Jun-Yan Zhu
@junyanz89
Followers
12K
Following
1K
Media
23
Statuses
352
Assistant Professor at Generative Intelligence Lab @CMU_Robotics @CarnegieMellon. Understanding and creating pixels.
Pittsburgh, PA
Joined April 2017
Check out our new unpaired learning method for instruction-following image editing models.
š New preprint! We present NP-Edit, a framework for training an image editing diffusion model without paired supervision. We use differentiable feedback from Vision-Language Models (VLMs) combined with distribution-matching loss (DMD) to learn editing directly. webpage:
1
3
74
Leaving Meta and PyTorch I'm stepping down from PyTorch and leaving Meta on November 17th. tl;dr: Didn't want to be doing PyTorch forever, seemed like the perfect time to transition right after I got back from a long leave and the project built itself around me. Eleven years
501
586
11K
We present MotionStream ā real-time, long-duration video generation that you can interactively control just by dragging your mouse. All videos here are raw, real-time screen captures without any post-processing. Model runs on a single H100 at 29 FPS and 0.4s latency.
37
150
1K
@AvaLovelace0 @RuixuanLiu_ @RamananDeva @ChangliuL @junyanz89 If you miss the talk or want to dive deeper, please also check out our poster and our interview! Poster Session Details: - Location: Exhibit Hall I #306 - Time: Wed 22 Oct 2:45 p.m. HST ā 4:45 p.m Read the interview in ICCV Daily: https://t.co/K791u7scoo
1
3
19
š Excited to share that BrickGPT ( https://t.co/yvi8cyrArX) received the ICCV Best Paper Award! Our first author, @AvaLovelace0, will present it from 1:30 to 1:45 p.m. today in Exhibit Hall III. Huge thanks to all the co-authors @RuixuanLiu_ @RamananDeva @ChangliuL @junyanz89
8
21
187
I will be presenting this work in today's afternoon poster session!! #ICCV2025 š
Oct 22 | šExhibit Hall 1 #157 | ā²ļø2:45 pm -4:45 pm HST Github: https://t.co/yzEy0XyqxF Please drop by the poster for any discussionš
github.com
SynCD: Generating Multi-Image Synthetic Data for Text-to-Image Customization (ICCV 2025) - nupurkmr9/syncd
Can we generate a training dataset of the same object in different contexts for customization? Check out our work SynCD, which uses Objaverse assets and shared attention in text-to-image models for the same. https://t.co/vHa2RTLHoX w/ @xi_yin_ @junyanz89 @imisra_ @smnh_azadi
0
1
14
#ICCV2025 best paper award & best paper honorable mention! RI researchers collaborated with @CSDatCMU and @CMU_ECE to bring some incredible work to the conference this yearšš§ š„ Check out the SCS news post on BrickGPT, which brought home best paper!
cs.cmu.edu
SCS researchers have developed a tool that uses text prompts to help people ā and even robots ā bring ideas to life with Lego bricks.
0
8
57
I'll be joining the faculty @JohnsHopkins late next year as a tenure-track assistant professor in @JHUCompSci Looking for PhD students to join me tackling fun problems in robot manipulation, learning from human data, understanding+predicting physical interactions, and beyond!
87
112
865
Really excited to be presenting this work as part of my talk in Higen workshop ( https://t.co/QZPZ1yqUgu) at 2 pm in Room 309 today, Oct 19 š Please check out if interested! #ICCV2025
Few-step diffusion isnāt only an acceleration trick. The big deal is that it can generate samples during training, like a GAN generatorāso you can apply any differentiable loss. We use this to train an image editor without paired data, with a VLM-based loss, without RL.
0
4
26
Few-step diffusion isnāt only an acceleration trick. The big deal is that it can generate samples during training, like a GAN generatorāso you can apply any differentiable loss. We use this to train an image editor without paired data, with a VLM-based loss, without RL.
9
40
399
HeyGen just hit $100M ARR this month, 29 months after we first reached $1M in April 2023. None of this happens without our incredible team, customers, partners, and community. Thank you š When we shared our first $1M milestone, it was to give back to the build-in-public
62
59
634
š New preprint! We present NP-Edit, a framework for training an image editing diffusion model without paired supervision. We use differentiable feedback from Vision-Language Models (VLMs) combined with distribution-matching loss (DMD) to learn editing directly. webpage:
2
29
173
More of what my colleagues and I have been working on in AI for fusion is public now
Weāre announcing a research collaboration with @CFS_energy, one of the worldās leading nuclear fusion companies. Together, weāre helping speed up the development of clean, safe, limitless fusion power with AI. āļø
53
131
3K
Over the past year, my lab has been working on fleshing out theory/applications of the Platonic Representation Hypothesis. Today I want to share two new works on this topic: Eliciting higher alignment: https://t.co/KY4fjNeCBd Unpaired rep learning: https://t.co/vJTMoyJj5J 1/9
10
119
696
Iām hiring PhD students for 2026 @TTIC_Connect. More details here:
22
172
720
Launch our first real product! Along with agent-based image editing workflows, our method natively allows region editing. https://t.co/CNqWna0yRN
8
15
146
We added š¤ demos for our group inference on FLUX.1 Schnell and FLUX.1 Kontext. Thanks @multimodalart for helping set this up so quickly! FLUX Schnell: https://t.co/Dgto1pR5ah FLUX Kontext: https://t.co/b89JFpJEBx GitHub:
github.com
Scalable group inference for generating high quality and diverse images with diffusion models. - GaParmar/group-inference
When exploring ideas with generative models, you want a range of possibilities. Instead, you often disappointingly get a gallery of near-duplicates. The culprit is standard I.I.D. sampling. We introduce a new inference method to generate high-quality and varied outputs. 1/n
0
6
35
When exploring ideas with generative models, you want a range of possibilities. Instead, you often disappointingly get a gallery of near-duplicates. The culprit is standard I.I.D. sampling. We introduce a new inference method to generate high-quality and varied outputs. 1/n
6
24
120
š„Ā Announcing the Creative Visual Content Generation, Editing & Understanding Workshop at #SIGGRAPH2025! For the first time in SIGGRAPH history, this pioneering workshop is officially part of theĀ Technical Programāand weāre celebrating with an exceptional lineup! āØ Join us as
0
9
30