AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Logit-Conditioned Diffusion Decoding for Frozen Discrete-Token VLMs

Ji Woo Hong, Hee Suk Yoon, Gwanhyeong Koo, Eunseop Yoon, SooHwan Eom, Qi Dai, Chong Luo, Chang Yoo

NeurIPS 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a logit-conditioned diffusion decoder that improves the visual fidelity of frozen discrete-token VLMs by replacing the VQ-VAE decoding stage, without retraining the foundation model.

Problem

Discrete-token VLMs that generate images via codebook indices are limited in visual fidelity by the VQ-VAE decoder; prior fixes require replacing the tokenizer or retraining foundation-scale models.

Approach

Conditions a diffusion decoder on the frozen VLM's output logits over codebook indices, so the bottleneck VQ-VAE decoding stage is upgraded while the discrete-token VLM stays frozen.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace VQ-VAE decoder with DCDD logit-conditioned diffusion decoderJanus-Pro 7B · MJHQ-30KgFID: 14.29 -> 9.42Improved
replace VQ-VAE decoder with DCDDJanus-Pro 7B · ImageNet-1K val reconstructionrFID: 2.8511 -> 0.7535Improved
replace VQ-VAE decoder with DCDDJanus-Pro 7B · GenEval/DPG-Bench/WISEsemantic alignment: GenEval 77.24 -> 78.31; DPG 86.94 -> 87.64No clear effect
condition diffusion decoder on full logit distribution instead of argmax codes (ablation)DCDD · MJHQ-30KgFID: 11.0837 -> 9.4160Improved
use diffusion decoder with argmax codes instead of VQ-VAE decoder (ablation)Janus-Pro 7B · MJHQ-30KgFID: 14.29 -> 11.08Improved
retain only top-1 token of logit distribution (ablation)DCDD · MJHQ-30K AnimalsgFID: 31.6866 -> 33.2015Hurt
include second-ranked token in conditioning (ablation)DCDD · MJHQ-30K AnimalsgFID: 33.2015 -> 27.0649Improved
add tokens ranked 3-8 and tail (ablation)DCDD · MJHQ-30K AnimalsgFID: 27.0649 -> 25.5681 -> 25.1060Improved
use FLUX.1 Kontext image-to-image refiner instead of logit conditioning (ablation)VQ-VAE output · MJHQ-30KgFID: 9.4160 (DCDD) vs 10.6678 (I2I generic)Hurt
apply 7B-trained DCDD to Janus-Pro 1B without retrainingJanus-Pro 1B · MJHQ-30K AnimalsgFID: 35.14 -> 26.09Improved
apply DCDD to LlamaGen-XL (Otsu rank 1 logits)LlamaGen-XL · MJHQ-30K AnimalsgFID: 74.20 -> 74.66No clear effect
use DCDD decoder with 30 denoising stepsJanus-Pro 7B · MJHQ-30K end-to-endlatency (s): 11.86 -> 15.56Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy