Yifan Pu

Ph.D. Student @LeapLabTHU @Tsinghua_Uni. Research Intern
@MSFTResearch, @canva CORE, Ex. On Job Market Now.

[netsky remix @iron_redux]

🚨🚨Breaking news from the math world! Open AI just proved nonsofic groups existed! GPT 5.6 Sol continues to replace mathematicians!!!

[levent @__alpoge__]

hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final

((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)

[Yifan Pu @keeplea58276507]

@osanseviero @_AndrewZhao

[Yifan Pu @keeplea58276507]

@ShenzhiWang_THU why you have so many best paper

[Yifan Pu @keeplea58276507]

One more best paper from our lab @LeapLabTHU

[ICML Conference @icmlconf]

🏆Outstanding Paper is awarded to the following research papers:

The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models

High-Accuracy Sampling for Diffusion Models and Log-Concave Distributions

[ICML Conference @icmlconf]

🏆Outstanding Paper is awarded to the following research papers:

The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models

High-Accuracy Sampling for Diffusion Models and Log-Concave Distributions

[Yifan Pu @keeplea58276507]

@ShenzhiWang_THU @ZanlinNi1 outstanding paper award @icmlconf ICML 2026

[Yifan Pu @keeplea58276507]

by our team in Canva CORE

[Canva @canva]

PSA 📣 You can now turn your ChatGPT-generated images into fully editable Canva designs with Magic Layers, without ever leaving the chat.

[Canva @canva]

PSA 📣 You can now turn your ChatGPT-generated images into fully editable Canva designs with Magic Layers, without ever leaving the chat.

[Yifan Pu @keeplea58276507]

@RainbowYuhui amazing work

[Ryan Yuan @RainbowYuhui]

We’re excited to launch #MagicLayers globally at Canva 🚀🚀🚀

This is currently the most advanced image-to-layer decomposition model in the world-purpose-built for design-native content.

⚡ 20–200× faster than Qwen-Image-Layer
🎨 Super strong on designs

https://t.co/vQ7cCPnGGZ

[Ryan Yuan @RainbowYuhui]

[1/2] Most image gen models output flat pixels. Users need layers. Our MRT (CVPR'2026) closes that gap and supports:
• text → layers
• image → layers
• layers → layers

🔥 New SOTA on all three tasks
🔥 Beats Qwen-Image-Layered
🔥 10–100× faster, 50–90% less GPU memory

[Yifan Pu @keeplea58276507]

@danielshkao what a pity

[Yifan Pu @keeplea58276507]

@JustinLin610 @_AndrewZhao do you know why

[Yifan Pu @keeplea58276507]

@jietang @bindureddy @_AndrewZhao

[Yifan Pu @keeplea58276507]

@danielshkao 牛的

[Shenzhi Wang🌟 @ShenzhiWang_THU]

🚨The "killer feature" of diffusion LLMs (arbitrary order) might actually be killing their reasoning potential.

🚀Introducing JustGRPO: An embarrassingly simple approach for better realizing reasoning potential in dLLMs.

🔗https://t.co/yQsYrUV61h

First author @ZanlinNi1

🧵⬇️

[Yifan Pu @keeplea58276507]

@ytz2024 👏

[Yifan Pu @keeplea58276507]

@_AndrewZhao @willknight @ZilongZheng @WIRED very good job, from very smart people

[Yifan Pu @keeplea58276507]

@LeapLabTHU

[Yifan Pu @keeplea58276507]

Bringing sCM & MeanFlow to Text-to-Image! 🎨📷

A new guide from @LeapLabTHU and AlibabaDAMO for distilling a 8B FLUX.1-lite into a high-performance, few-step generator.

Paper: https://t.co/bwzSq4X71n
Code: https://t.co/QWWh19dNVg