Vibe Spaces for Creatively Connecting and Expressing Visual Concepts
Huzheng Yang, Katherine Xu, Andrew Lu, Michael D. Grossberg, Yutong Bai, Jianbo Shi
2025-12-19
Summary
This paper introduces a new way to combine images to create novel visuals, focusing on capturing the 'vibe' or underlying feeling shared between different concepts.
What's the problem?
Currently, computer programs struggle to creatively blend images because they have trouble understanding how seemingly different ideas connect. They can't easily find a smooth path between these ideas in the complex digital space where images are represented, leading to blends that don't make sense or lack a consistent style.
What's the solution?
The researchers developed something called 'Vibe Space,' which is like a map that organizes images based on their shared vibes. This map isn't a simple grid, but a more flexible network that allows for smooth transitions between concepts. It helps the computer find the best way to blend images while maintaining a consistent and meaningful style, using a system similar to how CLIP understands images.
Why it matters?
This work is important because it moves us closer to computers being able to generate truly creative content. By understanding and blending 'vibes,' machines can create images that are not just technically correct, but also aesthetically pleasing and conceptually coherent, potentially useful for art, design, and other creative fields.
Abstract
Creating new visual concepts often requires connecting distinct ideas through their most relevant shared attributes -- their vibe. We introduce Vibe Blending, a novel task for generating coherent and meaningful hybrids that reveals these shared attributes between images. Achieving such blends is challenging for current methods, which struggle to identify and traverse nonlinear paths linking distant concepts in latent space. We propose Vibe Space, a hierarchical graph manifold that learns low-dimensional geodesics in feature spaces like CLIP, enabling smooth and semantically consistent transitions between concepts. To evaluate creative quality, we design a cognitively inspired framework combining human judgments, LLM reasoning, and a geometric path-based difficulty score. We find that Vibe Space produces blends that humans consistently rate as more creative and coherent than current methods.