MoWa: An Authoring Tool for Refining AI-Generated Human Avatar Motions Through Latent Waveform Manipulation

要旨

Creating expressive and realistic motion animations is a challenging task. Generative artificial intelligence (AI) models have emerged to address this challenge, offering the capability to synthesize human motion animations from text prompts. However, the effective integration of AI-generated motion into professional designer workflows remains uncertain. This study proposes MoWa, an authoring tool designed to refine AI-generated human motions to meet professional standards. A formative study with six professional motion designers identified the strengths and weaknesses of AI-generated motions. To address these weaknesses, MoWa utilizes latent space to enhance the expressiveness of motions, making them suitable for use in professional workflows. A user study involving twelve professional motion designers was conducted to evaluate MoWa's effectiveness in refining AI-generated motions. The results indicated that MoWa streamlines the motion design process and improves the quality of the outcomes. These findings suggest that incorporating latent space into motion design tasks can improve efficiency.

著者
Jeongseok Oh
Gwangju Institute of Science and Technology, Gwangju, Korea, Republic of
SeungJun Kim
Gwangju Institute of Science and Technology, Gwangju, Korea, Republic of
DOI

10.1145/3706598.3714253

論文URL

https://dl.acm.org/doi/10.1145/3706598.3714253

動画

会議: CHI 2025

The ACM CHI Conference on Human Factors in Computing Systems (https://chi2025.acm.org/)

セッション: XR and Virtual Characteristics

G314+G315
7 件の発表
2025-04-30 18:00:00
2025-04-30 19:30:00
日本語まとめ
読み込み中…