Open
Description
New Feature for FluxPipeline
Paper: Link
Code : Link
workflow : Link
This paper introduces Sample-specific Score-aware Noise Injection (SSNI) to improve diffusion-based purification (DBP) methods. Unlike existing approaches that use a fixed noise level (t*) for all samples, SSNI adapts t* based on how noisy or clean each sample is. Using a pre-trained score network, SSNI estimates a sample's deviation from the clean data and adjusts the noise level accordingly.
This have stunning Results with Flux