Speculative Coupled Decoding for Training-Free Lossless Acceleration of Autoregressive Visual Generation

Autoregressive (AR) modeling has recently emerged as a promising new paradigm in visual generation, but its practical adoption is severely constrained by the slow inference speed of per-token generation, which often requires thousands of steps to produce a single sample. While several Speculative Decoding (SD)-based methods have been proposed to solve this problem by generating multiple tokens in a single forward step, they suffer from limited speedup, degraded quality, or require the training of a draft model. To solve these problems, we propose a new training-free, lossless SD framework, Speculative Coupled Decoding (SCD), by extending the recently proposed Speculative Jacobi Decoding (SJD). While SJD shows strong potential for accelerating AR generation by combining Jacobi iteration and SD, we found that its acceptance rate is still significantly limited due to the instability arising from the independent sampling process used during draft token generation. To overcome this, we introduce an information-theoretic approach, Coupling, which stabilizes the drafting trajectory of SJD by maximizing the probability of sampling identical draft tokens across consecutive iterations, significantly enhancing the acceptance rate while preserving its lossless property. Remarkably, this method requires only a single-line modification to the existing algorithm with almost zero overhead, yet achieves substantial performance gains, delivering up to a 4.2× speedup in image generation and 13.6× speedup in video generation compared to standard AR decoding, without any degradation or the need for additional training.

Recent AI models can create images and videos piece by piece, much like how chatbots write text one word at a time. This approach can produce high-quality results, but it is slow because the model may need to make thousands of small decisions before one image or video is finished. A common shortcut is to let the model guess several next pieces at once and then check which guesses are safe to keep. However, these guesses often change too much from one round to the next, so many of them get rejected and the speedup remains limited. We propose a simple method that makes consecutive guesses more consistent with each other, like helping the model stay on a stable path instead of starting over randomly each time. Because the guesses become more stable, the main model can accept more pieces at once. Our method does not require training a new helper model and does not change the final results compared with the original slower process. In experiments, it made image generation up to 4.2 times faster and video generation up to 13.6 times faster without reducing quality. This could make advanced image and video generation faster and cheaper to use in real-world applications.