Research map / Video / Video generation
When Numbers Speak: Aligning Textual Numerals and Visual Instances in Text-to-Video Diffusion Models
Zhengyang Sun, Yu Chen, Xin Zhou, Xiaofan Li, Xiwu Chen, Dingkang Liang, Xiang Bai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add NUMINA layout guidance | Wan2.1-1.3B · CountBench | CountAcc (%): 42.3 -> 49.7 | Improved |
| add NUMINA layout guidance | Wan2.1-14B · CountBench | CountAcc (%): 53.6 -> 59.1 | Improved |
| add seed search of 5 seeds instead of NUMINA | Wan2.1-1.3B · CountBench | CountAcc (%): 42.3 -> 45.5 | Improved |
| add prompt enhancement | Wan2.1-1.3B · CountBench | CountAcc (%): 42.3 -> 47.2 | Improved |
| add attention-derived layout instead of GroundingDINO layout (ablation) | NUMINA · CountBench | CountAcc (%): 47.5 -> 49.7 | Improved |
| add temporal cost C_t to overlap cost (ablation) | NUMINA layout refinement · CountBench | CountAcc (%): 45.1 -> 48.9 | Improved |
| add center cost C_c to overlap cost (ablation) | NUMINA layout refinement · CountBench | CountAcc (%): 45.1 -> 46.9 | Improved |
| select Top-1 head instead of averaging all heads (ablation) | NUMINA head selection · CountBench | CountAcc (%): 43.0 -> 49.7 | Improved |
| select Top-1 head instead of random head (ablation) | NUMINA head selection · CountBench | CountAcc (%): 44.1 -> 49.7 | Improved |
| add object addition alongside removal (ablation) | NUMINA · CountBench | CountAcc (%): 43.8 -> 49.7 | Improved |
| add object removal alongside addition (ablation) | NUMINA · CountBench | CountAcc (%): 47.7 -> 49.7 | Improved |
| add EasyCache acceleration | NUMINA · CountBench | wall-clock time (s): 431 -> 355 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.88)
- VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Improving Long-Text Alignment for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.86)
- VideoDPO: Omni-Preference Alignment for Video Diffusion Generation (CVPR 2025 · similarity 0.86)
- VideoTetris: Towards Compositional Text-to-Video Generation (NeurIPS 2024 · similarity 0.86)
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
