提出自适应动态选 token 技术，让图像生成更高效且细节更准。
AViTS: Adaptive Spatiotemporal Token Selection for Efficient Dynamic-Resolution Generation
- 根据文本关联和时序变化动态选关键令牌升分辨率
- 在 FLUX 上节省 6.34 倍计算量，Qwen-Image-Edit 上近 9 倍
- 适合追求高效率与高质量生成的视觉生成研究者
扩散变换器（DiTs）虽能生成高质量结果，但因迭代采样成本高。动态分辨率采样通过低分辨率去噪降低早期开销；然而，统一升采所有潜在令牌会引入冗余计算，并影响细粒度一致性。现有部分升采策略通常依赖局部结构或单步统计，难以同时捕捉令牌与文本语义相关性及跨扩散步骤的表示动态。我们提出 AViTS，一种面向动态分辨率 DiTs 的自适应时空令牌选择框架。AViTS 通过潜在-文本注意力建模空间重要性，通过令牌级特征在扩散时间步的变化建模时间重要性，并融合二者实现时空感知的选择性升采：优先对关键令牌进行分辨率提升，延迟次要令牌，从而减少冗余高分辨率计算，改善质量-效率权衡。AViTS 在 FLUX 上实现最高 6.34 倍、在 Qwen-Image-Edit 与 FLUX.1-Kontext-dev 上接近 9 倍的浮点运算量（FLOPs）缩减，且与蒸馏、量化、特征缓存等方法正交，结合蒸馏模型可达 14.76 倍。代码已开源。
原文摘要 · Abstract (English)
Diffusion Transformers (DiTs) achieve high-quality generation but are costly due to iterative sampling. Dynamic-resolution sampling reduces early-stage cost by denoising at low resolution; however, uniformly upsampling all latent tokens at resolution transitions incurs redundant computation and may degrade fine-detail consistency. Existing partial upsampling strategies typically rely on local latent structure cues or single-step statistics, making it difficult to jointly capture token-text semantic relevance and token-wise representation dynamics across diffusion steps. We propose AViTS, an adaptive spatiotemporal token selection framework for dynamic-resolution DiTs. AViTS models spatial importance via latent-text attention and temporal importance via token-level feature variation across diffusion timesteps, and fuses them to enable spatiotemporal importance-aware selective upsampling: it prioritizes resolution refinement for critical tokens while deferring less important ones, thereby reducing redundant high-resolution computation and improving the quality-efficiency trade-off. AViTS achieves up to 6.34x on FLUX and nearly 9x FLOPs reduction on Qwen-Image-Edit and FLUX.1-Kontext-dev, orthogonal to distillation, quantization, and feature caching, and reaching 14.76x with distilled models. Code: https://github.com/QHR69/AViTS
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。