FrancisTu

CS Ph.D. at @FudanUniversity | Diffusion Model Pretraining | MoE/MoT Scaling | Video Generation
@TencentHunyuan Project UP Talent Program | @MSFTResearch

[FrancisTu @FrancisRing_Tu]

@KhoiNg05 Thanks!

[FrancisTu @FrancisRing_Tu]

StableAvatar NeurIPS 554 accept!
Baton NeurIPS 5554 accept!
Finally, my first-author paper got accepted at a top-3 ML conference 🩵
See u guys in Sydney! #NeurIPS2026

[FrancisTu @FrancisRing_Tu]

🚀 The first survey on Agentic Visual Generation is out!

We review 300+ works across Image, Video, 3D, World Models, Slides & Rendering, mapping the rapidly evolving landscape of agentic visual generation. @_akhaliq

📄 https://t.co/KrgajRKWdy ⭐ https://t.co/M0XJ5j4NMT

[FrancisTu @FrancisRing_Tu]

@ZhMlliang Thanks!

[yinminghuang @adamhuang1828]

Excited to share our work, VA-Judger, the first reward model for joint video-audio generation! 🎥🔊
Powered by VA-Judger, post-trained LTX-2 generates high-quality content that better aligns with human preferences.

📄Paper: https://t.co/qlb1w7xHGo
💻Code: https://t.co/znBST9dPH9

[yinminghuang @adamhuang1828]

🚀 Awesome Agentic Visual Generation is live!
A taxonomy-driven paper collection covering image, video, 3D & world models, organized into four levels of controller capability.

⭐ Explore, star & contribute:
https://t.co/agE81HgVEf

📄 Survey coming soon!

#AIAgents #GenerativeAI

[ZHYang @yang_zihan79147]

Excited to share our work: ArcFlow, a 2-Step Text-to-Image Generation Framework via High-Precision Non-Linear Flow Distillation.

Code: https://t.co/lt3CwyWii1

It ensures high-quality alignment with teacher, delivering 40× speedup and 4× faster convergence with <5% parameters.

[FrancisTu @FrancisRing_Tu]

Excited to share our work: FlashPortrait

It is more powerful than the Runway Act series and open-source, capable of synthesizing infinite-length videos while achieving up to 6X acceleration in inference speed.

Paper: https://t.co/RkG7SGeTdm
Code: https://t.co/YXAFd6kHJG

[FrancisTu @FrancisRing_Tu]

Have you ever seen Professor Snape like this at Hogwarts? With StableAvatar, anything is possible! We’re excited to share brand-new demos synthesized by StableAvatar!

Paper: https://t.co/6s8SwzFIKg
Code:https://t.co/i7jdxgZYcT

[yinminghuang @adamhuang1828]

🚀 Check out StableAvatar demo on Hugging Face!
🎭 Audio-driven avatar video generation with just 1 image + 1 audio.
👉 https://t.co/yhoURQJEAF

⚠️ Due to video generation limits, currently only Hugging Face Pro users can try it out.
Thanks @gradio for the awesome interface! 💡

[FrancisTu @FrancisRing_Tu]

Excited to share our work: StableAvatar

It is the first Wan2.1-1.3B-based video diffusion transformer, which synthesizes infinite-length high-quality avatar videos, conditioned on a reference image and audio.

Paper: https://t.co/6s8SwzFIKg
Code:https://t.co/i7jdxgZYcT

[FrancisTu @FrancisRing_Tu]

Website: https://t.co/ECZK4K06BX
Demo: https://t.co/47Mbhrx8iQ

[FrancisTu @FrancisRing_Tu]

Excited to share our #CVPR2025 paper: StableAnimator

It is the first end-to-end ID-preserving video diffusion framework, which synthesizes videos without any post-processing, conditioned on an image and poses!

Paper: https://t.co/mwQPTnpaok
Code: https://t.co/rbIqdQPHLk