moab.arar

FAIR (@AIatMeta) | Ph.D. | Generative World Models | ex Google

[Yael Vinker🎗 @YVinker]

Excited to share that Inspiration Seeds has received an Honorable Mention award this year at SIGGRAPH! 🎉

👉https://t.co/gE9MRvbKcV

Huge thanks and congrats to the best team! @kfir99 @EladRichardson

Looking forward to seeing you in LA in July!🌱

[moab.arar @ArarMoab]

@OmriAvr Sweet - looks good!

[moab.arar @ArarMoab]

@_amirbar @Meta Welcome back 😁😁

[Etai Sella @etai_sella]

Even today, with powerful image editing models, making fine-grained structural changes to 3D shapes remains a major challenge.

In our new #SIGGRAPH2026 paper, Prox-E, we use primitive-based abstraction to leverage VLMs for precise, reasoning-based 3D editing!

👇

[Gal Metzer @galmetz]

Excited to share our work accepted to #SIGGRAPH2026 ! Video generation models struggle with something few talk about: their transformations don't evolve smoothly. You get long boring stretches... then a sudden semantic jump where everything "catches up" at once.

[Shelly Golan @Shelly_Golan1]

1/7

When rewards conflict, what should RL post-training of diffusion models optimize?

In visual generation, objectives are often in tension:
Prompt adherence can conflict with source preservation.
Photorealism can conflict with stylization.

In our new paper, ParetoSlider, we introduce a multi-objective RL framework that trains a single diffusion model for continuous control over competing reward objectives 🧵

[Daniel Severo @_dsevero]

The FAIR Montréal office is expanding!

We're looking for folks to join us who are exciting about world modelling. Hiring for both recent grads as well as more experienced researchers.

Feel free to reach out, but please read the posting first :)

https://t.co/6yZUyej4VZ

[Inbar Gat @Gatinbar]

3D editing has long relied on workarounds: per-asset optimization, 2D view propagation, or hacking frozen priors. The bitter lesson is the same one image editing already learned. Train a native model, end-to-end.
Introducing ShapeUP, accepted to SIGGRAPH 2026 💫

[moab.arar @ArarMoab]

People are missing the fun of TPUs... JAX is just the way in.

[François Chollet @fchollet]

When looking at deep learning profiles, one of the most obvious tells between a mediocre and great candidate is whether they list PyTorch or JAX.

[AK @_akhaliq]

Pretrained Video Models as Differentiable Physics Simulators for Urban Wind Flows

paper: https://t.co/z3oKErhhD4

[Yael Vinker🎗 @YVinker]

I am *very* excited to announce our SIGGRAPH 2026 workshop:

Lines & Minds: Visual Abstraction in Art, Psychology, and Computer Graphics 🎨🧠🫖

🔗 https://t.co/dZaoDPv5Zt
📅 Sunday, July 19

Join us to explore how visual abstraction shapes how we think, create, and communicate.

[moab.arar @ArarMoab]

Video models as Physics simulators. 🌍🎥

[1/] In our latest work, WinDiNet, we finetuned a pre-trained video model into a differentiable physics engine. 1000x faster than traditional CFD solvers.

Project page: https://t.co/LAx7t00y3e
Abs: https://t.co/OdcgbKeQEG

[Sagi Polaczek 🦜 @PolaczekSagi]

surprising repurposing of video diffusion models
🌬️🏢

[moab.arar @ArarMoab]

[4/] Shout out to Janne Perini and Rafael Bischof for their hard work, and the great team: Ayça Duran, Michael A. Kraus, Siddhartha Mishra, and @bernd_bickel.

[moab.arar @ArarMoab]

Video models as Physics simulators. 🌍🎥

[1/] In our latest work, WinDiNet, we finetuned a pre-trained video model into a differentiable physics engine. 1000x faster than traditional CFD solvers.

Project page: https://t.co/LAx7t00y3e
Abs: https://t.co/OdcgbKeQEG

[moab.arar @ArarMoab]

[3/] We fine-tuned LTX with 10k 2D urban wind flow simulations by (i) fine-tuning the VAE decoder with physics-informed losses, and (ii) replacing the text encoder with a wind speed and direction embedder.

[Yael Vinker🎗 @YVinker]

Creative work often starts before we can describe what we're looking for. What role can generative models play at this stage?
🌱Our new work, Inspiration Seeds, reveals hidden visual connections between images, creating a purely visual exploration space.
🔗https://t.co/gE9MRvbKcV

[Andrey Voynov @kusichan]

Check out our most recent work on the training-free video editing from @GoogleDeepMind. One of the cool things about the approach is that it allows modifying the motions of particular scene elements while keeping the dynamics of other parts (check the 🎱 example!). The method is model-agnostic https://t.co/McaTNHmTT4

[moab.arar @ArarMoab]

@EthanHe_42 Congrats - Looks awesome!