Paper Detail
Zhengyang Sun, Yu Chen, Xin Zhou, Xiaofan Li, Xiwu Chen, Dingkang Liang, Xiang Bai
Text-to-video diffusion models have enabled open-ended video synthesis, but often struggle with generating the correct number of objects specified in a prompt. We introduce NUMINA , a training-free identify-then-guide framework for improved numerical alignment. NUMINA identifies prompt-layout inconsistencies by selecting discriminative self- and cross-attention heads to derive a countable latent layout. It then refines this layout conservatively and modulates cross-attention to guide regeneration. On the introduced CountBench, NUMINA improves counting accuracy by up to 7.4% on Wan2.1-1.3B, and by 4.9% and 5.5% on 5B and 14B models, respectively. Furthermore, CLIP alignment is improved while maintaining temporal consistency. These results demonstrate that structural guidance complements seed search and prompt enhancement, offering a practical path toward count-accurate text-to-video diffusion. The code is available at https://github.com/H-EmbodVis/NUMINA.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@article{sun2026when,
title = {When Numbers Speak: Aligning Textual Numerals and Visual Instances in Text-to-Video Diffusion Models},
author = {Zhengyang Sun and Yu Chen and Xin Zhou and Xiaofan Li and Xiwu Chen and Dingkang Liang and Xiang Bai},
year = {2026},
abstract = {Text-to-video diffusion models have enabled open-ended video synthesis, but often struggle with generating the correct number of objects specified in a prompt. We introduce NUMINA , a training-free identify-then-guide framework for improved numerical alignment. NUMINA identifies prompt-layout inconsistencies by selecting discriminative self- and cross-attention heads to derive a countable latent layout. It then refines this layout conservatively and modulates cross-attention to guide regeneratio},
url = {https://arxiv.org/abs/2604.08546},
keywords = {cs.CV, text-to-video diffusion models, numerical alignment, prompt-layout inconsistencies, self-attention heads, cross-attention heads, latent layout, attention modulation, CountBench, CLIP alignment, temporal consistency, code available, huggingface daily},
eprint = {2604.08546},
archiveprefix = {arXiv},
}
{}