image-conditioning

Visual Concept Fusion (VCF) enables dual conditioning on both an image and text prompt in diffusion models at inference time without retraining, using a lightweight aligner and fusion strategy.