LipsFlower Stabilizes Image Restoration using Lipschitz-Controlled Flow Matching
November 9, 2026
LipsFlower introduces LipHDiT, a diffusion transformer architecture designed to prevent small input perturbations from causing instability during image restoration. It uses stabilized approximate spectral normalization and bounded modulation to maintain stable dynamics.
HOW THIS AFFECTS YOU
●
builderYou can use this architecture to create more robust image restoration tools that don't break on noisy inputs.
●
researcherYou can apply these stability bounds to improve flow-matching models.