🎉 Excited to introduce TRON, a relighting framework for 3D captures. 💡TRON pairs a neural renderer with 3D Gaussian reconstructions, achieving realistic quality, with 3D, material, & lighting control at interactive frame rates. arxiv.org/abs/2606.11314 research.nvidia.com/labs/sil…

Reference-conditioned diffusion models rely on dense reference token grids. Are they really necessary? Surprisingly, dropping 80–90% of the reference tokens barely affects quality. A little fine-tuning is enough to recover nearly the same quality at a fraction of the cost.

🎬🎨 LTX-2.3-Sync-LoRA, dataset and weights are out! check it out yourself :) sagipolaczek.github.io/Sync-…

🌟🚀 Excited to share our latest work: "Keep The Essentials: Efficient Reference Conditioned Generation via Token Dropping"! TL;DR: Stop wasting compute on redundant tokens! We introduce SparseContext that drops reference tokens for speeding up reference-based image generation⚡

Generating images with AI is fun 🤓 Finding the image? Not so much… The first result is rarely the one. Or the second. Or the third. 🥵 So we built a way to generate diverse possibilities along meaningful semantic axes 🧭 and organize them into a gallery you can effortlessly browse 🧚🏻 Introducing our #ECCV2026 paper: Semantic Browsing: Controllable Diversity for Image Generation ✨ 1/5

We find a shortcut in training reference-conditioned audio flow matching: at low noise levels, the model assigns speakers based on acoustic similarity rather than text. Biasing timestep sampling toward higher noise removes this shortcut and restores text-driven speaker assignment

Conditional diffusion/flow models often produce outputs inconsistent with the very signal conditioning them. The error is easily measurable, yet models are never trained to act on it. In FlowBender (now on arXiv), we train the model to correct its own errors. 🧵

1/7 New paper 🧵 ScenA generates a multi-speaker audio scene: overlapping speech, laughter, real room noise, from a text description and a few reference voices. When trained in the obvious way, it ignores the text and decides who speaks on its own. We found out why and fixed it.

Technical Workshop: 'AI for Creative Visual Content Generation, Editing and Understanding' at SIGGRAPH 2026 marks the 10th installment of the CVEU workshop series at the intersection of generative AI and visual arts. Building on prior editions across SIGGRAPH, CVPR, ICCV, and ECCV, the workshop explores how generative models continue to evolve for creative workflows. This year focuses on a next-generation creator–machine co-learning paradigm, where AI becomes a collaborative partner that bridges artistic intent and physical realism. s2026.conference-schedule.or… photo by David Weng © 2025 ACM SIGGRAPH

Thrilled to share that our paper "SAEdit: Token-level control for continuous image editing via Sparse AutoEncoder" has been accepted to #ECCV2026! 🥳 We use sparse autoencoders to find fine-grained editing directions, enabling controlled & precise image edits.

FUN to share that Sync-LoRA has been accepted to #ECCV2026! 🤗✨🎬 the field moves fast, but the core seems to remain: 1️⃣ IC-LoRA enables diverse edits 2️⃣ high-quality data is key (even if it's a small amount!) Huge thanks to @arash_mham @OPatashnik @DanielCohenOr1

ישראלים בחו״ל — בסוף אוקטובר יתקיימו בישראל עשרות כנסים אקדמיים. המועד המדויק (ככה"נ 19 או 26 באוק') ייקבע בקרוב לרשימת עשרות הכנסים שמתארגנים: mind-il.org/

🎉 Happy to share our latest work: Bootstrap Your Generator: Unpaired Visual Editing with Flow Matching (accepted to #ICML2026)! TL;DR: We train image and video editing models without any paired data. No ground-truth edit data, no external reward models.

Excited to present two papers at #CVPR2026 this week! 🚀 📸 Image Generation from Contextually-Contradictory Prompts We introduce a training-free method that leverages the coarse-to-fine nature of diffusion models to better handle conflicting concepts in prompts. 🕦 Sat, Jun 6 | 11:45 AM – 1:45 PM 📍 ExHall F, Poster #60 🔗 Project Page: tdpc2025.github.io/SAP/ 🎥 SemanticMoments: Training-Free Motion Similarity via Third Moment Features We introduce a training-free method for motion similarity in videos, leveraging higher-order temporal statistics to better capture semantic motion dynamics. 🕢 Sun, Jun 7 | 7:30 AM – 9:00 AM 📍 ExHall A, Poster #178 📄 Paper: arxiv.org/abs/2602.09146 Come by and say hi! 👋 @OPatashnik @DanielCohenOr1 @MokadyRon @BenaimSagie @kfir99 @OmerDahary

Excited to be giving a talk today at 12 PM at the Personalization Workshop ✨

Join us on June 4, 8:30AM–12:30PM, Room 4CD, for the 2nd Personalization in Generative AI Workshop (P13N) at @CVPR! #CVPR2026 #P13N Talks, posters & a panel on the future of generative AI systems that adapt to users' preferences, identities & contexts! p13n-workshop.github.io

I'll be presenting some of our recent works at the AI for Visual Arts (Room 4AB, 9am), Spatial Intelligence for Cultural Heritage (Room 708, 2:15pm) and the What is Next in Multimodal Foundation Models? (Room 3A, 4pm) workshops at @CVPR tomorrow. Please join if you're around!

Join us on June 4, 8:30AM–12:30PM, Room 4CD, for the 2nd Personalization in Generative AI Workshop (P13N) at @CVPR! #CVPR2026 #P13N Talks, posters & a panel on the future of generative AI systems that adapt to users' preferences, identities & contexts!

The code for Prox-E is now available! - github.com/etaisella/Prox-E Check it out and let us know if you find any issues