Presentation

· Contributors · Organizations · Search Program · My Schedule · My Recommendations · Happening Now

Inspiration Seeds: Learning Non-Literal Visual Combinations for Generative Exploration

DescriptionInspiration Seeds is a generative framework that combines two input images into diverse, visually coherent compositions—without text prompts. Trained on synthetic triplets using CLIP Sparse Autoencoders, it uncovers latent visual relationships between inputs, supporting early-stage creative exploration where designers seek emergent ideas from loosely connected visual references.

Authors

Massachusetts Institute of Technology (MIT)