Drift-AR: Single-Step Visual Autoregressive Generation via Anti-Symmetric Drifting

Autoregressive (AR)-Dix001Busion hybrid paradigms combine AR'sstructured semantic modeling with dix001Busion's high-x001Cdelity synthesis, yetsux001Ber from a dual speed bottleneck: the sequential AR stage and theiterative multi-step denoising of the dix001Busion vision decode stage. Exist-ing methods address each in isolation without a unix001Ced principle design.We observe that the per-position prediction entropy of continuous-spaceAR models naturally encodes spatially varying generation uncertainty,which simultaneously governing draft prediction quality in the AR stageand rex001Decting the corrective ex001Bort required by vision decoding stage,which is not fully explored before. Since entropy is inherently tied toboth bottlenecks, it serves as a natural unifying signal for joint accel-eration. In this work, we propose Drift-AR, which leverages entropysignal to accelerate both stages: 1) for AR acceleration, we introduceEntropy-Informed Speculative Decoding that align draftx0015target entropydistributions via a causal-normalized entropy loss, resolving the entropymismatch that causes excessive draft rejection; 2) for visual decoder ac-celeration, we reinterpret entropy as the physical variance of the initialstate for an anti-symmetric drifting x001Celdx0016high-entropy positions acti-vate stronger drift toward the data manifold while low-entropy positionsyield vanishing driftx0016enabling single-step (1-NFE) decoding without it-erative denoising or distillation. Moreover, both stages share the sameentropy signal, which is computed once with no extra cost. Experimentson MAR, TransDix001B, and NextStep-1 demonstrate 3.8x00155.5× speedup withgenuine 1-NFE decoding, matching or surpassing original quality. Codeis available at https://github.com/aSleepyTree/Drift-AR.