生成式 AI
AI 图像生成、视频生成、音乐创作等 AIGC 领域最新动态。
Seedance2.0生成视频价格公布，生成视频一秒1块钱
字节跳动旗下火山引擎正式公布了其视频生成模型Seedance2.0的商用定价。服务分为含视频输入的编辑模式（28元/百万tokens）和不含视频输入的纯生成模式（46元/百万tokens）。根据官方数据，生成一段15秒的标准视频约消耗30....
Physically Valid Biomolecular Interaction Modeling with Gauss-Seidel Projection
Researchers have developed a differentiable AI module that enforces strict steric feasibility in biomolecular interactio...
Physically Valid Biomolecular Interaction Modeling with Gauss-Seidel Projection
A novel differentiable Gauss-Seidel projection module enforces physical constraints in AI-generated biomolecular structu...
Physically Valid Biomolecular Interaction Modeling with Gauss-Seidel Projection
A novel AI module integrates a differentiable Gauss-Seidel projection to enforce physical steric constraints in biomolec...
Higher Gauge Flow Models
Higher Gauge Flow Models are a novel class of generative AI models that extend ordinary Gauge Flow Models by incorporati...
Higher Gauge Flow Models
Higher Gauge Flow Models represent a novel class of generative artificial intelligence that incorporates L∞-algebra stru...
Higher Gauge Flow Models
Higher Gauge Flow Models represent a novel generative AI architecture that extends traditional Gauge Flow Models by inco...
Higher Gauge Flow Models
Higher Gauge Flow Models represent a groundbreaking class of generative AI that extends ordinary Gauge Flow Models by in...
Higher Gauge Flow Models
Higher Gauge Flow Models represent a novel class of generative AI that incorporates advanced geometric structures like L...
Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage
The Function-space Decoupled Diffusion Posterior Sampling (Fun-DDPS) framework combines generative diffusion models with...
Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage
The Function-space Decoupled Diffusion Posterior Sampling (Fun-DDPS) framework is a novel generative AI method that comb...
Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage
The Function-space Decoupled Diffusion Posterior Sampling (Fun-DDPS) framework is a novel AI method that combines genera...
Continual Unlearning for Text-to-Image Diffusion Models: A Regularization Perspective
A new study reveals that text-to-image diffusion models experience 'utility collapse' during continual unlearning, where...
Continual Unlearning for Text-to-Image Diffusion Models: A Regularization Perspective
New research reveals that text-to-image diffusion models suffer from rapid utility collapse when processing sequential u...
Fine-Tuning Diffusion Models via Intermediate Distribution Shaping
Researchers introduced P-GRAFT, a novel fine-tuning framework for diffusion models that shapes intermediate probability ...
Fine-Tuning Diffusion Models via Intermediate Distribution Shaping
A new unified mathematical framework for fine-tuning pre-trained diffusion and flow models demonstrates significant impr...
Entering the Era of Discrete Diffusion Models: A Benchmark for Schr\"odinger Bridges and Entropic Optimal Transport
Researchers have introduced CATSBench, the first dedicated benchmark for evaluating Schrödinger bridge solvers on discre...
Entering the Era of Discrete Diffusion Models: A Benchmark for Schr\"odinger Bridges and Entropic Optimal Transport
Researchers have introduced the first comprehensive benchmark for evaluating Schrödinger Bridge (SB) solvers on discrete...
CREPE: Controlling Diffusion with Replica Exchange
CREPE (Controlling with REPlica Exchange) is a novel algorithm that enables flexible, real-time control of diffusion mod...
CREPE: Controlling Diffusion with Replica Exchange
CREPE (Controlling with REPlica Exchange) is a novel algorithm that enables inference-time control of diffusion models w...
Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search
Researchers developed AIGB-Pearl, a novel AI auto-bidding method that integrates conditional generative planning with po...
Gauge Flow Models
Gauge Flow Models represent a novel class of generative AI that integrates a learnable gauge field into flow ordinary di...
Improving Classifier-Free Guidance in Masked Diffusion: Low-Dim Theoretical Insights with High-Dim Impact
New research reveals a fundamental flaw in Classifier-Free Guidance (CFG) for discrete diffusion models, where applying ...
RNE: plug-and-play diffusion inference-time control and energy-based training
The Radon-Nikodym Estimator (RNE) is a fundamental mathematical tool that addresses a core limitation in diffusion model...
Strictly Constrained Generative Modeling via Split Augmented Langevin Sampling
The Constrained Alternated Split Augmented Langevin (CASAL) framework is a novel algorithmic approach that rigorously en...
Strictly Constrained Generative Modeling via Split Augmented Langevin Sampling
Researchers developed the Constrained Alternated Split Augmented Langevin (CASAL) algorithm to ensure deep generative mo...
Strictly Constrained Generative Modeling via Split Augmented Langevin Sampling
Researchers developed the Constrained Alternated Split Augmented Langevin (CASAL) framework, enabling deep generative mo...
Strictly Constrained Generative Modeling via Split Augmented Langevin Sampling
The Constrained Alternated Split Augmented Langevin (CASAL) framework is a novel algorithmic approach that rigorously en...
Infinite dimensional generative sensing
Researchers have established a rigorous theoretical framework for generative compressed sensing in infinite-dimensional ...
Infinite dimensional generative sensing
Researchers have established a rigorous mathematical framework for using deep generative models to solve inverse problem...
Infinite dimensional generative sensing
Researchers have developed a rigorous mathematical framework for using deep generative models to solve inverse problems ...
Variance reduction in lattice QCD observables via normalizing flows
Normalizing flow machine learning models enable major variance reduction in lattice Quantum Chromodynamics (QCD) calcula...
Variance reduction in lattice QCD observables via normalizing flows
Researchers have applied normalizing flow machine learning models to significantly reduce variance in lattice Quantum Ch...
Learning Demographic-Conditioned Mobility Trajectories with Aggregate Supervision
ATLAS is a novel weakly supervised AI framework that generates synthetic human mobility trajectories conditioned on demo...
Inverse Reconstruction of Shock Time Series from Shock Response Spectrum Curves using Machine Learning
Researchers have developed a conditional variational autoencoder (CVAE) that reconstructs acceleration time series signa...
Inverse Reconstruction of Shock Time Series from Shock Response Spectrum Curves using Machine Learning
Researchers have developed a conditional variational autoencoder (CVAE) that performs inverse reconstruction of shock ti...
Improving Diffusion Planners by Self-Supervised Action Gating with Energies
SAGE (Self-supervised Action Gating with Energies) is a novel inference-time method that significantly improves diffusio...
Bridging Diffusion Guidance and Anderson Acceleration via Hopfield Dynamics
A new research paper introduces Geometry Aware Attention Guidance (GAG), a theoretical framework that bridges diffusion ...
Bridging Diffusion Guidance and Anderson Acceleration via Hopfield Dynamics
A new research paper introduces Geometry Aware Attention Guidance (GAG), a theoretically grounded method that stabilizes...
Bridging Diffusion Guidance and Anderson Acceleration via Hopfield Dynamics
A new research paper introduces Geometry Aware Attention Guidance (GAG), a theoretical framework that bridges attention-...
Manifold Aware Denoising Score Matching (MAD)
Manifold Aware Denoising Score Matching (MAD) is a novel AI method that simplifies learning data distributions on manifo...
Manifold Aware Denoising Score Matching (MAD)
Manifold Aware Denoising Score Matching (MAD) is a novel AI research method that decomposes the score function into anal...
Spectral Regularization for Diffusion Models
Spectral regularization is a novel training framework that improves diffusion model performance by incorporating differe...
Spectral Regularization for Diffusion Models
Researchers have developed a spectral regularization framework that improves diffusion model sample quality by incorpora...
Spectral Regularization for Diffusion Models
A new spectral regularization framework enhances diffusion models by incorporating Fourier and wavelet domain losses dur...
Rigidity-Aware Geometric Pretraining for Protein Design and Conformational Ensembles
RigidSSL is a rigidity-aware self-supervised learning framework that addresses three core limitations in AI-driven prote...
Rigidity-Aware Geometric Pretraining for Protein Design and Conformational Ensembles
RigidSSL is a novel rigidity-aware self-supervised learning framework that improves AI-driven protein design through two...
Rigidity-Aware Geometric Pretraining for Protein Design and Conformational Ensembles
RigidSSL (Rigidity-Aware Self-Supervised Learning) is a novel geometric pretraining framework that improves protein desi...
Diffusion-MPC in Discrete Domains: Feasibility Constraints, Horizon Effects, and Critic Alignment: Case study with Tetris
Researchers applied Diffusion-based Model Predictive Control (Diffusion-MPC) to Tetris, demonstrating that feasibility-c...
Diffusion-MPC in Discrete Domains: Feasibility Constraints, Horizon Effects, and Critic Alignment: Case study with Tetris
Researchers applied Diffusion-MPC to Tetris using a MaskGIT-style discrete denoiser with feasibility-constrained samplin...
Diffusion-MPC in Discrete Domains: Feasibility Constraints, Horizon Effects, and Critic Alignment: Case study with Tetris
A new study applies Diffusion-MPC (Model Predictive Control) to the discrete combinatorial domain of Tetris, revealing c...
Quantum-Inspired Fine-Tuning for Few-Shot AIGC Detection via Phase-Structured Reparameterization
Researchers have developed Q-LoRA and H-LoRA, quantum-inspired fine-tuning methods that enhance AI-generated content (AI...
Quantum-Inspired Fine-Tuning for Few-Shot AIGC Detection via Phase-Structured Reparameterization
Researchers developed Q-LoRA, a quantum-enhanced variant of Low-Rank Adaptation (LoRA) that improves few-shot AI-generat...
Generalized Discrete Diffusion with Self-Correction
The Self-Correcting Discrete Diffusion (SCDD) model is a novel framework that reformulates self-correction in discrete d...
Value Gradient Guidance for Flow Matching Alignment
VGG-Flow is a novel fine-tuning method that efficiently aligns flow matching models like Stable Diffusion 3 with human p...
PrismAudio: Decomposed Chain-of-Thoughts and Multi-dimensional Rewards for Video-to-Audio Generation
PrismAudio is a novel Reinforcement Learning framework that addresses Video-to-Audio (V2A) generation by decomposing the...
SceneStreamer: Continuous Scenario Generation as Next Token Group Prediction
SceneStreamer is an AI framework that generates continuous, dynamic traffic scenarios for autonomous vehicle simulation ...
Interaction Field Matching: Overcoming Limitations of Electrostatic Models
Interaction Field Matching (IFM) is a novel machine learning framework that generalizes the Electrostatic Field Matching...
RealOSR: Latent Guidance Boosts Diffusion-based Real-world Omnidirectional Image Super-Resolutions
RealOSR is a novel diffusion-based framework for omnidirectional image super-resolution (ODISR) that addresses critical ...
Uni-Animator: Towards Unified Visual Colorization
Uni-Animator is a unified AI framework that leverages a Diffusion Transformer (DiT) architecture to perform both image a...