1 paper · 1 filter
Ruolan Sun, Pawel Polak
Inference-time alignment of pretrained text-to-image models is typically performed along a single control axis, such as classifier-free guidance, attention editing, or reward-based…