Visual proxies decouple explicit world-state simulation from photorealistic video synthesis, offering a controllable interface for both offline clip creation and real-time interactive streaming. Yet a practical renderer must address two challenges. Upstream proxies vary from coarse layouts to fine articulated geometry depending on the interaction, and streaming inference must operate in a few causal steps without accumulating drift across self-generated history. We present Invoker-Render, pairing a bidirectional creation model with a causal four-step streaming model under a shared conditioning interface over structural proxies, an appearance reference, and language. Trained on an aligned approximately 200-hour multi-granularity corpus, Invoker-Render honors fine geometric constraints when supplied while synthesizing plausible local motion under coarse proxies, enabling runtime selection of input detail. To bridge bidirectional creation and real-time streaming without compounding step compression and rollout drift, we introduce Progressive Forcing (ProForcing). ProForcing first trains a causal predictor and distills it into a four-step renderer using histories derived from training videos. Training then shifts to continuous self-generated rollouts. A pipelined serving stack supports efficient streaming inference. Experiments show versatile control across proxy families, granularities, and camera paths, consistent gains over recent streaming baselines, and stable 30-minute racing rollouts that preserve scene and subject appearance.
Figures & tables
Figure 2: Source video frames and aligned structural conditions. (a) Public datasets pair RGB frames with Magpie white-box geometry, CARLA depth, OmniWorld-Game depth, and Virtual KITTI 2 semantic layouts [ 67 , 11 , 76 , 3 ] . (b) Self-collected GTA V and Cyberpunk 2077 examples pair each RGB source frame with matched coarse and fine proxies of the same event, where coarse proxies place canonical root-driven bodies and fine proxies track articulated limbs.
Figure 3: Capture and multi-granularity proxy construction pipeline. Synchronized gameplay capture and 3D scene reconstruction yield aligned appearance targets, geometry, entity dynamics, metadata, and text annotations. From a shared scene state, the pipeline renders matched coarse and fine structural proxies for each event.
Figure 4: Comparison of ProForcing with representative causal video training recipes. Rows align five training pipelines across four functional stages from causal adaptation (Stage 1) to long-sequence tuning (Stage 4). Self Forcing initializes from a bidirectional ODE teacher [ 18 ] , Causal Forcing/++ and Causal-rCM initialize from causal teachers [ 78 , 72 , 74 ] , and LongLive appends streaming long tuning [ 61 ] . ProForcing adapts block-causal attention via TF and RF (Stage 1), distills four-step RF-DMD with GAN supervision under data-anchored histories (Stage 2), and transfers the student, fake score model, discriminator, optimizer states, and student exponential moving average (EMA) directly into continuous long-rollout SF-DMD (Stages 3–4).
Figure 5: Streaming inference schedule on an eight-GPU colocated pool. Each ordinary continuation chunk runs four DiT passes under Ulysses-8 context parallelism. Pass 1 fuses deferred clean history Hn−1 , the current proxy condition Cn , and the first denoising step D1 ; Passes 2–4 finish denoising, and only clean H+C key/value states are committed. During the codec interval between DiT windows, proxy encoding for chunk n+1 overlaps with causal video decoding ( ×4 ) for chunk n .
Figure 7: Head-to-head overall preference against streaming baselines. Bars show pairwise win, tie, and loss proportions of Invoker-Render (Causal, 4-step) across 140 blinded test comparisons per baseline, with the dashed vertical line serving as a 50% reference line.
Figure 8: Qualitative comparison of causal streaming renderers on held-out rollouts. Each column shows one test sequence at a matched late-rollout timestamp (11 s to 18 s). The top shaded rows display the 0 s appearance reference and the synchronized input proxy, followed by frames from other works and Invoker-Render (outlined at bottom), which maintains both structural fidelity and reference appearance at the shown late-rollout timestamps.
Figure 11: Long-horizon causal rendering on two racing sequences. Each case shows six matched timestamps, read from left to right. Rows contain the input proxy, two prior works (A and B), and Invoker-Render. The upper and lower cases extend to 298 s and 299 s, respectively, showing vehicle appearance and trajectory adherence over approximately five minutes.
We introduce StreamForce, a streaming video generation framework that enables physically grounded control through continuous force inputs. Unlike prior video models that train separate models for different force types, assume fixed forces, or rely on non-causal processing, StreamForce is a causal and unified model that responds instantly and coherently to both local and global, time-varying forces. To achieve this, we design a unified force representation as a control signal and develop a distillation pipeline for force-controllable video generation. Our model combines autoregressive efficiency with force responsiveness, sustaining stable photometric and dynamic realism. StreamForce runs at up to 16.6 FPS on a single GPU, achieving state-of-the-art performance in both force adherence and motion realism. Project website: https://neu-vi.github.io/StreamForce/
Hanhui Wang, Yiming Xie, Haiwen Feng +3
1Northeastern University · 2Impossible Research · University of California, Berkeley +1
We present Astronex-World 1.0, an open controllable video world-model foundation. Given a text prompt (text-to-video) or an initial observation (image-to-video), the model predicts future visual states under frame-aligned camera trajectories, continuous actions, and an embodiment identifier, and accepts text events inserted at a specified position of a rollout. The family provides a bidirectional model for full-context generation and a causal model with block-causal attention and cross-block KV caching for persistent generation, both built on the Wan2.2-TI2V-5B prior. PRoPE injects camera intrinsics and extrinsics, while a 64-dimensional action stream modulates every Transformer layer. A five-stage training path develops bidirectional camera and action control, converts the backbone to block-causal generation, distills a few-step student, restores mixed-domain dynamics, and applies asymmetric DMD/DMD2 distribution matching. The causal model generates 832x480 video at 24 fps. All five training stages run on two NVIDIA L20 48 GB GPUs, and the causal model streams in real time on one. It scores 73.5 on WBench Navi and 70.0 on WBench Full. On Full, this 5B model is above the 13.6B LongCat-Video and the 14B Helios, within one point of the 22B LTX-2.3, and above YUME 1.5, which is post-trained from the same 5B prior on NVIDIA A100 GPUs. The reserved action input and output interfaces allow post-training for embodied intelligence and autonomous driving.
Xin Zhou, Cong Miao
Astronex Robotics · Nanjing University of Information Science and Technology
Camera-controlled video-to-video (V2V) generation enables dynamic viewpoint synthesis from monocular footage, holding immense potential for interactive filmmaking and live broadcasting. However, existing implicit synthesis methods fundamentally rely on non-causal, full-sequence processing and rigid prefix-style temporal concatenation. This architectural paradigm mandates bidirectional attention, resulting in prohibitive computational latency, quadratic complexity scaling, and inherent incompatibility with real-time streaming or variable-length inputs. To overcome these limitations, we introduce RealCam, a novel autoregressive framework for interactive, real-time camera-controlled V2V generation. We first design a high-fidelity teacher model grounded in a Cross-frame In-context Learning paradigm. By interleaving source and target frames into synchronized contextual pairs, our design inherently enables length-agnostic generalization and naturally facilitates causal adaptation, breaking the rigid prefix bottleneck. We then distill this teacher into a few-step causal student via Self-Forcing with Distribution Matching Distillation, enabling efficient, on-the-fly streaming synthesis. Furthermore, to mitigate severe loop inconsistency in closed-loop trajectories, we propose Loop-Closed Data Augmentation (LoopAug), a novel paradigm that synthesizes globally consistent loop sequences from existing multiview datasets. Extensive experiments demonstrate that RealCam achieves state-of-the-art visual fidelity and temporal consistency while enabling truly interactive camera control with orders-of-magnitude faster inference than existing paradigms. Our project page is at https://xyc-fly.github.io/RealCam/.
Youcan Xu, Jiaxin Shi, Zhen Wang +5
Zhejiang University, China · Xmax.AI Ltd., China · Hong Kong University of Science and Technology, China