Bowen Xue

I explore visual generation and am actively seeking Fall 2027 PhD opportunities.

[Bowen Xue @BowenXue_]

Thanks for sharing these important works, DaS and FlexAM!

I think UniAction addresses a different core research question.

UniAction is motivated by the question of how heterogeneous action data can be jointly learned within a single model. To this end, we introduce a unified action-conditioning mechanism that allows diverse action types, including applied forces, camera commands, hand poses, and robot poses, to directly condition video generation without first converting them into a common motion representation.

For example, for force conditioning, we directly specify the application point and direction of the force, letting the model predict how the object responds. For camera control, we use discrete keyboard commands directly, without requiring predefined 3D trajectories. This differs from the setting of DaS/FlexAM, which uses tracked 3D point motion to control video generation.

Furthermore, we investigate whether different types of action data can benefit from joint training. Our experiments show that a single jointly trained model outperforms separately trained action-specific models.

We will also include citations to and discussion of DaS and FlexAM in a future revision to better clarify the connections and distinctions between these works.

Thanks again for sharing these papers and for your thoughtful suggestion!

[Zizhang Li @zizhang_li]

🌍 Actions that condition world models are highly heterogeneous: forces on objects, camera movements, hand motions, robot arms, each with its own meaning and structure. So today's action-conditioned world models are usually built one action type at a time, each with its own conditioning design.

We introduce UniAction, a unified action-conditioning framework: one world model driven by all of these actions, and even by mixes of them in a single generation. Trained jointly, it outperforms the action-specific experts. Great work led by amazing @BowenXue_ .

More detailed threads below👇
🧵 1/10

[lllyasviel @lvminzhang]

You can now chat with my thoughts and perspectives using this!

https://t.co/dW3dgSbNwf

[Ion Stoica @istoica05]

High-quality video generation now runs on the hardware you already own. We cannot wait to see what you can do with it!

[Chenfeng_X @Chenfeng_X]

FreeToken+VDN = FreeVideo, nice work from @HaochengXiUCB @BowenXue_ Xiaoze Fan and @Andy_ShuoYang

[Shuo Yang @Andy_ShuoYang]

FreeVideo + FreeToken will enable a fully automatic pipeline for short-drama generation: use VLMs to write prompts and design storylines, generate high-quality videos and visual assets with FreeVideo, and orchestrate the whole workflow with FreeToken. Really excited about what this can unlock for automated short-drama creation!

Huge thanks to @BowenXue_ and @HaochengXiUCB for making this happen!

[Haocheng Xi @HaochengXiUCB]

Video generation now runs right on your laptop. 🎬

Introducing FreeVideo — bringing MiniMax H3 + Video DeltaNet to the hardware you already own.

As little as 8GB VRAM + 16GB RAM. ComfyUI integration, LoRAs & custom workflows.

Code: https://t.co/sIKSsAuOGw

[Bowen Xue @BowenXue_]

@IMG_5955 完全な動画は https://t.co/UhYN47jtVf にあります。

[Bowen Xue @BowenXue_]

@denosoft @IMG_5955 これは動画の数フレームに過ぎません。完全な動画は https://t.co/UhYN47iW5H にあります。

[Papers with Code @HuggingPapers]

Tencent's WeChat Vision just released Stand-In!

A lightweight, plug-and-play framework for high-fidelity identity-preserving video generation.

It achieves state-of-the-art results with only ~1% extra parameters.

[青龍聖者 @bdsqlsz]

Tencent open source Stand-In:
a lightweight, plug-and-play framework for identity-preserving video generation.
Can be seamlessly integrated into other tasks such as subject-driven video generation, pose-controlled video generation, video stylization, and face swapping.

[Wildminder @wildmindai]

Stand-In: plug-and-play identity T2V adapter. Hits SOTA face-sim + naturalness, beats full finetunes. Drop-in for subject/non-human preservation, pose-guided (VACE), stylized vids, and face swap. LoRA-friendly. Wan2.1, Wan2.2 soon

https://t.co/PZZYZN8NFj

[AIGCLINK @aigclink]

腾讯新开源了一个轻量级、即插即用的身份保留视频生成框架：Stand-In，也支持换脸视频

生成效果人物脸部特征能很好的保留

支持非人类主体的视频生成、姿态控制视频生成、视频风格化，以及人脸替换

可以通过调整参数来控制人物的外观、动作、背景等等的视频生成效果

即插即用，可能轻松集成到现有T2V模型中

#视频生成 #AI视频 #StandIn

[Linoy Tsaban @linoy_tsaban]

Stand-In just landed on the hub 🔥
✨identity preserving video generation
✨lightweight LoRA with minimal affect on latency!
✨apache 2.0 license
plug-and-play approach, LoRA weights for Wan 2.1 available 👇

[Bowen Xue @BowenXue_]

It can also be seamlessly integrated into other tasks such as subject-driven video generation, pose-guided video generation, stylization, and face swapping.

[Bowen Xue @BowenXue_]

Excited to share our recent work at Tencent — Stand-In, a lightweight, plug-and-play framework for identity-preserving video generation.

🔗 Code: https://t.co/UhYN47jtVf
🌍 Project Page: https://t.co/DItmBG45NI
📄 Paper: https://t.co/KiWXUxMDAr

[Bowen Xue @BowenXue_]

With only ~1% additional parameters trained, Stand-In outperforms full-parameter training methods in both video quality and identity preservation.

[Bowen Xue @BowenXue_]

We introduce a conditional image branch into pre-trained video generation models, combined with restricted self-attentions enhanced by conditional position mapping, to achieve precise identity control.