During token replacement, variations are injected via patch perturbations. This timestep-adaptive strategy balances identity preservation and generative flexibility.
@jiqizhixin
-
Personalize Anything: Mask-Guided Token Replacement for AI Models
By
–
Personalize Anything anchors subject identity in early denoising through mask-guided token replacement with preserved positional encoding, and transitions to multi-modal attention for semantic fusion with text in later steps.
-
Personalize Anything Free with Diffusion Transformer Technology
By
–
Tsinghua, Beihang, and RUC present: Personalize Anything for Free with Diffusion Transformer, enabling personalized image generation that produces user-specified concepts while allowing flexible editing.
-
VAMBA: Powerful Hybrid Mamba-Transformer Architecture
By
–
VAMBA is a powerful hybrid Mamba-Transformer architecture!
-
Li Auto’s MindVLA Redefines Autonomous Driving Technology
By
–
"Just like iPhone 4 redefined mobile phones,"Peng Jia, head of R&D of Li Auto’s autonomous driving technology, said, "For the automotive industry, MindVLA will also redefine autonomous driving."
-
MindVLA Navigates Complex Traffic Using RL and Multi-Agent Modeling
By
–
MindVLA can also navigate complex traffic environments by jointly modeling its own vehicle behavior and the trajectories of other vehicles. It employs RL to learn from mistakes, continuously optimizing its decision-making.
-
MindVLA: Fast-Slow Thinking with Action Token Diffusion
By
–
MindVLA integrates fast and slow thinking modes and can switch between them autonomously. When deployed, it uses Diffusion to decode Action Tokens into optimized trajectories.
-
Li Auto Unveils MindVLA Autonomous Driving Model at NVIDIA GTC
By
–
At NVIDIA GTC 2025, Li Auto, a Chinese car company, released more details about MindVLA, their next-generation autonomous driving model. MindVLA is a world model that deeply integrates their Novel View Restoration and Progressive Scene Reconstruction capabilities.
-
Research Paper and Data Available on arXiv
By
–
Paper: https://arxiv.org/pdf/2503.11346
Data and code: -
KG-Powered Multi-Agent System Reduces AI Hallucinations
By
–
Using a KG-powered multi-agent system to generate accurate biographies? That's exactly what a team from Zhejiang University just released. They built a historian AI called AIstorian, which has a much lower hallucination rate and significantly higher factual accuracy.