# Layer- and Timestep-Adaptive Differentiable Token Compression Ratios for Efficient Diffusion Transformers

arXiv: https://arxiv.org/abs/2412.16822 · HTML: https://mlsanity.com/paper/2412.16822/

## Key facts

- Cited by 22 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models.
- Builds on 46 of its 57 references that are in the map.
- First posted on arXiv 22 Dec 2024.

## Placed in

- Diffusion models / [Diffusion foundations and sampling](https://mlsanity.com/diffusion/diffusion-foundations-and-sampling/)
- Efficient LLM inference and serving / [Visual token pruning in multimodal models](https://mlsanity.com/inference/visual-token-pruning-in-multimodal-models/)

## Builds on (most cited first, 30 of 46)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Denoising Diffusion Implicit Models](https://mlsanity.com/paper/2010.02502/) (2020)
- [Diffusion Models Beat GANs on Image Synthesis](https://mlsanity.com/paper/2105.05233/) (2021)
- [Deep Unsupervised Learning using Nonequilibrium Thermodynamics](https://mlsanity.com/paper/1503.03585/) (2015)
- [Scalable Diffusion Models with Transformers](https://mlsanity.com/paper/2212.09748/) (2022)
- [Classifier-Free Diffusion Guidance](https://mlsanity.com/paper/2207.12598/) (2022)
- [SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis](https://mlsanity.com/paper/2307.01952/) (2023)
- [LAION-5B: An open large-scale dataset for training next generation image-text models](https://mlsanity.com/paper/2210.08402/) (2022)
- [Improved Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2102.09672/) (2021)
- [Elucidating the Design Space of Diffusion-Based Generative Models](https://mlsanity.com/paper/2206.00364/) (2022)
- [CLIPScore: A Reference-free Evaluation Metric for Image Captioning](https://mlsanity.com/paper/2104.08718/) (2021)
- [DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps](https://mlsanity.com/paper/2206.00927/) (2022)
- [LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs](https://mlsanity.com/paper/2111.02114/) (2021)
- [One-step Diffusion with Distribution Matching Distillation](https://mlsanity.com/paper/2311.18828/) (2023)
- [Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis](https://mlsanity.com/paper/2306.09341/) (2023)
- [DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification](https://mlsanity.com/paper/2106.02034/) (2021)
- [PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis](https://mlsanity.com/paper/2310.00426/) (2023)
- [Adversarial Diffusion Distillation](https://mlsanity.com/paper/2311.17042/) (2023)
- [On Distillation of Guided Diffusion Models](https://mlsanity.com/paper/2210.03142/) (2022)
- [Improved Distribution Matching Distillation for Fast Image Synthesis](https://mlsanity.com/paper/2405.14867/) (2024)
- [BranchyNet: Fast Inference via Early Exiting from Deep Neural Networks](https://mlsanity.com/paper/1709.01686/) (2017)
- [All are Worth Words: A ViT Backbone for Diffusion Models](https://mlsanity.com/paper/2209.12152/) (2022)
- [SkipNet: Learning Dynamic Routing in Convolutional Networks](https://mlsanity.com/paper/1711.09485/) (2017)
- [SmartBrush: Text and Shape Guided Object Inpainting with Diffusion Model](https://mlsanity.com/paper/2212.05034/) (2022)
- [Mixture-of-Depths: Dynamically allocating compute in transformer-based language models](https://mlsanity.com/paper/2404.02258/) (2024)
- [Imagen Editor and EditBench: Advancing and Evaluating Text-Guided Image Inpainting](https://mlsanity.com/paper/2212.06909/) (2022)

## Built on by (most cited first, 22 of 22)

- [Shifting AI Efficiency From Model-Centric to Data-Centric Compression](https://mlsanity.com/paper/2505.19147/) (2025)
- [Scale-wise Distillation of Diffusion Models](https://mlsanity.com/paper/2503.16397/) (2025)
- [Training-free Mixed-Resolution Latent Upsampling for Spatially Accelerated Diffusion Transformers](https://mlsanity.com/paper/2507.08422/) (2025)
- [Shallow-π: Knowledge Distillation for Flow-based VLAs](https://mlsanity.com/paper/2601.20262/) (2026)
- [RAPID^3: Tri-Level Reinforced Acceleration Policies for Diffusion Transformer](https://mlsanity.com/paper/2509.22323/) (2025)
- [UniMoD: Efficient Unified Multimodal Transformers with Mixture-of-Depths](https://mlsanity.com/paper/2502.06474/) (2025)
- [Are Image-to-Video Models Good Zero-Shot Image Editors?](https://mlsanity.com/paper/2511.19435/) (2025)
- [DDiT: Dynamic Patch Scheduling for Efficient Diffusion Transformers](https://mlsanity.com/paper/2602.16968/) (2026)
- [TRIM: Scalable 3D Gaussian Diffusion Inference with Temporal and Spatial Trimming](https://mlsanity.com/paper/2511.16642/) (2025)
- [DC-DiT: Adaptive Compute and Elastic Inference for Visual Generation via Dynamic Chunking](https://mlsanity.com/paper/2603.06351/) (2026)
- [Phase-Aligned RoPE for Mixed-Resolution Diffusion Transformer](https://mlsanity.com/paper/2511.19778/) (2025)
- [Shiva-DiT: Residual-Based Differentiable Top-k Selection for Efficient Diffusion Transformers](https://mlsanity.com/paper/2602.05605/) (2026)
- [DiffSparse: Accelerating Diffusion Transformers with Learned Token Sparsity](https://mlsanity.com/paper/2604.03674/) (2026)
- [CoReDiT: Spatial Coherence-Guided Token Pruning and Reconstruction for Efficient Diffusion Transformers](https://mlsanity.com/paper/2605.14191/) (2026)
- [Elastic Token Compression for Pixel-Space Diffusion Transformers](https://mlsanity.com/paper/2608.29281/) (2026)
- [NanoFLUX: Distillation-Driven Compression of Large Text-to-Image Generation Models for Mobile Devices](https://mlsanity.com/paper/2602.06879/) (2026)
- [HiLo-Token: Input-Adaptive High-Low Frequency Token Compression for Efficient Image Editing](https://mlsanity.com/paper/2606.13898/) (2026)
- [CSD: Content-aware Speculative Decoding for Efficient Image Generation](https://mlsanity.com/paper/2606.27829/) (2026)
- [Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers](https://mlsanity.com/paper/2607.00382/) (2026)
- [MOSAIK: Multi-Patch Content-Aware Spatial Allocation of Image Tokens for Efficient Generation](https://mlsanity.com/paper/2608.05450/) (2026)
- [RegionCache: Semantic-Aware Region Reuse for Efficient Multi-Turn Image Generation](https://mlsanity.com/paper/2608.29809/) (2026)
- [ForcingBooster: Adaptive Patch-wise Denoising for Fast Autoregressive Video Diffusion](https://mlsanity.com/paper/2610.12625/) (2026)

## Abstract (arXiv)

Diffusion Transformers (DiTs) have achieved state-of-the-art (SOTA) image generation quality but suffer from high latency and memory inefficiency, making them difficult to deploy on resource-constrained devices. One major efficiency bottleneck is that existing DiTs apply equal computation across all regions of an image. However, not all image tokens are equally important, and certain localized areas require more computation, such as objects. To address this, we propose DiffCR, a dynamic DiT inference framework with differentiable compression ratios, which automatically learns to dynamically route computation across layers and timesteps for each image token, resulting in efficient DiTs. Specifically, DiffCR integrates three features: (1) A token-level routing scheme where each DiT layer includes a router that is fine-tuned jointly with model weights to predict token importance scores. In this way, unimportant tokens bypass the entire layer's computation; (2) A layer-wise differentiable ratio mechanism where different DiT layers automatically learn varying compression ratios from a zero initialization, resulting in large compression ratios in redundant layers while others remain less compressed or even uncompressed; (3) A timestep-wise differentiable ratio mechanism where each denoising timestep learns its own compression ratio. The resulting pattern shows higher ratios for noisier timesteps and lower ratios as the image becomes clearer. Extensive experiments on text-to-image and inpainting tasks show that DiffCR effectively captures dynamism across token, layer, and timestep axes, achieving superior trade-offs between generation quality and efficiency compared to prior works. The project website is available at https://www.haoranyou.com/diffcr.