What Semantics Survive the Connector? Diagnosing VLM-to-DiT Alignment in Video Editing

• URL: http://arxiv.org/abs/2605.20795v1
• Date: Wed, 20 May 2026 06:42:15 GMT
• Title: What Semantics Survive the Connector? Diagnosing VLM-to-DiT Alignment in Video Editing
• Abstract summary: Flow matching based video generative models rely on prepended Vision-Language Models to handle complex, instruction-based video editing.<n>The prevailing assumption is that a connector module can seamlessly align the VLM's rich multi-modal reasoning with the original text embedding space of DiTs.<n>We propose a controlled data processing pipeline based on video composition that results in TRACE-Edit, a diagnostic dataset focusing on relation-based editing.
• Score: 59.40855219373921
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Flow matching based video generative models have been increasingly relying on prepended Vision-Language Models (VLMs) to handle complex, instruction-based video editing. The prevailing assumption underlying this paradigm is that a connector module can seamlessly align the VLM's rich multi-modal reasoning with the original text embedding space of DiTs. However, we hypothesize that this alignment acts as a severe semantic bottleneck, degrading fine-grained structural variables. Verifying this is challenging, as end-to-end evaluations conflate alignment failures with generation errors, and natural datasets lack disentangled annotations. To rigorously investigate this, we propose a controlled data processing pipeline based on video composition that results in TRACE-Edit, a diagnostic dataset focusing on relation-based editing. Leveraging this dataset, we propose a comprehensive diagnostic protocol to analyze two important designs of meta-query and connector in the existing video editing models. Systematic evaluation of four representative model cases reveals that fine-grained structural semantics can be severely degraded during alignment. Our findings overturn the assumption of lossless semantic transfer, identifying the VLM-to-DiT alignment as a major bottleneck and providing a new diagnostic foundation for future multi-modal alignment architectures.

Related papers

• MLLM-Guided Semantic Correction for Text-to-Video Generation [66.7723669725808]
  We introduce a training-free, interpretable mid-generation correction framework that integrates multimodal large language model feedback.<n>We propose two key modules: a Semantic Assessment Supervisor that generates intermediate preview frames for semantic evaluations and deviation diagnostics, and a Semantic Modification Assistant that corrects semantic drift during inference.<n>Our method improves semantic alignment, visual fidelity, and temporal consistency without modifying model parameters.
  arXiv  Detail & Related papers  (2026-08-17T12:54:10Z)
• Sparse Visual Thought Circuits in Vision-Language Models [2.5754366051855837]
  We develop a causal pipeline to localize and test sparse visual thought circuits in Qwen3-VL-8B.<n>We train SAEs at this layer, construct task-selective sets via an explicit rule, and perform inference time scaling and ablation while quantifying accuracy and drift.
  arXiv  Detail & Related papers  (2026-03-26T06:24:36Z)
• SAMA: Factorized Semantic Anchoring and Motion Alignment for Instruction-Guided Video Editing [76.349958946335]
  We present SAMA (factorized Semantic Anchoring and Motion Alignment), a framework that factorizes video editing into semantic anchoring and motion modeling.<n>First, we introduce Semantic Anchoring, which establishes a reliable visual anchor by jointly predicting semantic tokens and video latents at sparse anchor frames.<n>Second, Motion Alignment pre-trains the same backbone on motion-centric video restoration pretext tasks.
  arXiv  Detail & Related papers  (2026-03-19T17:59:51Z)
• Advancing Adaptive Multi-Stage Video Anomaly Reasoning: A Benchmark Dataset and Method [96.63801368613177]
  We present a new task that elevates video anomaly analysis from descriptive understanding to structured, multi-stage reasoning.<n>We present a new dataset with 8,641 videos, totaling more than 50,000 samples, making it one of the largest datasets for video anomaly understanding.<n>Building upon the proposed task and dataset, we develop an end-to-end MLLM-based VAR model termed Vad-R1-Plus, which supports adaptive hierarchical reasoning and risk-aware decision making.
  arXiv  Detail & Related papers  (2026-01-15T08:09:04Z)
• UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits [43.59555184340113]
  We introduce a lightweight data pipeline that replaces multi-toolchains with an end-to-end model and a unified post-verification stage.<n>For scalable quality control, we train a 7B dual-task expert model, textbfQwen-Verify, for efficient failure detection and instruction recaptioning.<n>This pipeline yields textbfUnicEdit-10M, a 10M-scale dataset spanning diverse basic and complex editing tasks.
  arXiv  Detail & Related papers  (2025-12-01T17:45:44Z)
• ViRectify: A Challenging Benchmark for Video Reasoning Correction with Multimodal Large Language Models [23.37951284612929]
  We construct a dataset of over 30K instances spanning dynamic perception, scientific reasoning, and embodied decision-making domains.<n>In ViRectify, we challenge MLLMs to perform step-wise error identification and generate rationales with key video evidence grounding.<n>In addition, we propose the trajectory evidence-driven correction framework, comprising step-wise error trajectory and reward modeling on visual evidence-grounded correction.
  arXiv  Detail & Related papers  (2025-12-01T09:05:02Z)
• CtrlVDiff: Controllable Video Generation via Unified Multimodal Video Diffusion [62.04833878126661]
  We tackle the dual challenges of video understanding and controllable video generation within a unified diffusion framework.<n>We propose CtrlVDiff, a unified diffusion model trained with a Hybrid Modality Control Strategy (HMCS) that routes and fuses features from depth, normals, segmentation, edges, and graphics-based intrinsics (albedo, roughness, metallic)<n>Across understanding and generation benchmarks, CtrlVDiff delivers superior controllability and fidelity, enabling layer-wise edits (relighting, material adjustment, object insertion) and surpassing state-of-the-art baselines while remaining robust when some modalities are unavailable.
  arXiv  Detail & Related papers  (2025-11-26T07:27:11Z)
• MoA-VR: A Mixture-of-Agents System Towards All-in-One Video Restoration [62.929029990341796]
  Real-world videos often suffer from complex degradations, such as noise, compression artifacts, and low-light distortions.<n>We propose MoA-VR, which mimics the reasoning and processing procedures of human professionals through three coordinated agents.<n>Specifically, we construct a large-scale and high-resolution video degradation recognition benchmark and build a vision-language model (VLM) driven degradation identifier.
  arXiv  Detail & Related papers  (2025-10-09T17:42:51Z)
• EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing [170.71134330650796]
  EdiVal-Agent is an object-centric evaluation framework for instruction-based image editing.<n>It is designed to assess not only standard single-turn but also multi-turn instruction-based editing with precision.<n>We build EdiVal-Bench, a benchmark covering 9 instruction types and 13 state-of-the-art editing models spanning in-context, flow-matching, and diffusion paradigms.
  arXiv  Detail & Related papers  (2025-09-16T17:45:39Z)
• VideoLights: Feature Refinement and Cross-Task Alignment Transformer for Joint Video Highlight Detection and Moment Retrieval [8.908777234657046]
  Large-language and vision-language models (LLM/LVLMs) have gained prominence across various domains.<n>Here we propose VideoLights, a novel HD/MR framework addressing these limitations through (i) Convolutional Projection and Feature Refinement modules.<n> Comprehensive experiments on QVHighlights, TVSum, and Charades-STA benchmarks demonstrate state-of-the-art performance.
  arXiv  Detail & Related papers  (2024-12-02T14:45:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.