Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Viggle unveils JST-2 Motion for text- and video-to-3D animation

AI summary

Viggle disclosed JST-2 Motion, a full-body 3D motion module paired with JST-2 that generates motion from text or reconstructs it from video, with outputs compatible with standard 3D character skeletons. In vendor-reported blind human evaluations, the text-to-motion test compared it with HY-Motion-1.0 and Cartwheel Swing, while the video-to-motion test used DeepMotion Animate 3D and Cartwheel Comic 4; Viggle reported overall scores of 80.0% and 89.1% against one baseline in each task. The company also acknowledged slight foot sliding, weaker performance than Cartwheel Comic 4 on some fast sports footage, and limited control over direction, transitions, and rhythm.

In-depth insight

JST-2 Motion gives animation, game, and embodied-AI workflows a unified method for producing full-body 3D motion from either natural-language instructions or real footage.

(文/杨亮)10月28日消息,近日,物理原生世界模型公司 Viggle 公开 JST‑2 Motion 技术细节,作为 JST‑2 世界模型配套的全身 3D 动作生成模块,同时支持文本生成动作与视频重建 3D 人体运动,生成结果可直接适配 JST‑2 角色,无需额外动作重定向,原生兼容行业标准 3D 角色骨骼,面向动画、游戏、具身 AI 训练等场景。

JST‑2 Motion 具备两大核心能力。文本转全身动作,输入自然语言描述即可输出完整人体运动,艺术家不用再手动调关键帧或者从动作库筛选素材,AI 智能体也可以直接输出文本指令生成动作。

视频转 3D 动作重建,不仅还原画面可见肢体运动,还依靠学习到的人体动力学,推理被遮挡的身体部位姿态;视频素材不需要人物正对镜头,侧面、局部遮挡画面均可处理,能够降低动捕成本,同时为具身 AI 供给大量人体运动训练数据。

研究采用盲测人工评估方案,统一使用无纹理 MetaHuman 骨骼展示动作,排除角色外观干扰。文本转动作对比 HY‑Motion‑1.0、Cartwheel Swing 两大基线,测试覆盖日常移动、手势交互、格斗特技、舞蹈表演八大类场景;在手势、舞蹈类动作表现尤为突出,指令遵循度、足部接地、动作流畅度综合占优。对比 Cartwheel Swing 综合得分 74.8%,对比 HY‑Motion‑1.0 综合得分达到 80.0%。

#Viggle#JST-2 Motion#JST-2#MetaHuman

Original · 2 min readRead the original
Group chat

No comments yet. Start the conversation.