# Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs **Authors:** Donghyeon Cho, Joowon Kim, Seungho Shin, Eunho Yang **Lab:** KAIST · **Published:** 26 Oct 2026 · **Upvotes:** 101 · **Level:** Intermediate **Links:** [arXiv](https://arxiv.org/abs/2610.36671) · [PDF](https://arxiv.org/pdf/2610.36671) · [Code](https://github.com/hyeon9698/LOWN) · [This page](https://omen-bit.github.io/ai-paper-daily/paper/2610.36671.html) ## Curator's notes **TL;DR:** LOWN skips image processing during diffusion decoding unless needed, cutting FLOPs while keeping performance. - **What's new:** Introduces certificate‑guided token unmasking that decides per step whether to use visual tokens, unlike prior static pruning methods. - **Key result:** Retains 99.6% of LLaDA‑V's score on ten benchmarks while using only 35.3% of decoding FLOPs on average. - **Why it matters:** Reduces compute cost for vision‑language diffusion models, enabling faster inference on limited hardware. ## Key numbers - **22.8%**: …image changes a token being unmasked at only 22.8% of steps on average and at as… - **7.9%**: …token being unmasked at only 22.8% of steps on average and at as few as 7.9%. - **99.6%**: LOWN retains 99.6% of LLaDA-V's score on ten benchmarks at 35.3% of the decoding FLOPs on… ## Abstract Diffusion vision-language models (DVLMs) process their entire input at every denoising step. Since visual tokens make up 89-97% of this input on the benchmarks we study, most of the decoding computation is spent repeatedly processing the same image. Visual token pruning methods partially relieve this burden by removing part of the visual tokens, but they lose the removed information and still process the remaining image at every step. To measure how often the image is actually needed, we compare LLaDA-V's predictions with and without the image at every decoding step and find that removing the image changes a token being unmasked at only 22.8% of steps on average and at as few as 7.9%. We therefore propose LOWN, a decoding method that decides at every step whether to process the image and how many tokens to unmask. LOWN predicts the masked tokens without the image and processes the image only when this prediction disagrees with the last image-conditioned prediction. Agreement certifies the earlier prediction for reuse, and the number of certified tokens sets how many are unmasked in that step. We further observe that the need for the image differs across benchmarks, so LOWN's adaptive decision is needed rather than a fixed schedule. Without training or a tuned threshold, LOWN retains 99.6% of LLaDA-V's score on ten benchmarks at 35.3% of the decoding FLOPs on average. LOWN also generalizes across LLaDA-based DVLMs, composes orthogonally with visual token pruning, and runs under key-value caching for further efficiency gains. Code and project page are available at https://hyeon9698.github.io/LOWN/ ## What you could build 1. Implement a real‑time image captioning service that uses LOWN to lower latency 2. Create a mobile app that generates visual stories with diffusion VLMs using LOWN for efficiency 3. Build a benchmark suite comparing static token pruning vs. LOWN on various VLM tasks ## Quiz yourself How often does removing the image change an unmasked token during decoding according to the authors? **Answer:** Only 22.8% of steps on average, and as few as 7.9% on some benchmarks. --- Notes written by AI from the paper, numbers copied from the abstract. Check the paper before you rely on them. From AI Papers Daily.