Tali Dekel

We raised a $7M pre-seed to make shopping fun again. @daletlabs generates media just for you: like walking into a mall built around your taste. And you can buy everything in-app. Built by researchers from Google, Runway and MIT. Backed by @lightspeedvp and amazing angels.

1/5 Excited to introduce IR4RL! Image-to-code models build images step by step. We show that visual progress provides a powerful learning signal enabling RL post-training from intermediate renders. With amazing @kate_feingold, @phillip_isola & @talidekel ir4rl.github.io/

Very excited to share my first paper! 🎉 FoldingAgent has been accepted to SIGGRAPH Asia 2026 ✨ 🔹 Reconstructs parametric origami from demonstration videos 🔹 Uses a VLM agent + origami simulator Project: maya-moriya.github.io/origam… Joint work with @sigal_raab @YVinker @talidekel

Excited to present DRoPS! A novel method for dynamic 3D reconstruction from monocular videos, accompanied by a static-prescan of the dynamic object. 🌐 Project: drops-dynamics.github.io 📄 Paper: arxiv.org/abs/2603.24770 🧵 [1/7]

Video editing just got more dynamic! 🚀 Thrilled to share DynaEdit: a training-free, text-based method for non-rigid video editing. Work done during my internship at @GoogleDeepMind with @Roni_Paiss, @kusichan, @inbar_mosseri, @talidekel, @t_michaeli dynaedit.github.io

Sharing that DynVFX, our framework for augmenting real videos with dynamic content, is accepted to SIGGRAPH Asia 2025!🥳 Who is this framework for? EVERYONE. It's tuning-free, zero-shot, and fully automated – All you need to do is give an input video + simple text describing the new content. What's the trick? enforcing the generation to be content-aware of existing scene anchors via selectively extending the attention mechanism in a T2V DiT model. And code is finally out!: github.com/DynVFX/dynvfx Read the paper for more details: arxiv.org/abs/2502.03621 Website: dynvfx.github.io/ Grateful for presenting it recently at #MIT and next week at #SIGGRAPHAsia2025! With the amazing @RafailFridman, @omerbartal, @talidekel

Image editing? Video editing? What about image collections? Excited to share our latest work: Match-and-Fuse 🪇 match-and-fuse.github.io/ Check out how we tackle the largely unexplored task of set-to-set generation in a training-free mask-free manner by leveraging the off-the-shelf 2D matches and the emergent grid prior of T2I models in a joint framework. Work with the amazing @omri_kaduri and @talidekel 🧵 [1/3]

So much is already possible in image generation that it's hard to get excited. TokenVerse has been a refreshing exception! Disentangling complex visual concepts (pose, lighting, materials, etc.) from a single image — and mixing them across others with plug-and-play ease!

Prompt Theory (Made with Veo 3) What if AI-generated characters refused to believe they were AI-generated?

Excited to share that "TokenVerse: Versatile Multi-concept Personalization in Token Modulation Space" got accepted to SIGGRAPH 2025! It tackles disentangling complex visual concepts from as little as a single image and re-composing concepts across multiple images into a coherent result. token-verse.github.io #SIGGRAPH2025

Understanding the inner workings of foundation models is key for unlocking their full potential. While the research community has explored this for LLMs, CLIP, and text-to-image models, it's time to turn our focus to VLMs. Let's dive in! 🌟 vision-of-vlm.github.io/

🔍 Unveiling new insights into Vision-Language Models (VLMs)! In collaboration with @OneViTaDay & @talidekel, we analyzed LLaVA-1.5-7B & InternVL2-76B to uncover how these models process visual data. 🧵 vision-of-vlm.github.io/

Excited to introduce our new paper, Generative Omnimatte: Learning to Decompose Video into Layers, with the amazing team at Google DeepMind! Our method decomposes a video into complete layers, including objects and their associated effects (e.g., shadows, reflections).

We are excited to introduce "VidPanos: Generative Panoramic Videos from Casual Panning Videos" VidPanos converts phone-captured panning videos into (fully playing) video panoramas, instead of the usual (static) image panoramas. Website: vidpanos.github.io/ Paper: arxiv.org/abs/2410.13832 1/n

I think that's my favorite squish it example so far 📊😆

Introducing✨Still-Moving✨—our work from @GoogleDeepMind that lets you apply *any* image customization method to video models🎥 Personalization (DreamBooth)🐶stylization (StyleDrop) 🎨 ControlNet🖼️—ALL in one method! Plus… you can control the amount of generated motion🏃‍♀️ 🧵👇

Excited to present DINO-Tracker (accepted to #ECCV2024)! A novel self-supervised method for long-range dense tracking in video, which harnesses the powerful visual prior of DINO. Project page: dino-tracker.github.io. [1/4] @assaf_singer @OneViTaDay @talidekel