Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching

Video generation models have demonstrated remarkable performance, yet their broader adoption remains constrained by slow inference speeds and substantial computational costs, primarily due to the iterative nature of the denoising process. Addressing this bottleneck is essential for democratizing advanced video synthesis technologies and enabling their integration into real-world applications. This work proposes EasyCache, a training-free acceleration framework for video diffusion models. EasyCache introduces a lightweight, runtime-adaptive caching mechanism that dynamically reuses previously computed transformation vectors, avoiding redundant computations during inference. Unlike prior approaches, EasyCache requires no offline profiling, pre-computation, or extensive parameter tuning. We conduct comprehensive studies on various large-scale video generation models, including OpenSora, Wan2.1, and HunyuanVideo. Our method achieves leading acceleration performance, reducing inference time by up to 2.1-3.3$\times$ compared to the original baselines while maintaining high visual fidelity with a significant up to 36% PSNR improvement compared to the previous SOTA method. This improvement makes our EasyCache a efficient and highly accessible solution for high-quality video generation in both research and practical applications. The code is available at https://github.com/H-EmbodVis/EasyCache.

DepthSync: Diffusion Guidance-Based Depth Synchronization for Scale- and Geometry-Consistent Video Depth Estimation

Diffusion-based video depth estimation methods have achieved remarkable success with strong generalization ability. However, predicting depth for long videos remains challenging. Existing methods typically split videos into overlapping sliding windows, leading to accumulated scale discrepancies across different windows, particularly as the number of windows increases. Additionally, these methods rely solely on 2D diffusion priors, overlooking the inherent 3D geometric structure of video depths, which results in geometrically inconsistent predictions. In this paper, we propose DepthSync, a novel, training-free framework using diffusion guidance to achieve scale- and geometry-consistent depth predictions for long videos. Specifically, we introduce scale guidance to synchronize the depth scale across windows and geometry guidance to enforce geometric alignment within windows based on the inherent 3D constraints in video depths. These two terms work synergistically, steering the denoising process toward consistent depth predictions. Experiments on various datasets validate the effectiveness of our method in producing depth estimates with improved scale and geometry consistency, particularly for long videos.

SketchColour: Channel Concat Guided DiT-based Sketch-to-Colour Pipeline for 2D Animation

The production of high-quality 2D animation is highly labor-intensive process, as animators are currently required to draw and color a large number of frames by hand. We present SketchColour, the first sketch-to-colour pipeline for 2D animation built on a diffusion transformer (DiT) backbone. By replacing the conventional U-Net denoiser with a DiT-style architecture and injecting sketch information via lightweight channel-concatenation adapters accompanied with LoRA finetuning, our method natively integrates conditioning without the parameter and memory bloat of a duplicated ControlNet, greatly reducing parameter count and GPU memory usage. Evaluated on the SAKUGA dataset, SketchColour outperforms previous state-of-the-art video colourization methods across all metrics, despite using only half the training data of competing models. Our approach produces temporally coherent animations with minimal artifacts such as colour bleeding or object deformation. Our code is available at: https://bconstantine.github.io/SketchColour .

Training-Free Motion Customization for Distilled Video Generators with Adaptive Test-Time Distillation

Distilled video generation models offer fast and efficient synthesis but struggle with motion customization when guided by reference videos, especially under training-free settings. Existing training-free methods, originally designed for standard diffusion models, fail to generalize due to the accelerated generative process and large denoising steps in distilled models. To address this, we propose MotionEcho, a novel training-free test-time distillation framework that enables motion customization by leveraging diffusion teacher forcing. Our approach uses high-quality, slow teacher models to guide the inference of fast student models through endpoint prediction and interpolation. To maintain efficiency, we dynamically allocate computation across timesteps according to guidance needs. Extensive experiments across various distilled video generation models and benchmark datasets demonstrate that our method significantly improves motion fidelity and generation quality while preserving high efficiency. Project page: https://euminds.github.io/motionecho/

SmoothSinger: A Conditional Diffusion Model for Singing Voice Synthesis with Multi-Resolution Architecture

Singing voice synthesis (SVS) aims to generate expressive and high-quality vocals from musical scores, requiring precise modeling of pitch, duration, and articulation. While diffusion-based models have achieved remarkable success in image and video generation, their application to SVS remains challenging due to the complex acoustic and musical characteristics of singing, often resulting in artifacts that degrade naturalness. In this work, we propose SmoothSinger, a conditional diffusion model designed to synthesize high quality and natural singing voices. Unlike prior methods that depend on vocoders as a final stage and often introduce distortion, SmoothSinger refines low-quality synthesized audio directly in a unified framework, mitigating the degradation associated with two-stage pipelines. The model adopts a reference-guided dual-branch architecture, using low-quality audio from any baseline system as a reference to guide the denoising process, enabling more expressive and context-aware synthesis. Furthermore, it enhances the conventional U-Net with a parallel low-frequency upsampling path, allowing the model to better capture pitch contours and long term spectral dependencies. To improve alignment during training, we replace reference audio with degraded ground truth audio, addressing temporal mismatch between reference and target signals. Experiments on the Opencpop dataset, a large-scale Chinese singing corpus, demonstrate that SmoothSinger achieves state-of-the-art results in both objective and subjective evaluations. Extensive ablation studies confirm its effectiveness in reducing artifacts and improving the naturalness of synthesized voices.

Improving Progressive Generation with Decomposable Flow Matching

Generating high-dimensional visual modalities is a computationally intensive task. A common solution is progressive generation, where the outputs are synthesized in a coarse-to-fine spectral autoregressive manner. While diffusion models benefit from the coarse-to-fine nature of denoising, explicit multi-stage architectures are rarely adopted. These architectures have increased the complexity of the overall approach, introducing the need for a custom diffusion formulation, decomposition-dependent stage transitions, add-hoc samplers, or a model cascade. Our contribution, Decomposable Flow Matching (DFM), is a simple and effective framework for the progressive generation of visual media. DFM applies Flow Matching independently at each level of a user-defined multi-scale representation (such as Laplacian pyramid). As shown by our experiments, our approach improves visual quality for both images and videos, featuring superior results compared to prior multistage frameworks. On Imagenet-1k 512px, DFM achieves 35.2% improvements in FDD scores over the base architecture and 26.4% over the best-performing baseline, under the same training compute. When applied to finetuning of large models, such as FLUX, DFM shows faster convergence speed to the training distribution. Crucially, all these advantages are achieved with a single model, architectural simplicity, and minimal modifications to existing training pipelines.

STAGE: A Stream-Centric Generative World Model for Long-Horizon Driving-Scene Simulation

The generation of temporally consistent, high-fidelity driving videos over extended horizons presents a fundamental challenge in autonomous driving world modeling. Existing approaches often suffer from error accumulation and feature misalignment due to inadequate decoupling of spatio-temporal dynamics and limited cross-frame feature propagation mechanisms. To address these limitations, we present STAGE (Streaming Temporal Attention Generative Engine), a novel auto-regressive framework that pioneers hierarchical feature coordination and multi-phase optimization for sustainable video synthesis. To achieve high-quality long-horizon driving video generation, we introduce Hierarchical Temporal Feature Transfer (HTFT) and a novel multi-stage training strategy. HTFT enhances temporal consistency between video frames throughout the video generation process by modeling the temporal and denoising process separately and transferring denoising features between frames. The multi-stage training strategy is to divide the training into three stages, through model decoupling and auto-regressive inference process simulation, thereby accelerating model convergence and reducing error accumulation. Experiments on the Nuscenes dataset show that STAGE has significantly surpassed existing methods in the long-horizon driving video generation task. In addition, we also explored STAGE's ability to generate unlimited-length driving videos. We generated 600 frames of high-quality driving videos on the Nuscenes dataset, which far exceeds the maximum length achievable by existing methods.

InterActHuman: Multi-Concept Human Animation with Layout-Aligned Audio Conditions

End-to-end human animation with rich multi-modal conditions, e.g., text, image and audio has achieved remarkable advancements in recent years. However, most existing methods could only animate a single subject and inject conditions in a global manner, ignoring scenarios that multiple concepts could appears in the same video with rich human-human interactions and human-object interactions. Such global assumption prevents precise and per-identity control of multiple concepts including humans and objects, therefore hinders applications. In this work, we discard the single-entity assumption and introduce a novel framework that enforces strong, region-specific binding of conditions from modalities to each identity's spatiotemporal footprint. Given reference images of multiple concepts, our method could automatically infer layout information by leveraging a mask predictor to match appearance cues between the denoised video and each reference appearance. Furthermore, we inject local audio condition into its corresponding region to ensure layout-aligned modality matching in a iterative manner. This design enables the high-quality generation of controllable multi-concept human-centric videos. Empirical results and ablation studies validate the effectiveness of our explicit layout control for multi-modal conditions compared to implicit counterparts and other existing methods.

CGVQM+D: Computer Graphics Video Quality Metric and Dataset

While existing video and image quality datasets have extensively studied natural videos and traditional distortions, the perception of synthetic content and modern rendering artifacts remains underexplored. We present a novel video quality dataset focused on distortions introduced by advanced rendering techniques, including neural supersampling, novel-view synthesis, path tracing, neural denoising, frame interpolation, and variable rate shading. Our evaluations show that existing full-reference quality metrics perform sub-optimally on these distortions, with a maximum Pearson correlation of 0.78. Additionally, we find that the feature space of pre-trained 3D CNNs aligns strongly with human perception of visual quality. We propose CGVQM, a full-reference video quality metric that significantly outperforms existing metrics while generating both per-pixel error maps and global quality scores. Our dataset and metric implementation is available at https://github.com/IntelLabs/CGVQM.

StgcDiff: Spatial-Temporal Graph Condition Diffusion for Sign Language Transition Generation

Sign language transition generation seeks to convert discrete sign language segments into continuous sign videos by synthesizing smooth transitions. However,most existing methods merely concatenate isolated signs, resulting in poor visual coherence and semantic accuracy in the generated videos. Unlike textual languages,sign language is inherently rich in spatial-temporal cues, making it more complex to model. To address this,we propose StgcDiff, a graph-based conditional diffusion framework that generates smooth transitions between discrete signs by capturing the unique spatial-temporal dependencies of sign language. Specifically, we first train an encoder-decoder architecture to learn a structure-aware representation of spatial-temporal skeleton sequences. Next, we optimize a diffusion denoiser conditioned on the representations learned by the pre-trained encoder, which is tasked with predicting transition frames from noise. Additionally, we design the Sign-GCN module as the key component in our framework, which effectively models the spatial-temporal features. Extensive experiments conducted on the PHOENIX14T, USTC-CSL100,and USTC-SLR500 datasets demonstrate the superior performance of our method.