Search arXiv⌕ Search

arXiv subjects

Songchun Zhang

Publications and source records attributed to Songchun Zhang.

18 recordsLinked to original sources
arXiv2026

Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout

Autoregressive (AR) video diffusion models have shown great potential in real-time video generation. Recent methods distill pretrained bidirectional video diffusion models into causal AR students through Distribution Matching Distillation (DMD), but the generated videos often suffer from over-saturation and over-smoothing issues, resulting in limited visual quality and realism. The key contributing factor is the mode-seeking behavior of the reverse KL objective in DMD, which can cause the student distribution to collapse onto only a few modes of the teacher distribution. To address this, we propose Mask Forcing, a Dual-Noise Masking Rollout strategy that perturbs the AR student self-rollout to mitigate mode collapse induced by reverse-KL mode seeking. The core idea is to inject cleaner signals into noisy rollout inputs via random masks along spatial and temporal axes during the self-rollout process of AR diffusion distillation. Such perturbations encourage the student rollouts to explore more regions of the teacher distribution, allowing DMD to provide learning signals beyond the modes already covered by the student. Moreover, the cleaner tokens act as denoising guidance for other noisier tokens, improving the intermediate rollout predictions and reducing error accumulation. Extensive experiments demonstrate that our method improves multiple AR video diffusion distillation methods with higher visual quality efficiently, without incorporating real video data or additional post-training stages.

cs.CV↗
arXiv2026

Building Pretraining Data for World Models: An Unreal Engine-Based Pipeline for Action-Conditioned Video Generation

Action-conditioned video models require large-scale visual data paired with control signals that are temporally aligned with the resulting scene transitions. Such supervision is difficult to obtain from ordinary real-world video because the actions that caused each visual change are typically unknown. We present a large-scale synthetic data production pipeline built on Unreal Engine for generating action-conditioned, multi-view video. To accommodate the different execution requirements of real-time physics and high-quality offline rendering, the pipeline executes trajectory generation and final rendering in two stages: Stage I runs real physics in PIE and records per-frame character states, control inputs, and camera states into an intermediate trajectory representation; Stage II replays those trajectories in a new engine process and renders them offline with Movie Render Queue (MRQ). Around this core, we develop a distributed production system with cache-aware task partitioning, node-local slot scheduling, automated scene screening, aesthetic and luminance filtering, partial-output recovery, asynchronous upload, and continuous cluster health monitoring. The production cluster contains 25 servers with eight NVIDIA RTX 5090 GPUs per server. From 2,384 asset packs, 429 levels were retained for production together with a pool of 40 humanoid characters. The pipeline has produced 2,691 hours of 1080p video and 6,076 hours of 720p video. We describe the system architecture, the implementation decisions that emerged from production failures, and the limitations of using perceptual quality proxies for world-model data curation. The pipeline described in this report constitutes the Unreal Engine synthetic-data production component used in EchoWM.

cs.CV↗
arXiv2026

Long-Horizon Audio-Visual Generation for Persistent Stories and Interactive Worlds

Video generation is progressing beyond isolated clips toward long-form narratives and interactive worlds, requiring models to preserve identities, follow user controls, and remain stable over extended rollouts. We present JoyAI-Echo-1.5, a unified audio-visual generation system with two purpose-built variants. The long-video variant introduces composable cross-shot memory that aggregates visual evidence across multiple prior shots and speaker cues derived from speech-filtered full-shot audio, enabling persistent character appearance and voice identity across flexible combinations of text, image, and memory conditioning. The world-model variant converts heterogeneous navigation inputs into calibrated metric 6-DoF camera trajectories and injects them through a geometry-aware conditioning pathway, enabling controller-agnostic interaction across flexible viewpoints. To support efficient long-horizon generation, we transform a bidirectional audio-visual backbone into a causal few-step generator using progressive teacher forcing and short- and long-horizon Self-Gradient Forcing on self-generated rollouts. Experiments demonstrate strong performance in both settings. JoyAI-Echo-1.5 achieves improvements over existing long-video baselines in cross-shot consistency, visual quality, text alignment, and speech fidelity. Its world-model variant ranks first on WBench, with an average score of 81.7, and achieves leading visual quality and long-horizon persistence on SANA-WM-Bench. Together, these results indicate that memory, geometric control, and rollout-aware training provide a practical foundation for generating coherent stories and continuously evolving interactive worlds. Project page: https://echo-team-joy-future-academy-jd.github.io/Echo-1.5-Page/.

cs.CV↗
arXiv2026

ShotVerse: Advancing Cinematic Camera Control for Text-Driven Multi-Shot Video Creation

Text-driven video generation has democratized film creation, but camera control in cinematic multi-shot scenarios remains a significant block. Implicit textual prompts lack precision, while explicit trajectory conditioning imposes prohibitive manual overhead and often triggers execution failures in current models. To overcome this bottleneck, we propose a data-centric paradigm shift, positing that aligned (Caption, Trajectory, Video) triplets form an inherent joint distribution that can connect automated plotting and precise execution. Guided by this insight, we present ShotVerse, a ``Plan-then-Control'' framework that decouples generation into two collaborative agents: a VLM (Vision-Language Model)-based Planner that leverages spatial priors to obtain cinematic, globally aligned trajectories from text, and a Controller that renders these trajectories into multi-shot video content via a camera adapter. Central to our approach is the construction of a data foundation: we design an automated multi-shot camera calibration pipeline aligns disjoint single-shot trajectories into a unified global coordinate system. This facilitates the curation of ShotVerse-Bench, a high-fidelity cinematic dataset with a three-track evaluation protocol that serves as the bedrock for our framework. Extensive experiments demonstrate that ShotVerse effectively bridges the gap between unreliable textual control and labor-intensive manual plotting, achieving superior cinematic aesthetics and generating multi-shot videos that are both camera-accurate and cross-shot consistent.

cs.CV↗
arXiv2026

EchoWM: Open and Enterable Omnimodal World Models

We present EchoWM, an omnimodal world model for enterable generative media that responds to continuous navigation while jointly generating 720p video, environmental sound, music and speech. We organize interaction around camera intent: in first-person scenes, it specifies observer motion, while in third-person scenes, camera--character dynamics are learned from data without view-specific controllers. Discrete commands and continuous poses are mapped to a shared metric-scale relative 6-DoF trajectory, with dataset-level calibration preserving motion magnitude across heterogeneous data. To jointly learn audio-visual generation and trajectory control, we construct a complementary data engine and adopt progressive training followed by autoregressive post-training for long-horizon generation. Extensive evaluations show that \model achieves strong trajectory following and high visual quality on public world-model benchmarks, supporting both first- and third-person interaction across varied subjects, and maintaining synchronized environmental sound and speech over long-horizon generation.

cs.CV↗
arXiv2026

LiveLight: Real-time Streaming Video Relighting with Interactive Control

We present LiveLight, the first diffusion-based framework for real-time streaming video relighting with interactive 3D lighting control. Achieving this is non-trivial, as it requires overcoming three critical challenges: effectively injecting dynamic 3D lighting into a diffusion model, maintaining high-fidelity generation under an extremely low NFE (Number of Function Evaluations) budget for real-time speed, and facilitating continuous streaming for interactive control. To address these pain points, we propose three key designs. First, for accurate lighting injection, we propose a lightweight adapter that feeds Multi-Plane Light Irradiance (MPLI) conditions-depth-aware irradiance maps encoding 3D lighting geometry-directly into the diffusion backbone. Second, to prevent rendering quality degradation at low NFEs towards real-time distillation, we introduce a geometry-guided feedback branch. This training-time constraint leverages a frozen geometry estimator to enforce depth- and normal-consistent relighting, ensuring geometrically plausible shading without adding inference overhead. Finally, to enable streaming interaction, we develop a progressive rolling-window strategy that maintains a denoising ladder of latent chunks at varying noise levels. By propagating intermediate states, this strategy guarantees temporal coherence and supports arbitrarily long video relighting with per-frame reference refresh. Extensive experiments on real-world and synthetic benchmarks demonstrate that LiveLight achieves state-of-the-art relighting quality while running at real-time speed, significantly outperforming offline baselines in temporal stability, lighting controllability, and user preference. To foster real-time interactive relighting research, we will publicly release our models, training data, and synthetic data generator.

cs.CV↗
arXiv2026

FlexComposer: Unified Video Compositing from Images to Dynamic Footage with Flexible Trajectory Control

Generative video compositing, which involves inserting external assets seamlessly into existing video sequences, is essential for content creation and visual effects. However, existing approaches suffer from a control-fidelity trade-off: they either hallucinate motion from static images, failing to preserve the dynamics of pre-animated assets, or lack fine-grained spatial control for precise asset placement along user-defined trajectories. We propose FlexComposer, a unified framework that standardizes video compositing as a trajectory-guided conditional generation task, enabling the seamless integration of both static images and dynamic footage. Our approach introduces three key designs: (1) a Unified Canonical Foreground Representation that decouples an object's intrinsic motion from its global displacement, standardizing heterogeneous inputs into a stabilized, centered latent space; (2) a Spatial-Aware Latent Injection strategy that exploits the translation equivariance of VAE latent spaces to transport canonical features onto target trajectories via a parameter-free mechanism; and (3) a Hybrid Dataset and Synthetic-to-Real Curriculum that synergizes procedural simulation, real-world cinematic footage, and generative data to implicitly learn physically plausible illumination and shadow harmonization. This unified design handles diverse inputs from product photos to dynamic subjects achieving high-fidelity motion control and environmental integration without the need for explicit 3D reconstruction or auxiliary learnable adapters. Extensive experiments demonstrate that FlexComposer outperforms state-of-the-art methods in visual quality, temporal consistency, and trajectory adherence.

cs.CV↗
arXiv2026

Self Gradient Forcing: Native Long Video Extrapolation

Recent autoregressive video diffusion methods are increasingly built upon Self Forcing, where the student is trained on histories produced by its own rollout rather than ground-truth video contexts. This reduces exposure bias, but the historical key-value cache is still used by future frames only as frozen rollout state. As a result, future losses cannot supervise how earlier generated latents should be written into more useful keys and values for later video-latent generation. We call this the historical context-gradient gap. We propose Self Gradient Forcing (SGF), a two-pass training strategy that restores this missing supervision signal without backpropagating through the full serial rollout. Pass 1 performs a no-gradient autoregressive rollout matching inference and, at a sampled denoising exit step, records both the self-generated context and the noisy latents fed to the model. Pass 2 performs parallel context-gradient reconstruction for the recorded exit step. The generated context is used as stop-gradient clean-latent input, while the model recomputes the context KV representations and future-to-context causal attention. Thus, SGF provides the missing memory-writing supervision within the native autoregressive training objective, using losses on future video latents to train the model to encode context into more effective causal memory. Across extensive long-horizon frame-wise and chunk-wise experiments under different initializations, SGF achieves stronger native long-video extrapolation than Self Forcing, especially in subject identity, background/layout consistency, and temporal stability. Remarkably, using only a 5-second training window, SGF can extrapolate to videos lasting several minutes. Code and models will be released to advance research on autoregressive video generation.

cs.CV↗
arXiv2026

GeoEdit: Geometry-Aware Object Editing via Dual-Branch Denoising

Precisely manipulating objects in a single photograph (translation, rotation, scaling) while obeying 3D physical constraints remains unsolved for diffusion-based editors. Current 2D methods lack spatial awareness and produce perspective violations. Forcing structural proxies into the latent space also disrupts variance homogeneity, and the resulting self-attention leakage leads to ghosting and background blur. The core difficulty is asymmetric: the relocated object must follow a rigid geometry, yet the uncovered background needs freedom to synthesize plausible content. We present GeoEdit, a training-free Lift-Manipulate-Render-Denoise pipeline that satisfies both constraints. We decouple scene and object in 3D, align them through point correspondence, and render a geometry-aligned proxy with a structural depth map. A Dual-Branch Denoising stage then refines this proxy: a video diffusion backbone preserves object identity, while 3D constraints are injected into the foreground within a narrow denoising window at matching noise variance (variance-homogeneous injection). The background denoises freely. Because the injected signal matches the native latent statistics, self-attention stays undisturbed. We also introduce GeoEditBench, a pose-aware benchmark covering object translation, object rotation, and camera movement with pose-aware evaluation metrics. Experiments confirm consistent gains in geometric accuracy, identity fidelity, and background quality. Our codes are available at https://github.com/Heey731/GeoEdit.

cs.CV↗
arXiv2026

Ultra Flash: Scaling Real-Time Streaming Video Generation to High Resolutions

While recent autoregressive video diffusion models achieve remarkable streaming quality, they remain confined to low resolutions (e.g., 480P), leaving efficient, scalable, real-time high-resolution video generation a fundamental open challenge. To bridge this gap, we present Ultra Flash, a cascaded streaming framework capable of real-time high-resolution video generation. Ultra Flash achieves ~30 FPS at 1K resolution and ~18 FPS at 2K resolution on a single GPU through three key contributions: (1) an architecture-preserving T2V-to-TV2V super-resolution training paradigm coupled with an AIGC-oriented data degradation pipeline that effectively preserves the generative capability of the base model, enabling enhanced high-resolution detail when cascaded after mainstream low-resolution generative models; (2) a causal streaming latent upsampler paired with a high-resolution decoder, which enhances spatiotemporal coherence while enabling efficient latent spatial scaling and precise high-resolution decoding with negligible computational overhead; and (3) a cascade high-resolution streaming video generation optimization scheme that first performs hybrid-reward-enhanced sparse causalization and single-step distillation of the super-resolution model, then introduces cascaded streaming self-forcing preference optimization with dynamic cache management, jointly enhancing overall coherence, improving quality, and enabling real-time high-resolution streaming video generation. Extensive experiments demonstrate that Ultra Flash reliably produces ultra-high-resolution streaming video while maintaining state-of-the-art visual quality and superior efficiency. Project Page: https://xin1u.github.io/UltraFlash/

cs.CV↗
arXiv2026

Echo-Memory: A Controlled Study of Memory in Action World Models

We present \textbf{Echo-Memory}, a controlled study of memory mechanisms in action-conditioned world models. These models generate multi-segment videos from a first frame, text prompt, and camera-action sequence, but their central failure is often memory rather than local image synthesis: after the camera leaves and returns, the scene or salient object may silently change. Existing memory designs are hard to compare because gains are entangled with backbone, training, retrieval, and evaluation differences. Echo-Memory fixes the action-to-video interface and varies only how history is stored and read by the generator. Under a shared video diffusion backbone, optimizer, camera-action representation, sampler, and evaluation pipeline, we compare raw context, compression-based memory, spatial summaries with different read-out paths, and state-space recurrence. This matched matrix separates four otherwise conflated axes: \emph{capacity}, \emph{compression}, \emph{read-out}, and \emph{recurrence}. We also evaluate memory through a three-branch protocol: replay quality, in-domain loop revisit, and open-domain return probes. The branches routinely disagree, showing that replay fidelity is not a sufficient proxy for remembering a world. Three findings follow. Raw context is a strong capacity baseline and improves open-domain return far more than it improves replay metrics. Compactness is not a free substitute for capacity: aggressive spatial and hybrid-compression memories lose the salient evidence needed for return. Finally, block-wise state-space recurrence is the strongest open-domain return mechanism in our matrix, showing that the structure of implicit memory matters as much as the decision to use it. These results provide a compact protocol for studying memory in action world models beyond isolated replay metrics.

cs.CV↗
arXiv2026

Echo-Infinity: Learning Evolving Memory for Real-Time Infinite Video Generation

We present Echo Infinity, an autoregressive (AR) framework towards real-time infinite video generation that employs a learnable evolving memory to dynamically filter, abstract, and compress any-length history at constant cost. Existing methods mainly curate memory with predefined KV-cache schedules, fixed-ratio heuristic compression, or inference-time RoPE adaptation. These designs inevitably lose historical information and amplify compounding errors due to their limited cache window and ignorance of autoregressive generation noise. Inspired by human memory consolidation, Echo-Infinity replaces handcrafted memory curation with learnable Memory Query, which are updated by attention and a gating mechanism when past frames are evicted from the local window. The queries are optimized end-to-end with the video diffusion transformers (DiTs), forming an evolving memory that supports arbitrary compression ratios with constant computation independent of video length. They also act as a generalizable generation prior, improving quality even when only the optimized initial state is used. We further introduce Unified Relative RoPE Recipe, which anchors the sink frames to start from id 0 and lets the newest frame id grow at most to the DiTs' pretrained maximum temporal RoPE id throughout training and inference, freeing the model from the finite RoPE constraint and closing the train-test RoPE extrapolation gap. In long and short video generation, Echo-Infinity achieves state-of-the-art performance, and, to our knowledge, demonstrates promising 24-hour (>1.3 M frames) real-time rollouts for the first time, suggesting a practical path toward infinite video generation.

cs.MM↗
arXiv2026

Astrolabe: Steering Forward-Process Reinforcement Learning for Distilled Autoregressive Video Models

Distilled autoregressive (AR) video models enable efficient streaming generation but frequently misalign with human visual preferences. Existing reinforcement learning (RL) frameworks are not naturally suited to these architectures, typically requiring either expensive re-distillation or solver-coupled reverse-process optimization that introduces considerable memory and computational overhead. We present Astrolabe, an efficient online RL framework tailored for distilled AR models. To overcome existing bottlenecks, we introduce a forward-process RL formulation based on negative-aware fine-tuning. By contrasting positive and negative samples directly at inference endpoints, this approach establishes an implicit policy improvement direction without requiring reverse-process unrolling. To scale this alignment to long videos, we propose a streaming training scheme that generates sequences progressively via a rolling KV-cache, applying RL updates exclusively to local clip windows while conditioning on prior context to ensure long-range coherence. Finally, to mitigate reward hacking, we integrate a multi-reward objective stabilized by uncertainty-aware selective regularization and dynamic reference updates. Extensive experiments demonstrate that our method consistently enhances generation quality across multiple distilled AR video models, serving as a robust and scalable alignment solution.

cs.CV↗
arXiv2026

Composing Concepts from Images and Videos via Concept-prompt Binding

Visual concept composition, which aims to integrate different elements from images and videos into a single, coherent visual output, still falls short in accurately extracting complex concepts from visual inputs and flexibly combining concepts from both images and videos. We introduce Bind & Compose, a one-shot method that enables flexible visual concept composition by binding visual concepts with corresponding prompt tokens and composing the target prompt with bound tokens from various sources. It adopts a hierarchical binder structure for cross-attention conditioning in Diffusion Transformers to encode visual concepts into corresponding prompt tokens for accurate decomposition of complex visual concepts. To improve concept-token binding accuracy, we design a Diversify-and-Absorb Mechanism that uses an extra absorbent token to eliminate the impact of concept-irrelevant details when training with diversified prompts. To enhance the compatibility between image and video concepts, we present a Temporal Disentanglement Strategy that decouples the training process of video concepts into two stages with a dual-branch binder structure for temporal modeling. Evaluations demonstrate that our method achieves superior concept consistency, prompt fidelity, and motion quality over existing approaches, opening up new possibilities for visual creativity.

cs.CV↗
arXiv2025

SpatialCrafter: Unleashing the Imagination of Video Diffusion Models for Scene Reconstruction from Limited Observations

Novel view synthesis (NVS) boosts immersive experiences in computer vision and graphics. Existing techniques, though progressed, rely on dense multi-view observations, restricting their application. This work takes on the challenge of reconstructing photorealistic 3D scenes from sparse or single-view inputs. We introduce SpatialCrafter, a framework that leverages the rich knowledge in video diffusion models to generate plausible additional observations, thereby alleviating reconstruction ambiguity. Through a trainable camera encoder and an epipolar attention mechanism for explicit geometric constraints, we achieve precise camera control and 3D consistency, further reinforced by a unified scale estimation strategy to handle scale discrepancies across datasets. Furthermore, by integrating monocular depth priors with semantic features in the video latent space, our framework directly regresses 3D Gaussian primitives and efficiently processes long-sequence features using a hybrid network structure. Extensive experiments show our method enhances sparse view reconstruction and restores the realistic appearance of 3D scenes.

cs.CV↗
arXiv2024

Pragmatist: Multiview Conditional Diffusion Models for High-Fidelity 3D Reconstruction from Unposed Sparse Views

Inferring 3D structures from sparse, unposed observations is challenging due to its unconstrained nature. Recent methods propose to predict implicit representations directly from unposed inputs in a data-driven manner, achieving promising results. However, these methods do not utilize geometric priors and cannot hallucinate the appearance of unseen regions, thus making it challenging to reconstruct fine geometric and textural details. To tackle this challenge, our key idea is to reformulate this ill-posed problem as conditional novel view synthesis, aiming to generate complete observations from limited input views to facilitate reconstruction. With complete observations, the poses of the input views can be easily recovered and further used to optimize the reconstructed object. To this end, we propose a novel pipeline Pragmatist. First, we generate a complete observation of the object via a multiview conditional diffusion model. Then, we use a feed-forward large reconstruction model to obtain the reconstructed mesh. To further improve the reconstruction quality, we recover the poses of input views by inverting the obtained 3D representations and further optimize the texture using detailed input views. Unlike previous approaches, our pipeline improves reconstruction by efficiently leveraging unposed inputs and generative priors, circumventing the direct resolution of highly ill-posed problems. Extensive experiments show that our approach achieves promising performance in several benchmarks.

cs.CV↗
arXiv2023

Cross-Video Contextual Knowledge Exploration and Exploitation for Ambiguity Reduction in Weakly Supervised Temporal Action Localization

Weakly supervised temporal action localization (WSTAL) aims to localize actions in untrimmed videos using video-level labels. Despite recent advances, existing approaches mainly follow a localization-by-classification pipeline, generally processing each segment individually, thereby exploiting only limited contextual information. As a result, the model will lack a comprehensive understanding (e.g. appearance and temporal structure) of various action patterns, leading to ambiguity in classification learning and temporal localization. Our work addresses this from a novel perspective, by exploring and exploiting the cross-video contextual knowledge within the dataset to recover the dataset-level semantic structure of action instances via weak labels only, thereby indirectly improving the holistic understanding of fine-grained action patterns and alleviating the aforementioned ambiguities. Specifically, an end-to-end framework is proposed, including a Robust Memory-Guided Contrastive Learning (RMGCL) module and a Global Knowledge Summarization and Aggregation (GKSA) module. First, the RMGCL module explores the contrast and consistency of cross-video action features, assisting in learning more structured and compact embedding space, thus reducing ambiguity in classification learning. Further, the GKSA module is used to efficiently summarize and propagate the cross-video representative action knowledge in a learnable manner to promote holistic action patterns understanding, which in turn allows the generation of high-confidence pseudo-labels for self-learning, thus alleviating ambiguity in temporal localization. Extensive experiments on THUMOS14, ActivityNet1.3, and FineAction demonstrate that our method outperforms the state-of-the-art methods, and can be easily plugged into other WSTAL methods.

cs.CV↗
arXiv2023

Dyna-DepthFormer: Multi-frame Transformer for Self-Supervised Depth Estimation in Dynamic Scenes

Self-supervised methods have showed promising results on depth estimation task. However, previous methods estimate the target depth map and camera ego-motion simultaneously, underusing multi-frame correlation information and ignoring the motion of dynamic objects. In this paper, we propose a novel Dyna-Depthformer framework, which predicts scene depth and 3D motion field jointly and aggregates multi-frame information with transformer. Our contributions are two-fold. First, we leverage multi-view correlation through a series of self- and cross-attention layers in order to obtain enhanced depth feature representation. Specifically, we use the perspective transformation to acquire the initial reference point, and use deformable attention to reduce the computational cost. Second, we propose a warping-based Motion Network to estimate the motion field of dynamic objects without using semantic prior. To improve the motion field predictions, we propose an iterative optimization strategy, together with a sparsity-regularized loss. The entire pipeline achieves end-to-end self-supervised training by constructing a minimum reprojection loss. Extensive experiments on the KITTI and Cityscapes benchmarks demonstrate the effectiveness of our method and show that our method outperforms state-of-the-art algorithms.

cs.CV↗