Trending Papers

The study reveals that in text-to-image generation, CFG Augmentation is the primary driver of few-step distillation in Distribution Matching Distillation (DMD), while the distribution matching term acts as a regularizer.

Z-Image, a 6B-parameter Scalable Single-Stream Diffusion Transformer (S3-DiT) model, achieves high-performance image generation with reduced computational cost, offering sub-second inference and compatibility with consumer hardware.

Wan-Move enhances motion control in video generative models by integrating motion-aware features into latent space, enabling high-quality and scalable video synthesis.

Visionary is an open web-native platform enabling real-time rendering of 3D Gaussian Splatting and meshes with efficient GPU-based inference, supporting dynamic content and generative models.

Live Avatar uses a 14-billion-parameter diffusion model with Timestep-forcing Pipeline Parallelism and Rolling Sink Frame Mechanism to achieve real-time, high-fidelity avatar generation.

PaperDebugger is an in-editor academic writing assistant that integrates large language models, enabling direct interaction within LaTeX editors for document state management, revision, and literature search.

DeepCode, a fully autonomous framework, addresses the challenges of document-to-codebase synthesis by optimizing information flow through source compression, structured indexing, knowledge injection, and error correction, achieving state-of-the-art performance and surpassing human experts.

LightRAG improves Retrieval-Augmented Generation by integrating graph structures for enhanced contextual awareness and efficient information retrieval, achieving better accuracy and response times.

Segment Anything Model 3 achieves state-of-the-art performance in promptable concept segmentation and tracking by leveraging a unified model architecture with decoupled recognition and localization.

TwinFlow is a 1-step generative model framework that enhances inference efficiency without requiring fixed pretrained teacher models or standard adversarial networks, achieving high performance on text-to-image tasks and scaling efficiently.

LongCat-Image is a bilingual open-source foundation model for image generation that addresses multilingual text rendering, photorealism, and deployment efficiency through rigorous data curation, compact design, and comprehensive open-source support.

Semantic-First Diffusion (SFD) enhances image generation by asynchronously denoising semantic and texture latents, improving convergence and quality.

MinerU is an open-source tool that enhances document content extraction using fine-tuned models and pre/postprocessing rules across diverse document types.

PaddleOCR-VL, a vision-language model combining NaViT-style dynamic resolution and ERNIE, achieves state-of-the-art performance in document parsing and element recognition with high efficiency.

MinerU2.5, a 1.2B-parameter document parsing vision-language model, achieves state-of-the-art recognition accuracy with computational efficiency through a coarse-to-fine parsing strategy.

Bind & Compose uses Diffusion Transformers with hierarchical binders and temporal strategies to accurately compose complex visual concepts from images and videos.

LlamaFactory is a unified framework enabling efficient fine-tuning of large language models across various tasks using a web-based user interface.

IndexTTS, an enhanced text-to-speech system combining XTTS and Tortoise models, offers improved naturalness, enhanced voice cloning, and controllable usage through hybrid character-pinyin modeling and optimized vector quantization.

A novel GPT-based model, OmniFlatten, enables real-time natural full-duplex spoken dialogue through a multi-stage post-training technique that integrates speech and text without altering the original model's architecture.

Mem0, a memory-centric architecture with graph-based memory, enhances long-term conversational coherence in LLMs by efficiently extracting, consolidating, and retrieving information, outperforming existing memory systems in terms of accuracy and computational efficiency.

Terrain Diffusion uses diffusion models and a novel algorithm called InfiniteDiffusion to generate realistic, seamless, and boundless procedural worlds with constant-time random access.

SAM 3D is a generative model that reconstructs 3D objects from single images using a multi-stage training framework that includes synthetic pretraining and real-world alignment, achieving high performance in human preference tests.

A multi-agent framework using large language models for stock trading simulates real-world trading firms, improving performance metrics like cumulative returns and Sharpe ratio.

MobiAgent, a comprehensive mobile agent system, achieves state-of-the-art performance in real-world mobile scenarios through its MobiMind-series models, AgentRR framework, and MobiFlow benchmarking suite, while also reducing data annotation costs.

RealGen is a photorealistic text-to-image framework that uses an LLM for prompt optimization and a diffusion model for image generation, enhanced by a Detector Reward mechanism and RealBench for automated evaluation.

Zep, a memory layer service, outperforms MemGPT in the DMR benchmark and LongMemEval by excelling in dynamic knowledge integration and temporal reasoning, critical for enterprise use cases.

A vision-language model (VLM) named GLM-4.1V-Thinking, developed with a reasoning-centric training framework, achieves state-of-the-art performance across various tasks, including STEM problem solving, video understanding, and long document understanding, outperforming larger models on many benchmarks.

MVP, a scalable multi-view transformer architecture, efficiently reconstructs large 3D scenes from multiple images using dual hierarchies and achieves state-of-the-art quality.

A framework called Easy Dataset synthesizes fine-tuning data from unstructured documents using a GUI and LLMs, improving domain-specific performance of LLMs while maintaining general knowledge.

UI-TARS, a native GUI agent model using screenshots as input, outperforms commercial models in various benchmarks through enhanced perception, unified action modeling, system-2 reasoning, and iterative training with reflective online traces.

DAComp is a benchmark of 210 tasks that evaluates the capabilities of agents in real-world data engineering and data analysis workflows, revealing significant deficiencies in both areas.

Depth Anything 3 (DA3) uses a plain transformer for geometry prediction from visual inputs, achieving state-of-the-art results in camera pose estimation, any-view geometry, visual rendering, and monocular depth estimation.

RAG-Anything is a unified framework that enhances multimodal knowledge retrieval by integrating cross-modal relationships and semantic matching, outperforming existing methods on complex benchmarks.

Flawed-Aware Policy Optimization (FAPO) enhances reinforcement learning with verifiable rewards by penalizing flawed-positive rollouts, improving reasoning capability and training stability in large language models.

Agent Lightning is a flexible RL framework for training LLMs in various agents, using a hierarchical RL algorithm and decoupling execution from training to handle complex interactions.

PyTorch Fully Sharded Data Parallel (FSDP) enables efficient and scalable training of large models across hardware configurations.

OneThinker, an all-in-one multimodal reasoning model, unifies image and video understanding across various tasks using RL and demonstrates strong performance and knowledge transfer.

The PyTorch distributed data parallel module optimizes large-scale model training using techniques like gradient bucketing, computation-communication overlap, and selective synchronization to achieve near-linear scalability.

GigaBrain-0, a VLA foundation model, uses world model-generated data to enhance cross-task generalization and policy robustness, improving real-world performance on complex manipulation tasks.

InternVL3 is a multimodal pre-trained language model that jointly learns from both multimodal data and text, improving performance and scalability through advanced techniques and setting a new state-of-the-art in multimodal tasks.

InkSight converts offline handwriting to digital ink using novel reading and writing priors, effectively derendering handwritten text in diverse and challenging conditions.

DeepAnalyze-8B, an agentic LLM, autonomously completes the data science pipeline from raw data to research reports using curriculum-based training and data-grounded trajectory synthesis.

AgentScope enhances agentic applications by providing flexible tool-based interactions, unified interfaces, and advanced infrastructure based on the ReAct paradigm, supporting efficient and safe development and deployment.

A simple head-specific sigmoid gate applied after Scaled Dot-Product Attention improves performance, stability, and scaling in large models, mitigating 'attention sink' and enhancing long-context extrapolation.

LucidFlux, a caption-free UIR framework using a diffusion transformer, achieves robust image restoration through adaptive conditioning and SigLIP features without text prompts.

SAM-Body4D is a training-free framework that enhances 3D human mesh recovery from videos by ensuring temporal consistency and robustness to occlusions through masklet generation and refinement.

A vision-action policy using correlated noise for flow matching and learnable mixed-layer attention wins the 2025 BEHAVIOR Challenge with high performance across diverse household tasks.

VideoCoF, a Chain-of-Frames approach, improves video editing precision and instruction-to-region mapping by using reasoning tokens without requiring user-provided masks.

Today's denoising diffusion models do not "denoise" in the classical sense, i.e., they do not directly predict clean images. Rather, the neural networks predict noise or a noised quantity. In this paper, we suggest that predicting clean data and predicting noised quantities are fundamentally different. According to the manifold assumption, natural data should lie on a low-dimensional manifold, whereas noised quantities do not. With this assumption, we advocate for models that directly predict clean data, which allows apparently under-capacity networks to operate effectively in very high-dimensional spaces. We show that simple, large-patch Transformers on pixels can be strong generative models: using no tokenizer, no pre-training, and no extra loss. Our approach is conceptually nothing more than "Just image Transformers", or JiT, as we call it. We report competitive results using JiT with large patch sizes of 16 and 32 on ImageNet at resolutions of 256 and 512, where predicting high-dimensional noised quantities can fail catastrophically. With our networks mapping back to the basics of the manifold, our research goes back to basics and pursues a self-contained paradigm for Transformer-based diffusion on raw natural data.

NEO, a novel family of native Vision-Language Models, addresses fundamental constraints and integrates vision and language within a unified framework, achieving competitive performance with limited data.