SpaceXAI · Grok Imagine TeamGrok Imagine Image v2.0
Precise image generation and editing built for real creative work, with editing-first tools (magic wand, multi-ref, smart resize) and sharp typography.
Core Contributor
Member of Technical Staff · SpaceXAI
Xian Liu
I am a Member of Technical Staff at SpaceXAI on the Grok Imagine team, working on omni-model pre-training and visual generation. Before that, I was a Research Scientist on the Cosmos team at NVIDIA Research. I received my Ph.D. from the Multimedia Laboratory (MMLab) at The Chinese University of Hong Kong (CUHK) in 2025, and my Bachelor's degree from Zhejiang University in 2021.
My research interests lie in multimodal AI and foundation generative models, with a focus on omni-model pre-training and post-training. I am a core contributor to Grok Imagine, a leading image and video generation model, and to the NVIDIA Cosmos series, a family of open-source world foundation models and their post-trained variants.
News
- Grok Imagine Image v1.5 (Quality Mode) is released, ranking top-3 on the Image Edit Arena!
- Grok Imagine v1.0 is released on LMArena, ranking Top-3 in image generation and Top-1 in video generation!
- Grok Imagine v1.0 image and video generation is released!
- Joined xAI (now SpaceXAI) as a Member of Technical Staff, working on omni and visual generation models. Stay tuned for our release!
- We released Cosmos 2.5, an improved world foundation model. Check out our GitHub and technical report.
- One paper is accepted to NeurIPS 2025.
- One paper is accepted to ICCV 2025 as an oral presentation.
- Passed my Ph.D. defense and officially became Dr. Liu!
- We released Cosmos-Predict2, a world foundation model with improved quality. Models are open-sourced on GitHub and Hugging Face.
- We released Cosmos-Transfer1, a world model with multimodal controllability. Models are open-sourced on GitHub and Hugging Face.
- Two papers are accepted to CVPR 2025.
- Cosmos won the Best of CES, Best of AI, and Best Overall awards from CNET in 2025!
- We released Cosmos, a world foundation model platform for Physical AI. Models are open-sourced on GitHub and Hugging Face.
- Four papers are accepted to ICLR 2025.
- One paper is accepted to AAAI 2025.
- We released Cosmos-Tokenizer, a suite of state-of-the-art image and video tokenizers, with models available on GitHub and Hugging Face.
- Honored to receive the ECCV 2024 Outstanding Reviewer Award. Many thanks for the recognition!
- Two papers are accepted to ECCV 2024.
- Joined NVIDIA Research as a full-time Research Scientist, building large-scale foundation models. Stay tuned for our release!
- One paper is accepted to ICML 2024.
- Started my internship at NVIDIA Research. See you in Santa Clara!
- Two papers are accepted to CVPR 2024, with HumanGaussian selected as a Highlight (Top 2.8%). See you in Seattle!
- One paper is accepted to ICLR 2024, with HyperHuman receiving review scores of 6, 6, 8, 10 (Top 1.6%, rank).
- I will intern at Deep Imagination Research @ NVIDIA Research in Spring 2024 with Ming-Yu Liu. See you in Santa Clara!
- HumanGaussian, a high-quality 3D human generation framework, is released with all code and models available!
- HyperHuman, a hyper-realistic human generation foundation model developed with Snap Research, is on arXiv!
- One paper is accepted to ICCV 2023.
- Started my internship at Snap Research. See you in Los Angeles!
- Two papers are accepted to CVPR 2023.
- One paper is accepted to TMLR.
- One paper is accepted to NeurIPS 2022, with ANGIE selected as a Spotlight (Top 5%)!
- Three papers are accepted to ECCV 2022, with SSP-NeRF selected for an Oral (Top 2.7%)!
- One paper is accepted to CVPR 2022.
- One paper is accepted to AAAI 2022.
Foundation Models
Large-scale models I have built as a core contributor.
Grok Imagine · SpaceXAI
SpaceXAI · Grok Imagine Team
SpaceXAI · Grok Imagine TeamGrok Imagine Image v1.5 (Quality Mode)
Production-grade image generation and editing behind Grok Imagine's Quality Mode.
Core Contributor
SpaceXAI · Grok Imagine TeamGrok Imagine v1.0
A leading image and video generation model.
Core Contributor · Grok Imagine Image
Cosmos · NVIDIA Research
NVIDIA · Cosmos TeamCosmos: World Foundation Model Platform for Physical AI
Core Contributor · Auto-regressive foundation model pre-training & post-training
NVIDIA · Cosmos TeamCosmos 2.5: Improved World Simulation with Video Foundation Models for Physical AI
Core Contributor · Data processing pipelines, captioning, long video generation, evaluation, transfer post-training
NVIDIA · Cosmos TeamCosmos-Transfer: Conditional World Generation with Adaptive Multimodal Control
Core Contributor · Adaptive multimodal control, data processing pipelines, open-source repo
Selected Publications
* Equal contribution. Full list on Google Scholar.
HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion
International Conference on Learning Representations (ICLR), 2024 Scores 6/6/8/10 · Top 1.6%
Semantic-Aware Implicit Neural Audio-Driven Video Portrait Generation
European Conference on Computer Vision (ECCV), 2022 Oral · Top 2.7%
Audio-Driven Co-Speech Gesture Video Generation
Advances in Neural Information Processing Systems (NeurIPS), 2022 Spotlight · Top 5%
HMAR: Efficient Hierarchical Masked AutoRegressive Image Generation
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025
TC4D: Trajectory-Conditioned Text-to-4D Generation
European Conference on Computer Vision (ECCV), 2024
Experience
Industry
Member of Technical StaffDec 2025 – Present
SpaceXAI · Omni Pretrain Team
Research ScientistJun 2024 – Dec 2025
NVIDIA Research · Cosmos Team
Research Internships
Generative AI Research InternMar 2024 – Jun 2024
NVIDIA Research · Deep Imagination Research Group
Supervisors: Ting-Chun Wang, Ming-Yu Liu
Research Visiting StudentDec 2023 – Mar 2024
University of Toronto · Toronto Computational Imaging Group
Hosts: Sherwin Bahmani, David B. Lindell
Research InternMay 2023 – Sep 2023
Snap Research · Creative Vision Group
Supervisors: Jian Ren, Sergey Tulyakov
Education
Invited Talks
-
Towards Multi-Modal Visual Generation: From Human Modeling to World Foundation Models The University of Hong Kong · Nanjing University · SenseTime2025
-
HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion AI TIME2024
-
Audio-Driven Co-Speech Gesture Video Generation AI TIME2023
-
Semantic-Aware Implicit Neural Audio-Driven Video Portrait Generation TechBeat2022
Selected Honors & Awards
- CNET Best of CES 2025: Best of AI and Best Overall (Cosmos)2025
- ECCV Outstanding Reviewer Award2024
- CVPR Travel Award2024
- ICLR Travel Award2024
- Hong Kong Ph.D. Fellowship Scheme (HKPFS)2021 – 2025
- Outstanding Graduate of Zhejiang Province2021
- Outstanding Bachelor Thesis Award of Zhejiang University (Top 1%)2021
- National Scholarship2019, 2020
- First Class Scholarship for Academic Excellence2019, 2020
- UCLA CSST Scholarship Program2020
- SenseTime Scholarship2020
- Tang Lixin Scholarship2019
Professional Service
- Area Chair / Senior Program Committee
- AAAI
- Conference Reviewer
- CVPR, ECCV, ICCV, WACV, SIGGRAPH, SIGGRAPH Asia, ACL, NeurIPS, ICML, ICLR, AISTATS, AAAI, ACM MM
- Journal Reviewer
- TPAMI, IJCV, TVCG, TIP, TMM, EG, CGF, PG
Teaching
- ENGG 1120 · Linear Algebra for Engineers (TA, CUHK)Spring 2022
- ENGG 2440 · Discrete Mathematics for Engineers (TA, CUHK)Fall 2021