Negative Guidance
A class of sampling techniques in diffusion models that steer the generation process away from undesired conditions, such as a particular style or content. It is implemented through repulsion terms or an explicit target density.
Timeline (last 90 days)
The paper proposes using Positive-Unlabeled Learning to reformulate negative guidance, giving the sampling process a well-defined target density and achieving asymptotic exactness in the particle limit.
Negative guidance typically faces an awkward situation: the harder it avoids, the worse the generation quality.
All Facts (2)
The paper proposes using Positive-Unlabeled Learning to reformulate negative guidance, giving the sampling process a well-defined target density and achieving asymptotic exactness in the particle limit.
50%UnverifiedNegative guidance typically faces an awkward situation: the harder it avoids, the worse the generation quality.
50%