generative video models.

Lego: Learning to Disentangle and Invert Concepts Beyond Object Appearance in Text-to-Image Diffusion Models

Saman Motamed, Danda Pani Paudel, Luc Van Gool

ECCV, 2024

code  /  arxiv

A method for textual inversion of adjectives and verbs in text-to-image diffusion models.

Investigating the Effectiveness of Cross-Attention to Unlock Zero-Shot Editing of Text-to-Video Diffusion Models

Saman Motamed, Wouter Van Gansbeke, Luc Van Gool

CVPR Generative Models Workshop, 2024

code  /  arxiv

Zero-shot control over object shape, position and movement in text-to-video models via cross-attention maps.

A Unified and Interpretable Emotion Representation and Expression Generation

Reni Paskaleva, Mykyta Holubakha, Andela Ilic, Saman Motamed, Luc Van Gool, Danda Paudel

CVPR, 2024

arxiv

Fine-grained generation of expressions in conjunction with other textual inputs and offers a new label space for emotions at the same time.

D3GU: Multi-Target Active Domain Adaptation via Enhancing Domain Alignment

Lin Zhang, Linghan Xu, Saman Motamed, Shayok Chakraborty, Fernando De la Torre

WACV, 2024

arxiv

A Multi-Target Active Domain Adaptation (MT-ADA) framework for image classification.

Personalized Face Inpainting With Diffusion Models by Parallel Visual Attention

Jianjin Xu, Saman Motamed, Praneetha Vaddamanu, Chen Henry Wu, Christian Haene, Jean-Charles Bazin, Fernando De la Torre

WACV, 2024

code  /  arxiv

Fast, identity preserving face inpainting with diffusion models.

PATMAT: Person Aware Tuning of Mask-Aware Transformer for Face Inpainting

Saman Motamed, Jianjin Xu, Chen Henry Wu, Fernando De la Torre

ICCV, 2023

ICCV  /  code  /  arxiv

A tuning method for personalizing inpainting of the face and preserving the identity of a subject.

Generative Visual Prompt: Unifying Distributional Control of Pre-Trained Generative Models

Chen Henry Wu, Saman Motamed, Shaunak Srivastava, Fernando De La Torre

NeurIPS, 2022

NeurIPS  /  code  /  arxiv

A framework for defining control over latent-based generative models.

Talks

Invited talks and presentations.