Chengyuan Wang

Yufeng Cui, Honghao Chen, Haoge Deng et al.

We introduce Emu3.5, a large-scale multimodal world model that natively predicts the next state across vision and language. Emu3.5 is pre-trained end-to-end with a unified next-token prediction objective on a corpus of vision-language interleaved data containing over 10 trillion tokens, primarily derived from sequential frames and transcripts of internet videos. The model naturally accepts interleaved vision-language inputs and generates interleaved vision-language outputs. Emu3.5 is further post-trained with large-scale reinforcement learning to enhance multimodal reasoning and generation. To improve inference efficiency, we propose Discrete Diffusion Adaptation (DiDA), which converts token-by-token decoding into bidirectional parallel prediction, accelerating per-image inference by about 20x without sacrificing performance. Emu3.5 exhibits strong native multimodal capabilities, including long-horizon vision-language generation, any-to-image (X2I) generation, and complex text-rich image generation. It also exhibits generalizable world-modeling abilities, enabling spatiotemporally consistent world exploration and open-world embodied manipulation across diverse scenarios and tasks. For comparison, Emu3.5 achieves performance comparable to Gemini 2.5 Flash Image (Nano Banana) on image generation and editing tasks and demonstrates superior results on a suite of interleaved generation tasks. We open-source Emu3.5 at https://github.com/baaivision/Emu3.5 to support community research.

Jiaqi Wang, Haoge Deng, Ting Pan et al.

Uniform Discrete Diffusion Model (UDM) has recently emerged as a promising paradigm for discrete generative modeling; however, its integration with reinforcement learning remains largely unexplored. We observe that naively applying GRPO to UDM leads to training instability and marginal performance gains. To address this, we propose UDM-GRPO, the first framework to integrate UDM with RL. Our method is guided by two key insights: (i) treating the final clean sample as the action provides more accurate and stable optimization signals; and (ii) reconstructing trajectories via the diffusion forward process better aligns probability paths with the pretraining distribution. Additionally, we introduce two strategies, Reduced-Step and CFG-Free, to further improve training efficiency. UDM-GRPO significantly improves base model performance across multiple T2I tasks. Notably, GenEval accuracy improves from $69\%$ to $96\%$ and PickScore increases from $20.46$ to $23.81$, achieving state-of-the-art performance in both continuous and discrete settings. On the OCR benchmark, accuracy rises from $8\%$ to $57\%$, further validating the generalization ability of our method. Code is available at https://github.com/Yovecent/UDM-GRPO.

Zhihui He, Chengyuan Wang, Shidong Yang et al.

Tooth arrangement is an essential step in the digital orthodontic planning process. Existing learning-based methods use hidden teeth features to directly regress teeth motions, which couples target pose perception and motion regression. It could lead to poor perceptions of three-dimensional transformation. They also ignore the possible overlaps or gaps between teeth of predicted dentition, which is generally unacceptable. Therefore, we propose DTAN, a differentiable collision-supervised tooth arrangement network, decoupling predicting tasks and feature modeling. DTAN decouples the tooth arrangement task by first predicting the hidden features of the final teeth poses and then using them to assist in regressing the motions between the beginning and target teeth. To learn the hidden features better, DTAN also decouples the teeth-hidden features into geometric and positional features, which are further supervised by feature consistency constraints. Furthermore, we propose a novel differentiable collision loss function for point cloud data to constrain the related gestures between teeth, which can be easily extended to other 3D point cloud tasks. We propose an arch-width guided tooth arrangement network, named C-DTAN, to make the results controllable. We construct three different tooth arrangement datasets and achieve drastically improved performance on accuracy and speed compared with existing methods.

Xianchao Wu, Chengyuan Wang, Qinying Lei

Current state-of-the-art AI based classical music creation algorithms such as Music Transformer are trained by employing single sequence of notes with time-shifts. The major drawback of absolute time interval expression is the difficulty of similarity computing of notes that share the same note value yet different tempos, in one or among MIDI files. In addition, the usage of single sequence restricts the model to separately and effectively learn music information such as harmony and rhythm. In this paper, we propose a framework with two novel methods to respectively track these two shortages, one is the construction of time-valued note sequences that liberate note values from tempos and the other is the separated usage of four sequences, namely, former note on to current note on, note on to note off, pitch, and velocity, for jointly training of four Transformer-XL networks. Through training on a 23-hour piano MIDI dataset, our framework generates significantly better and hour-level longer music than three state-of-the-art baselines, namely Music Transformer, DeepJ, and single sequence-based Transformer-XL, evaluated automatically and manually.

Zhang Chen, Anpei Chen, Guli Zhang et al.

We present a novel Relightable Neural Renderer (RNR) for simultaneous view synthesis and relighting using multi-view image inputs. Existing neural rendering (NR) does not explicitly model the physical rendering process and hence has limited capabilities on relighting. RNR instead models image formation in terms of environment lighting, object intrinsic attributes, and light transport function (LTF), each corresponding to a learnable component. In particular, the incorporation of a physically based rendering process not only enables relighting but also improves the quality of view synthesis. Comprehensive experiments on synthetic and real data show that RNR provides a practical and effective solution for conducting free-viewpoint relighting.