Guy Levy retweeted
Multi-turn image editing is notoriously challenging: as edits accumulate, unintended changes and artifacts build up.
Not anymore!😁
Introducing ACEdit ♠️
A training approach that counteracts accumulated drift across successive image edits, preserving scene details as the changes add up.
🌐 saradorfman1.github.io/ACEdi…
1/5
Guy Levy retweeted
When generating an image with an MM-DiT, the text tokens absorb information from the evolving image through multimodal attention. But what do these contextual tokens actually capture?
In our new work, “Learning to Read,” we show that contextual tokens can be read as plain text, revealing surprisingly detailed visual semantics. Building on this finding, we introduce a new way to train diffusion models: supervising their contextual tokens to improve generation quality and distributional coverage.
1/n
Guy Levy retweeted
This is a step toward physically grounded image editing!!
Glad to share our work, IntrinSync!
We jointly decomposes an image into editable intrinsic maps and renders them back to RGB, with cycle consistency to keep the maps coherent and support physically grounded image editing in a unified framework.
arxiv.org/abs/2610.11138
Guy Levy 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
Guy Levy retweeted
You ask for a talking parrot, but the model makes the pirate talk AND gives him a beak! 🦜😮💨
-> Introducing The Attention Triangle in Audio-Video Models: how semantics leak between text, audio, and video, and how we steer them back. 🧵 [1/5]
Guy Levy retweeted
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
Guy Levy retweeted
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!
👇
Guy Levy retweeted
High-frequency RoPE quietly hijacks attention. But RoPE is a spectrum: different frequencies play different roles and can be modulated.
We “untwist” it with frequency control.
Shared attention is a powerful way to transfer style in diffusion models: let tokens attend to a reference image, and the model can pick up stylistic cues.
⚠️ But in RoPE-based DiTs, this often breaks badly.
Instead of transferring style, the model starts copying the reference content.
In this work, Untwisting RoPE, we explore how positional bias and semantic understanding become twisted together in DiTs, and how frequency control can untwist them.
Had an immense pleasure collaborating with @Gatinbar and the team on this paper! A simple, holistic and effective approach. What’s better than that?
Check out our results 👀
inbar-2344.github.io/ShapeUp…
