
Not All Prediction Targets Keep Training-Free Diffusion Guidance on the Manifold
Prediction target determines whether training-free guidance fails gracefully or catastrophically.
A research collective
ManLuML follows interesting questions across fields, wherever they lead.
Explore the researchTwo friends · Side research
ManLuML is a research side project by two friends. We follow questions we genuinely enjoy, take them as far as we can, and share what we learn.
Featured research · ECCV 2026

Prediction target determines whether training-free guidance fails gracefully or catastrophically.