Xintao
Xintao Wang is an AI researcher and solo creator/research-project leader focused on practical image and video restoration, controllable and multimodal generation, open-source video generation, and—currently—multimodal video foundation models and post-training. His official site describes work on restoration/enhancement, image/video/3D generation, RL post-training, and unified multimodal models.
At a Glance
- AI/computer-vision researchers
- Open-source developers
- Academic and industry restoration teams
- Creative-media and image/video production users
- +1 more
AI Tools by Xintao
(1)Real-ESRGAN
Open Source Image Upscaler
Discussions
No discussions yet
Be the first to start a discussion about Xintao
Latest News
Two Wang-affiliated papers, Native4DVA and VR-Thinker, accepted to ICML 2026.
VFXMaster accepted to SIGGRAPH 2026.
Six papers accepted to CVPR 2026, including DUO-VSR, CineScene, GRPO-Guard and VideoCanvas.
Kling 3.0-Omni released, according to Wang's official homepage.
Products & Services
Open-source practical algorithm and toolkit for real-world blind image/video super-resolution and restoration, trained using a high-order degradation model and pure synthetic data; supports general, anime, arbitrary-scale, tiled, grayscale, alpha-channel, and 16-bit inference, with optional face enhancement via GFPGAN.
Generative facial-prior method and open-source toolkit for practical blind face restoration; provides pretrained models and an inference pipeline for restoring degraded faces.
Open-source image and video restoration toolbox covering super-resolution, denoising, deblurring and related algorithms, including EDSR, RCAN, SRGAN, ESRGAN, EDVR, BasicVSR, SwinIR and ECBSR.
Lightweight controllable-generation adapters that add conditions such as sketches, keypoints and other controls to text-to-image diffusion models; the SDXL release was done with Hugging Face.
Market Position
Xintao Wang's work is positioned around practical, widely adopted open-source restoration and generation research rather than a standalone company. Real-ESRGAN targets general real-world restoration and is presented alongside alternatives such as waifu2x for anime-oriented enhancement; T2I-Adapter is described as an influential controllable-generation approach developed around the same period as ControlNet. His later Kling work competes in multimodal and controllable video generation with closed and open video-foundation-model efforts.
Founding Story
Xintao Wang's public profile is an individual researcher's homepage rather than a separately incorporated company. His work grew from academic research at CUHK's Multimedia Lab and subsequent Tencent ARC Lab/AI Lab research into practical, open-source restoration and generation systems.
Business Model
Revenue Model
Public evidence describes an individual researcher publishing open-source code, pretrained models, papers and demos; no standalone commercial revenue model is identified.
Target Markets
- AI/computer-vision researchers
- Open-source developers
- Academic and industry restoration teams
- Creative-media and image/video production users
- Organizations building multimodal and generative-video systems
- Restoring and upscaling degraded photographs and video
- Face restoration and enhancement
- Anime image/video enhancement
- Research and fine-tuning of image-restoration models
- Controllable image and video generation for creators and researchers
- Reference-based video generation, camera control and video editing