{"ID":6023607,"CreatedAt":"2026-07-08T01:00:23.257252134Z","UpdatedAt":"2026-07-10T14:27:52.484755835Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.06335","arxiv_id":"2607.06335","title":"Bridging Diffusion Pruning and Step Distillation with Teacher-Aligned Repair","abstract":"Diffusion models generate high-quality images, but their inference cost comes from two sources: large denoising networks and repeated denoising steps. Existing compression pipelines usually attack these costs separately. Pruning reduces the network, but most pruning methods still rely on a long post-pruning retraining stage to recover a many-step sampler. Step distillation reduces the number of denoising steps, but it usually assumes a student that can already follow the teacher well enough to receive useful distillation gradients. This paper asks whether post-pruning retraining can be replaced by step distillation. We find that the direct replacement fails: after pruning an EDM2-XS teacher, starting SiDA from the pruned checkpoint produces unusable samples. We introduce a short teacher-alignment repair stage as a bridge between pruning and step distillation. The bridge matches the pruned generator to the teacher on noisy real-image latents, then hands the repaired checkpoint to one-step distillation. On ImageNet-512, the original EDM2-XS baseline uses 124.713M parameters and 63 network evaluations, reaching an FID of 3.53. With a suitable distillation objective, our 20% pruned one-step generator uses 98.826M parameters and one network evaluation, reaching an FID of 3.12. With 30% pruning, the model uses 88.029M parameters and one network evaluation, with an FID of 4.26.","short_abstract":"Diffusion models generate high-quality images, but their inference cost comes from two sources: large denoising networks and repeated denoising steps. Existing compression pipelines usually attack these costs separately. Pruning reduces the network, but most pruning methods still rely on a long post-pruning retraining...","url_abs":"https://arxiv.org/abs/2607.06335","url_pdf":"https://arxiv.org/pdf/2607.06335v1","authors":"[\"Jincheng Ying\",\"Li Wenlin\",\"Minghui Xu\",\"Yinhao Xiao\"]","published":"2026-07-07T14:30:24Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Diffusion Model\"]","has_code":false}