Compositional Video Generation via Inference-Time Guidance

• URL: http://arxiv.org/abs/2605.14988v1
• Date: Thu, 14 May 2026 15:50:25 GMT
• Title: Compositional Video Generation via Inference-Time Guidance
• Abstract summary: Text-to-video diffusion models often fail on prompts requiring compositional understanding.<n>We propose textbfCVG, an inference-time guidance method for improving compositional faithfulness in frozen text-to-video models.
• Score: 69.53614395025632
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Text-to-video diffusion models generate realistic videos, but often fail on prompts requiring fine-grained compositional understanding, such as relations between entities, attributes, actions, and motion directions. We hypothesize that these failures need not be addressed by retraining the generator, but can instead be mitigated by steering the denoising process using the model's own internal grounding signals. We propose \textbf{CVG}, an inference-time guidance method for improving compositional faithfulness in frozen text-to-video models. Our key observation is that cross-attention maps already encode how prompt concepts are grounded across space and time. We train a lightweight compositional classifier on these attention features and use its gradients during early denoising steps to steer the latent trajectory toward the desired composition. Built on a frozen VLM backbone, the classifier transfers across semantically related composition labels rather than relying only on narrow category-specific features. CVG improves compositional generation without modifying the model architecture, fine-tuning the generator, or requiring layouts, boxes, or other user-supplied controls. Experiments on compositional text-to-video benchmarks show improved prompt faithfulness while preserving the visual quality of the underlying generator.

Related papers

• InstructionCrafter: Generating Consistent and High-Fidelity Visual Instructions [6.198298884034955]
  Existing text-to-image generation approaches rarely meet step faithfulness, cross-image consistency, and per-frame visual quality.<n>We propose InstructionCrafter, a diffusion-based framework with the key idea of separating the optimization of temporal and instructional alignment.<n>Our code and trained models will be publicly available.
  arXiv  Detail & Related papers  (2026-08-09T04:01:21Z)
• 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)
• AlcheMinT: Fine-grained Temporal Control for Multi-Reference Consistent Video Generation [58.844504598618094]
  We propose AlcheMinT, a unified framework that introduces explicit timestamps conditioning for subject-driven video generation.<n>Our approach introduces a novel positional encoding mechanism that unlocks the encoding of temporal intervals, associated in our case with subject identities.<n>We incorporate subject-descriptive text tokens to strengthen binding between visual identity and video captions, mitigating ambiguity during generation.
  arXiv  Detail & Related papers  (2025-12-11T18:59:34Z)
• Layer-Aware Video Composition via Split-then-Merge [55.12521724893102]
  Split-then-Merge (StM) is a framework designed to enhance control in generative video composition.<n>StM splits a large corpus of unlabeled videos into dynamic foreground and background layers, then self-composes them to learn how dynamic subjects interact with diverse scenes.
  arXiv  Detail & Related papers  (2025-11-25T19:53:15Z)
• JointTuner: Appearance-Motion Adaptive Joint Training for Customized Video Generation [13.168628936598367]
  JointTuner is a framework that enables joint optimization of appearance and motion components.<n>AiT Loss disrupts the flow of appearance-related components, guiding the model to focus exclusively on motion learning.<n>JointTuner is compatible with both UNet-based models and Diffusion Transformer-based models.
  arXiv  Detail & Related papers  (2025-03-31T11:04:07Z)
• APLA: Additional Perturbation for Latent Noise with Adversarial Training Enables Consistency [9.07931905323022]
  We propose a novel text-to-video (T2V) generation network structure based on diffusion models. Our approach only necessitates a single video as input and builds upon pre-trained stable diffusion networks. We leverage a hybrid architecture of transformers and convolutions to compensate for temporal intricacies, enhancing consistency between different frames within the video.
  arXiv  Detail & Related papers  (2023-08-24T07:11:00Z)
• RIGID: Recurrent GAN Inversion and Editing of Real Face Videos [73.97520691413006]
  GAN inversion is indispensable for applying the powerful editability of GAN to real images. Existing methods invert video frames individually often leading to undesired inconsistent results over time. We propose a unified recurrent framework, named textbfRecurrent vtextbfIdeo textbfGAN textbfInversion and etextbfDiting (RIGID) Our framework learns the inherent coherence between input frames in an end-to-end manner.
  arXiv  Detail & Related papers  (2023-08-11T12:17:24Z)
• TVTSv2: Learning Out-of-the-box Spatiotemporal Visual Representations at Scale [59.01246141215051]
  We analyze the factor that leads to degradation from the perspective of language supervision. We propose a tunable-free pre-training strategy to retain the generalization ability of the text encoder. We produce a series of models, dubbed TVTSv2, with up to one billion parameters.
  arXiv  Detail & Related papers  (2023-05-23T15:44:56Z)

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.