Pinned Tweet
🎬 Excited to share our paper: Joints, Camera, Action! (JCA)
JCA generates video & 3D motion together for any rig!
Video models generate motion you can watch. But what if you could use that motion to animate a 3D asset, with motion that responds to its surroundings?
research.nvidia.com/labs/par…
🧵
Super exciting to see Time-to-Move extended to MiniMax H3! The results look awesome! 🚀
Minimax h3 😲 Motion Time to move
implementation of time-to-move dual clock denoising for Minimax H3, allowing for motion control by dragging objects into motion, or mask part of a video
and change what happens.
github:
github.com/PxTicks/h3-motion
workflow:
drive.google.com/file/d/1mOH…
Meet ReLift, our latest work on 3D tracking! 🏋️
Rethinking 3D tracking with the right combination:
Strong 2D trackers
＋ 3D geometry
＋ Efficient trajectory corrector
──────────────
Dense 3D tracking,
even through occlusions!
🚀 Introducing ReLift: dense 3D tracking from monocular video.
Strong 2D trackers provide correspondences.
Geometry models produce depth & cameras.
We show how to combine them, repair 3D errors, and recover occluded motion for SOTA 3D tracking.
All it takes: one small corrector + off-the-shelf models.
🌐 Project: relift3d.github.io/
@MarounKhriesh @NoamRot @yoni_kasten @orlitany
Noam Rotstein retweeted
Coding agents can build an app. What if they could give it a body?
Meet Metatron: a versatile System 1 controller for humanoids. A goal-based interface connects System 2 planners, including coding agents, to whole-body movement.
A few examples below. 1/5
Noam Rotstein retweeted
1/4 We trained an 8B safety classifier on just 1,275 carefully annotated prompts.
It beat GPT-5.4, Claude Sonnet 4.6, and leading open-weight guardrails trained on far more data, on average across 5 safety benchmarks.
Quality > quantity.
For more details, check out the paper!
Project page: research.nvidia.com/labs/par…
Paper: arxiv.org/abs/2610.28632
Big shoutout to my collaborators!
@YoadTewel @ChenTessler @HaggaiMaron @GalChechik @orlitany @yoni_kasten @AtzmonYuval at @nvidia Research
We build ~200K video–motion pairs across diverse rigs.
A major part is “Objaverse in Video”: a video editor places existing animated asset renders in suitable scenes, creating data at scale for arbitrary rigs. We add human and robot data.
A graph encoder maps the supplied rig to joint features for a transformer motion denoiser.
Bidirectional cross-attention links it to a pretrained video model. Bridge guidance amplifies the prediction difference with these links on vs. off to improve video–motion agreement.
And JCA works in both directions! ↔️
🎥 Video → 3D motion: Given a video and a rig, recover the subject's motion.
🦴 3D motion → Video: Given a rig and its motion, generate a matching video, optionally conditioned on a reference image.
All within the same model.
Want more control over the animation?
Give JCA a few chosen image keyframes, along with the rig and a text prompt.
It generates the motion between them, producing both a video and a synchronized 3D trajectory.
What if a 3D character could move in response to its surroundings?
With a scene image, JCA generates motion that fits the scene. Jumping over a bar, sliding down a slope, or reaching for an object.
We recover a camera to place and replay the generated 3D motion in that scene.
Noam Rotstein retweeted
Do vision-language models look at the image regions they describe? In our new paper, we present Self-Saliency, a method for aligning visual attention with reasoning text.
research.nvidia.com/labs/par… w/ @GalChechik @DalalGal
Noam Rotstein retweeted
Excited to share our work, “Token-to-Token Alignment of Text Embeddings for Semantic Blending,” accepted as a NeurIPS 2026 Spotlight 🌟
Can we directly interpolate between semantic concepts?
Text embedding spaces already encode a continuous semantic structure, the missing piece is alignment.
Once aligned, simple interpolation becomes meaningful, naturally enabling multiple applications.
Joint work with @MokadyRon @OPatashnik @DanielCohenOr1
Noam Rotstein retweeted
Meet DyRAD: Radar NVS for dynamic driving scenes.
📡🚗 Radar directly senses motion through Doppler, but prior radar NVS methods ignore this signal where it matters most: dynamic scenes.
🌐 dyrad-nvs.github.io
📄 arxiv.org/abs/2609.39841
💻 github.com/Dyrad-NVS/DyRAD
🧵
To Pareto and beyond! 🚀
Excited to share that our paper "ParetoSlider: Diffusion Models Post-Training for Continuous Reward Control" has been accepted to @NeurIPSConf as an oral! 🎉
ParetoSlider enables continuous inference-time control over competing reward trade-offs with a single diffusion model, without retraining or maintaining multiple checkpoints.
Huge thanks to the great team @mikikiFin, @ArielBAMath, @YotamNitzan, @OPatashnik
Paper: arxiv.org/abs/2604.20816
Project page: shelley-golan.github.io/Pare…
Code: github.com/Shelley-Golan/Par…
Noam Rotstein retweeted
1/11 Can a stronger coding model rescue a run started by a cheaper one?
Can a cheaper model continue a stronger model's work - retaining quality at lower cost?
The two directions behave very differently...
Check out our new paper: The Handoff Tax 🧵
📄 arxiv.org/abs/2608.24358
Noam Rotstein retweeted
🥳 Excited to share UniProbe: A learnable, token-level hallucination detection for large VLMs 🔎
UniProbe looks inside a VLM’s internals to identify hallucinated response tokens and mitigate them before they reach the user.
📉 55% fewer hallucinations
⚡ Just 1.06× latency
Noam Rotstein retweeted
I'm going to present Fast-AR at #ICML26 tomorrow (Tuesday) @ 2pm - poster #1204!
Come and say hi 👋
If you're around this week, feel free to DM me.
Project page: dvirsamuel.github.io/fast-au…
Details below ⬇️
