面壁智能开发者 @_Andresio 在 #BuildSmall 黑客松中打造 PAREIDOLIA,用 MiniCPM-V 4.5 识别物体特征并生成个性,结合 VoxCPM2 无需参考音频即可设计独特角色声音,让日常物品“活”成共享虚拟动物园中的角色。该项目展示了视觉与语音模型融合的趣味 AI 交互新形态。
PAREIDOLIA — Giving Everything a Face and a Voice 👁️
Created by developer @_Andresio during the #BuildSmall Hackathon, PAREIDOLIA turns everyday objects into expressive characters by finding faces hidden in the world around us.
Point a camera at an object, and PAREIDOLIA can:
👁️ Use MiniCPM-V 4.5 to identify visible features and create a unique personality for each object
🎙️ Use VoxCPM2 to design a distinct character voice from text descriptions, without reference audio
✨ Bring objects to life with blinking eyes, expressions, and voices based on their real-world features
🪄 Turn every awakened object into a character in a shared virtual “Menagerie”
PAREIDOLIA demonstrates how vision and speech models can transform ordinary objects into living characters — combining visual understanding, personality creation, and voice generation to create a playful new form of AI interaction.
Explore it:
🤗 https://huggingface.co/spaces/build-small-hackathon/pareidolia
Models:
🔗 MiniCPM-V 4.5: https://huggingface.co/openbmb/MiniCPM-V-4_5
🔗 VoxCPM2: https://huggingface.co/openbmb/VoxCPM2
来源:OpenBMB · x.com