KD-CVG: A Knowledge-Driven Approach for Creative Video Generation

ICASSP 2026

Linkai Liu^†  ·  Wei Feng^★  ·  Xi Zhao^★  ·  Shen Zhang^★  ·  Xingye Chen^★  ·  Zheng Zhang^★
Jingjing Lv^★  ·  Junjie Shen^★  ·  Ching Law^★  ·  Yuchen Zhou^†  ·  Zipeng Guo^†  ·  Chao Gou^†‡

^† Sun Yat-sen University     ^★ JD.COM
^‡ Corresponding author

---

Abstract

Creative Generation (CG) leverages generative models to automatically produce advertising content that highlights product features, and it has been a significant focus of recent research. However, while CG has advanced considerably, most efforts have concentrated on generating advertising text and images, leaving Creative Video Generation (CVG) relatively underexplored. This gap is largely due to two major challenges faced by Text-to-Video (T2V) models: (a) ambiguous semantic alignment, where models struggle to accurately correlate product selling points with creative video content, and (b) inadequate motion adaptability, resulting in unrealistic movements and distortions.

To address these challenges, we develop a comprehensive Advertising Creative Knowledge Base (ACKB) as a foundational resource and propose a knowledge-driven approach (KD-CVG) to overcome the knowledge limitations of existing models. KD-CVG consists of two primary modules: Semantic-Aware Retrieval (SAR) and Multimodal Knowledge Reference (MKR). SAR utilizes the semantic awareness of graph attention networks and reinforcement learning feedback to enhance the model's comprehension of the connections between selling points and creative videos. Building on this, MKR incorporates semantic and motion priors into the T2V model to address existing knowledge gaps. Extensive experiments have demonstrated KD-CVG's superior performance in achieving semantic alignment and motion adaptability, validating its effectiveness over other state-of-the-art methods.

---

Quantitative Comparison

Method	Textual Alignment	Temporal Consistency	Dynamic Degree	Motion Smoothness	Min-Max Score
Show-1	31.34%	96.83%	44.44%	98.04%	55.66%
VideoCrafter2	30.43%	97.03%	88.89%	91.91%	44.69%
Open-Sora	28.84%	97.94%	22.22%	98.92%	47.06%
KD-CVG (Ours)	30.68%	97.95%	72.00%	98.65%	81.80%

User Study & Inference Cost

Method	Text Alignment↑	Video Quality↑	Motion Regularity↑	Highlight Degree↑	Time Cost↓
Show-1	3.70	3.23	3.44	3.59	3692s
VideoCrafter2	3.63	3.81	2.26	3.26	161s
Open-Sora	3.26	3.07	2.63	3.30	52s
KD-CVG (Ours)	3.74	3.70	3.81	3.89	102s

---

Dataset (ACKB)

We publicly release the Advertising Creative Knowledge Base (ACKB), containing approximately 10K selling-point / ACV pairs with associated creative scripts and metadata, covering 377 daily product categories (e.g., toothbrushes, facial masks, shampoos).

• Download link: https://3.cn/1186-864O
• Access password: qj2w7y
• data format: path: video_path, video_name is video_md5_code.mp4, caption: description of the corresponding video.

Please use the dataset for academic / research purposes only and cite our paper if you use ACKB in your work.

---

Citation

If you find this work useful, please consider citing:

@inproceedings{liu2026kdcvg,
  title     = {KD-CVG: A Knowledge-Driven Approach for Creative Video Generation},
  author    = {Liu, Linkai and Feng, Wei and Zhao, Xi and Zhang, Shen and Chen, Xingye and
               Zhang, Zheng and Lv, Jingjing and Shen, Junjie and Law, Ching and
               Zhou, Yuchen and Guo, Zipeng and Gou, Chao},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year      = {2026}
}

---