arXiv:2609.23658v1 Announce Type: cross
Abstract: Despite impressive visual quality, state-of-the-art video diffusion models often generate content that violates real-world physical laws. While exist...
By Yueyan Li, Haibo Wang, Caixia Yuan, Xiaojie Wang
CleanVideo introduces a selective erasure framework for text-to-video diffusion models, addressing the challenge of removing undesired visual concepts from videos. The method uses a low-dimensional subspace intervention guided by a tri-modal gating mechanism that jointly considers spatiotemporal visual features, timestep signals, and textual semantics to decide where, when, and whether to intervene. Experiments on three video diffusion models demonstrate that CleanVideo effectively erases target concepts while preserving visual fidelity, temporal coherence, and outperforming existing baselines in both frame-level and video-level evaluations, even under concept-recovery attacks.
By Junchi Liao, Hongji Li, Wenrui Zhou, Lijie Hu
arXiv:2607. 14194v1 Announce Type: cross Abstract: Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos, but controllably removing a target concept from a generator remains difficult.
By Wenxuan Chen, Wenjie Feng
EraseSAE introduces a surgical concept erasure method for text-to-video diffusion models, using sparse autoencoders to decompose activations into interpretable, monosemantic features. The framework employs a contrastive attribution mechanism to isolate concept-specific kernels and applies timestep-resolved masks during inference to remove target concepts while preserving unrelated content. Experiments show that EraseSAE achieves precise, robust concept removal with minimal quality loss, outperforming existing methods.
EraseSAE is a framework for surgical concept erasure in text-to-video diffusion models. It uses a Partitioned Convolutional Sparse Autoencoder to decompose activations into interpretable sparse features, a contrastive attribution mechanism to isolate concept-specific kernels, and timestep‑resolved masks to confine erasure to active regions. Experiments show precise removal with minimal quality loss, outperforming existing methods.
By Xinghao Wang, Dong Li, Wei Yu, Yingwei Pan, Tao Gong, Qi Chu, Nenghai Yu, Ting Yao
BiMoGen introduces a unified masked discrete diffusion framework for bidirectional motion‑text generation, addressing limitations of autoregressive models by allowing iterative bidirectional prediction. The method employs a two‑stage training strategy—Decoupled Uni‑ and Cross‑Modal Training followed by supervised fine‑tuning—to establish cross‑modal correspondence and specialize for bidirectional generation. To mitigate train‑test mismatch, Generation‑Aware Self‑Correction exposes the model to its own predictions during training and applies correction passes early in sampling to refine unmasked tokens, achieving competitive results on HumanML3D and KIT‑ML.
By Wanjiang Weng, Yongliang Wu, Xiaofeng Tan, Xingyu Zhu, Wenbo Zhu, Hongsong Wang
FOMO is a training‑based selective video unlearning method that prioritizes preserving the original scene while removing targeted concepts. It localizes concept‑related representations for modification and employs a preservation mechanism that maintains non‑target scene information without auxiliary data. The approach extends to motion unlearning, enabling removal of concepts defined by temporal behavior, and achieves a strong balance between concept removal and scene preservation.
By {\L}ukasz Rudnik, Agnieszka Polowczyk, Alicja Polowczyk, Przemys{\l}aw Spurek
The paper introduces Factor-Aware Predictive Learning (FAPL) to improve cinematic video generation by effectively leveraging multimodal large language model (MLLM) latents. FAPL trains a Video Diffusion Transformer (DiT) in three stages: grounding with frozen MLLM latents, factor-span replacement for factor-aware control, and temporal inference with limited-input representations. Experiments show FAPL outperforms baselines on sparse-to-video, factor-aware editing, and multi-shot generation tasks.
By Songlin Yang, Zhe Wang, Xiaotong Zhao, Xi Wang, Ran Zhang, Shuai Li, Taiyi Wu, Songlian Li, Eric Liu, Alan Zhao, Anyi Rao
Video-Mirai is a training-only method that improves autoregressive video diffusion models by adding foresight. It uses a frozen foresight encoder to read completed rollouts and a lightweight predictor to distill future-informed targets into causal states, without altering the generator’s inference process. The approach raises the 5‑second VBench Total Score from 83.8 to 84.6 and improves subject and background consistency on 30‑second rollouts.
By Yonghao Yu, Lang Huang, Runyi Li, Zerun Wang, Toshihiko Yamasaki
MT‑WAM enhances the Fast‑WAM framework by adding complementary supervision for future 2‑D point trajectories and visual features while keeping the original training objectives. A lightweight dual‑stream branch and structured attention mask isolate motion‑specific processing, and motion‑stream tokens provide additional dynamics cues to the action expert. During inference, MT‑WAM skips future‑video prediction, using cached video and motion information to achieve higher success rates on LIBERO, LIBERO‑Plus, RoboTwin 2.0 Clean2Rand, and several real‑world tasks.
By Yiguang Yang, Jiankun Peng, Xiaoming Wang, Yiran Zhang, Zhibo Fang
BiMoGen introduces a unified masked discrete diffusion framework for bidirectional motion‑text generation, addressing the limitations of autoregressive models in capturing bidirectional dependencies between language and motion. The approach employs a two‑stage training strategy—decoupled uni‑ and cross‑modal pretraining followed by supervised fine‑tuning—to establish robust cross‑modal correspondence, and incorporates Generation‑Aware Self‑Correction to mitigate error propagation during inference. Experiments on HumanML3D and KIT‑ML show competitive performance on both text‑to‑motion and motion‑to‑text tasks, demonstrating the effectiveness of the proposed training and correction mechanisms.
The paper introduces Factor-Aware Predictive Learning (FAPL) to improve cinematic video generation by effectively leveraging multimodal large language model (MLLM) latents. FAPL trains a Video Diffusion Transformer (DiT) in three stages: grounding the DiT with frozen MLLM latents, enforcing factor-aware control through factor-span replacement, and optimizing temporal inference with limited-input representations while preserving full-observation reconstruction. Experiments show that FAPL outperforms baselines on sparse-to-video, factor-aware editing, and multi-shot generation tasks, demonstrating enhanced MLLM feature use, temporal control, and factor coordination.