Taeksoo Kim

[Taeksoo Kim @taeksu98]

Looking forward to presenting Dexterous World Models in Denver! #CVPR2026

👥 Byungjun Kim(@byungjun__kim), Taeksoo Kim(co-first), Junyoung Lee(@junc0ng), and Hanbyul Joo(@jhugestar)

📍 Main Poster - Jun 7 (Sun), 11:45–13:45 (ExHall F, #97)
Also at:
🎬VideoWorldModel Workshop - Jun 3 (Wed), 9:50–10:40 (Room 705/707)
🤖 H2R Workshop - Jun 4 (Thu), 12:15–13:30 (Mile High 2A)

Feel free to stop by and chat. See you in Denver!

🌐 Project Page:
https://t.co/Lsk3rBkq7q
💻 Code:
https://t.co/EkfbKKYCHQ

[Byungjun Kim @byungjun__kim]

Heading to #CVPR2026!

We present 🖐️🌏 𝐃𝐞𝐱𝐭𝐞𝐫𝐨𝐮𝐬 𝐖𝐨𝐫𝐥𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐃𝐖𝐌) — a scene-action-conditioned video diffusion model that simulates human manipulation in static 3D scenes from egocentric hand motions.

👥 Byungjun Kim, Taeksoo Kim(co-first, @taeksu98), Junyoung Lee(@junc0ng), and Hanbyul Joo(@jhugestar)

📄 Paper: https://t.co/eIkb1Fo9RC
🌐 Project Page: https://t.co/FXO4olPBvC
💻 Code: https://t.co/xat7oAzy7U

📍 Main Poster - Jun 7 (Sun), 11:45–13:45 (ExHall F, #97)

Also at:
🎬VideoWorldModel Workshop - Jun 3 (Wed), 9:50–10:40 (Room 705/707)
🤖 H2R Workshop - Jun 4 (Thu), 12:15–13:30 (Mile High 2A)

See you in Denver!

[Jisoo Kim @jisooslog]

Introducing ZeroDex: Zero-Shot Long-Horizon Dexterous Manipulation via Multi-View 3D-Grounded VLM Reasoning

VLMs know what to do. ZeroDex grounds that reasoning in multi-view 3D, enabling robots to perform dexterous actions.

No task-specific demos. No policy fine-tuning.

[Hanbyul (Han) Joo @jhugestar]

We present our Vanast paper at #CVPR2026 as a highlight paper!

Please come see our paper and talk to our amazing first author @hyunssu_cha.

Project page: https://t.co/kWo9ULQkWT

[Hyunsoo Cha @hyunssu_cha]

🚀 Excited to share that our paper,
“Vanast: Virtual Try-On with Human Image Animation via Synthetic Triplet Supervision”
will be presented at CVPR 2026 as a Highlight 🎉

In this work, we introduce a unified framework that directly generates garment-transferred human animation videos from a single human image, garment images, and a pose guidance video—eliminating the limitations of conventional two-stage pipelines.

🔑 Key contributions:
• A single-stage framework for virtual try-on with human animation
• A scalable synthetic triplet supervision pipeline enabling large-scale training
• A Dual Module architecture that improves garment fidelity, pose adherence, and identity preservation
• Support for zero-shot garment interpolation and multi-garment transfer

Our approach achieves strong performance across multiple benchmarks and produces high-fidelity, identity-consistent animation results.

🙏 I’m truly grateful to my collaborators @wonjung_woo, @byungjun__kim, and my advisor @jhugestar for their invaluable support.

🔗 Project page: https://t.co/Dv9LubUZfM
📰 I’ll be presenting at Poster Session 1 in the Exhibit Hall!
📍 ExHall A–F | 🕙 10:45 AM – 12:45 PM | 📌 Poster #369

#CVPR2026 #Highlight #ComputerVision #GenerativeAI #VirtualTryOn #HumanAnimation #DiffusionModels

[Hanbyul (Han) Joo @jhugestar]

We present 𝐃𝐞𝐱𝐭𝐞𝐫𝐨𝐮𝐬 𝐖𝐨𝐫𝐥𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐃𝐖𝐌) at #CVPR 2026!

As this cool video by Byungjun shows, our model takes dexterous hand motion as input and produces a wide range of unseen manipulation tasks. Come check out our talk and poster!

Led by amazing co-first authors @byungjun__kim and @taeksu98. Joint work with @RLWRLD_ai

[Hanbyul (Han) Joo @jhugestar]

In TAViD, we introduce a target-aware video diffusion model that generates videos of an actor interacting with a specified target --- surprisingly hard for existing models.

We see target awareness as a key missing piece for controllable World Models. Check out our paper and code!

Huge credit to @taeksu98, currently seeking research internship opportunities. Taeksoo is an expert in world models, video models, and 3D human modeling. Feel free to reach out!

[Taeksoo Kim @taeksu98]

🌐Project: https://t.co/6qYhBV5hXX
💻Code: https://t.co/KCWtRPLqeI
Joint work with my advisor @jhugestar.
Would love to chat more at ICLR - feel free to reach out! (5/5)

[Taeksoo Kim @taeksu98]

One prompt. Any target.
Introducing TAViD: Target-Aware Video Diffusion Models
Presenting at #ICLR2026 🎉
w/ @jhugestar (1/5)

[Taeksoo Kim @taeksu98]

TAViD generations → 3D pose → imitation learning.
Video models become effective motion planners for robotics. (4/5)

[Byungjun Kim @byungjun__kim]

We introduce 🖐️🌏 𝐃𝐞𝐱𝐭𝐞𝐫𝐨𝐮𝐬 𝐖𝐨𝐫𝐥𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐃𝐖𝐌) — a scene-action-conditioned video diffusion model that simulates human manipulation in static 3D scenes from egocentric hand motions.

📄 Paper: https://t.co/u2IldpfYTD
🌐 Project Page: https://t.co/YSE3rOmV1s

[Hanbyul (Han) Joo @jhugestar]

One of the biggest challenges in building practical World Models is incorporating dexterous hand actions to simulate dynamic human-scene interactions.

We introduce DWM (Dexterous World Models) toward that goal.

Paper: https://t.co/unjFrWm5UM
Project Page: https://t.co/ILh1mf3pQb

Led by amazing @byungjun__kim and @taeksu98. Supported by @RLWRLD_ai.

[Taeksoo Kim @taeksu98]

The work has been done in collaboration with my wonderful co-authors,
📍Byungjun Kim(@byungjun_kim),
📍Shunsuke Saito(@psyth91), and
📍Hanbyul Joo(@jhugestar).

🔗Project Page: https://t.co/el2eVWlHfC
💻Code: https://t.co/tUvK1kedRC

(End/🧵)

[Taeksoo Kim @taeksu98]

Thrilled to present our work "GALA" at #CVPR2024!
GALA generates animatable layered assets from a single 3D human scan, which can be recomposed to make novel avatars.
📌 Main conference: Wed 10:30-12:00 #133
📌 Virtual Try-on Workshop: Mon 15:30-16:00, Arch 309, short talk
(1/🧵)

[Taeksoo Kim @taeksu98]

By composing the decomposed assets, GALA enables efficient generation of novel human avatars, which are ready to be animated.
(4/🧵)