Xingang Pan

Introducing StoryMem — a memory-augmented framework for multi-shot long video storytelling. StoryMem carefully injects compact memory into the generation process with minimal overhead, enabling: • Cross-shot consistency • Smooth transitions • Narrative coherence across minutes-long videos Awesome work by Kaiwen @sze68zkw Project: kevin-thu.github.io/StoryMem/ arXiv: arxiv.org/pdf/2512.19539 Code: github.com/Kevin-thu/Stor…

StoryMem Multi-shot Long Video Storytelling with Memory huggingface.co/papers/2512.19…

We introduce Log-linear Sparse Attention (LLSA), a trainable sparse attention mechanism that reduces attention complexity from O(N²) to O(N log N). 📄 Paper: arxiv.org/abs/2512.16615 💻 Code: github.com/SingleZombie/L…

Can video generative models exhibit visuospatial intelligence? 🤔 Introducing Video4Spatial — a video-only framework that tackles spatial tasks. With just video context, our model can: 🔍 Ground objects by planning geometry-consistent paths 📸 Follow camera-pose instructions for scene navigation 🌐 Generalize to long contexts & unseen outdoor scenes A step toward video models as visual-spatial reasoners. Project: xizaoqu.github.io/video4spatial/ arXiv: arxiv.org/pdf/2512.03040

Introducing 📦𝗔𝗿𝘁𝗶𝗟𝗮𝘁𝗲𝗻𝘁🔧 (SIGGRAPH Asia 2025) — a high-quality 3D diffusion model that explicitly models object articulation, paving the way for richer, more realistic assets in embodied AI and simulation: – Generates fully articulated 3D objects – Physically plausible joints & motion – High-fidelity 3D Gaussian appearance – Supports generation from a single real image arXiv: arxiv.org/pdf/2510.21432 Project: chenhonghua.github.io/MyProjects/Art… Code (coming soon): github.com/chenhonghua/Ar…

STream3R Scalable Sequential 3D Reconstruction with Causal Transformer

Cool work that connects the idea of volume rendering with image diffusion!

Introducing 𝗦𝗧𝗿𝗲𝗮𝗺𝟯𝗥, a new 3D geometric foundation model for efficient 3D reconstruction from streaming input. Similar to LLMs, STream3R uses casual attention during training and KVCache at inference. No need to worry about post-alignment or reconstructing from scratch. You can easily add new frames and update the reconstruction incrementally. Great work by Yushi @GROS17121524 and Yihang @TheYihangLuo ! Project: nirvanalan.github.io/projects/strea… arXiv: arxiv.org/abs/2508.10893 Code: github.com/NIRVANALAN/STr… See a streaming reconstruction of our S-Lab lobby below!

Grok 4 one shots building a gemma-3-270m chatbot with transformers.js one click deploy in anycoder

Directly training Video Diffusion Models on long videos faces huge memory and learning challenges. How do we model long-range temporal distribution then? Our ICCV 2025 work, 🎞️𝗧𝗼𝗸𝗲𝗻𝘀𝗚𝗲𝗻, offers a solution. We compress videos into a highly condensed token space, enabling a DiT to model token distribution over much longer time ranges. A second-stage DiT then "decodes" these tokens back into long videos. While academic computing constraints currently prevent us from matching industrial VDM quality, TokensGen demonstrates an efficient and scalable design. We're optimistic it will pave the way for generating exceptionally long videos, like movies, in the future. Project: vicky0522.github.io/tokensgen-webp… arXiv: arxiv.org/abs/2507.15728 Github (to be released): github.com/Vicky0522/Toke…

𝗪𝗼𝗿𝗹𝗱𝗠𝗲𝗺 is mainly created by @zeqi_xiao Project page: xizaoqu.github.io/worldmem/ ArXiv: arxiv.org/abs/2504.12369 Github: github.com/xizaoqu/WorldM… Demo: huggingface.co/spaces/yslan/w…

Synthesizing worlds with video diffusion models is often inconsistent — moving the camera back and forth leads to different scenes. We propose 🌐𝗪𝗼𝗿𝗹𝗱𝗠𝗲𝗺, a memory-based approach that ensures consistent world simulation without relying on explicit 3D reconstruction.

Diffusion models are sensitive to small changes in the input noise. We introduce Alias-Free Latent Diffusion Models (𝗔𝗙-𝗟𝗗𝗠) at #CVPR2025. It achieves shift-equivariance and generates consistent outputs. Project: zhouyifan.net/AF-LDM-Page/ arXiv: arxiv.org/abs/2503.09419

arXiv: arxiv.org/abs/2503.08434

The Bokeh Effect is so important in photography, yet existing text2image diffusion models do not support controling bokeh strength. We introduce 𝗕𝗼𝗸𝗲𝗵 𝗗𝗶𝗳𝗳𝘂𝘀𝗶𝗼𝗻, a T2I diffusion model that supports flexible background blur control! Project: atfortes.github.io/projects/bokeh…

💥 Consistent Multi-View Diffusion for 3D Enhancement 💥 Introducing our work #3DEnhancer @CVPR: a multi-view diffusion model that enhances multi-view images to improve 3D models. 📰arXiv: arxiv.org/abs/2412.18565 🔥Project: yihangluo.com/projects/3DEnh…

🎉Excited to share Neural LightRig!🎉 It allows for accurate and fast estimation of surface normals and PBR materials from just one image. We achieve this by generating multi-light images with a diffusion model, overcoming the estimation ambiguity of inverse rendering.🚀 Page: projects.zxhezexin.com/neural-lightrig