Research
HybridStitch: Pixel and Timestep Level Model Stitching for Diffusion Acceleration
arXiv:2603.07815v2 Announce Type: replace Abstract: Iterative diffusion generation requires different amounts of model capacity at different token positions. Prior model combinations allocate capacity
arXiv:2603.07815v2 Announce Type: replace Abstract: Iterative diffusion generation requires different amounts of model capacity at different token positions. Prior model combinations allocate capacity by denoising step or by drafting and verification role. We present HybridStitch, a training-free framework that uses token-level feedback to select the model supplying each adopted update. Continuous image refinement and discrete language commitment operate on a shared generation state, while model-specific KV buffers provide attention context for selected queries. HybridStitch achieves 1.63--1.68imes faster image generation and 1.15--1.58imes diffusion large language model (dLLM) throughput on four main benchmarks. Estimated dLLM decode query cost decreases by 50--66% with comparable generation quality.
Related
- TC-Pade: Trajectory-Consistent Pade Approximation for Diffusion Acceleration
- Advanced Pixel Diffusion Model with Guided Sparse Global Refinement
- MoECa: Aligning Feature Reuse with Expert Decomposition in Diffusion Transformers
- Predict to Skip: Linear Multistep Feature Forecasting for Efficient Diffusion Transformers
- Adaptive Spectral Feature Forecasting for Diffusion Sampling Acceleration
Source: arXiv cs.CV | 2026-10-14