Diffusion transformers are widely used in image-editing services. To preserve both efficiency and privacy, existing systems commonly rely on cloud-device collaboration. Keeping request-specific inference on the device avoids sending personal images and prompts to the cloud. These workloads often use common templates, such as in face swapping and virtual try-on, enabling KV state to be precomputed and reused across requests. However, this reusable state is large and scales with the number of templates, making it impractical to store all of them on mobile devices. Storing them in the cloud addresses this capacity issue, but cache transfer is placed onto the inference path and can even be slower than on-device computation, especially under fluctuating mobile networks. We introduce SAKURA, a cloud-edge architecture that delivers reusable attention state to mobile devices under a bandwidth budget. It first uses a fine-grained token-importance-aware compression method to preserve image quality under limited transfer budgets. At runtime, SAKURA adaptively selects the cache representation for each upcoming shard based on network bandwidth. To enable efficient execution, we develop a pipelined runtime that overlaps cache transfer and reconstruction with on-device execution. We evaluate SAKURA using three editing tasks and two diffusion models, using a Snapdragon 8 Elite Gen 5 phone. SAKURA reduces latency by 1.82x and 3.27x over CacheGen adapted to DiT and on-device generation, respectively. It also reduces cache storage by 3.06x over uncompressed cache, while preserving image quality.
Figures & tables
Figure 1. Cloud–edge KV reuse for template-based editing.
Figure 2. Reusable template tokens have highly nonuniform downstream impact, while their relative importance remains stable across request-specific references.
Figure 3. Mobile network variation motivates adaptive cache streaming. (a) Required compression ratio. (b) The underlying 5G bandwidth trace.
Figure 4. SAKURA system overview.
Figure 5. Offline token-importance-aware cache construction. Masked-query attention is aggregated into reusable-token importance scores and converted into q -controlled row scales. These scales bias the low-rank factorization toward important cache rows, after which the factors are quantized independently.
Figure 6. Progressive cache packing. SAKURA shares ordered rank components and precision bit planes across supported configurations.
Figure 7. Workflow of online adaptive cache streaming.
Figure 8. Rank–precision tradeoff. Different precision pairs are optimal at different byte budgets.
Figure 9. Pipelined mobile execution. CPU reception and GPU cache reconstruction for the next shard overlap NPU execution of the current shard.
Workload
Task
Input Type
Model
Mask %
FS
Face swap
Image+Image
FLUX2
11.2
VTON
Virtual try-on
Image+Image
FLUX2
26.3
MB
MagicBrush
Image+Text
FLUX2
19.3
MB-SD
SD 3.5
Table 1. Models and image editing tasks.
Figure 10. Network and workload distribution.
Figure 11. An example showing FLUX2 latency under VTON workload during a 60-second network trace.
Figure 12. Generation latency under 5G trace.
Figure 13. Per-template cache storage on server.
Figure 14. Example outputs of requests from three tasks using FLUX2 with different systems.
Figure 15. Transferred cache size vs. image quality.
5G (main)
4G+5G mix
WiFi
Systems
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
FS
SAKURA
27.9
0.90
24.9
0.85
32.7
0.97
INTQ
21.3
0.79
21.0
0.78
21.4
0.80
CacheGen
20.8
0.70
16.2
0.51
23.5
0.81
VTON
SAKURA
30.4
0.95
26.8
0.91
37.3
0.99
INTQ
24.1
0.91
23.1
0.89
25.6
0.93
Table 2. Mean image quality on the real device (vs. full-cache output), averaged over all requests.
Method
FS
VTON
MB
MD-SD
SAKURA
86.0%
88.0%
90.0%
85.1%
INTQ
56.5%
72.8%
79.7%
47.4%
CacheGen
35.2%
55.8%
46.6%
43.5%
Table 3. User study comparing perceptual similarity to uncompressed cache.
Figure 16. Latency and quality under different editable-mask ratio.
Figure 17. Generation latency under different network traces.
Figure 18. Latency and quality under different resolutions.
Figure 19. Effect of attention-weighting exponent q .
The remarkable generation quality of modern diffusion models often comes at the cost of massive parameter counts, which necessitate server-side inference with significant computational costs and potential privacy risks. Consequently, there is growing momentum toward developing efficient on-device alternatives. While recent efforts have optimized text-to-image models for mobile hardware, they remain relatively bulky, typically ranging from 0.5B to 1B parameters. We present BlazeEdit, a highly efficient, generalist image-to-image diffusion model tailored for on-device deployment. By identifying that many practical image editing tasks do not require text-based guidance, we eliminate the text-conditioning components and develop a multi-task architecture that consolidates object removal, outpainting, tone correction, relighting, and sticker generation into a single, compact model of only 195M parameters. BlazeEdit achieves a substantial reduction in download size and memory overhead while maintaining competitive generation quality. It completes a full inference pass in just 290ms on a Pixel 10, delivering a seamless, privacy-preserving, and lightning-fast experience for generalist image editing on the edge.
Real-world image generation often involves multi-turn editing, where users iteratively modify small regions while most image content remains unchanged. However, existing diffusion transformer (DiT)-based editing pipelines recompute the entire image at every turn, causing substantial redundant computation. Existing DiT acceleration methods further ignore semantic correspondence across prompts, leading to unnecessary recomputation or unsafe reuse that harms editing quality. To address this, we propose RegionCache, a semantic-aware reuse framework for multi-turn image editing that selectively reuses diffusion states from unchanged regions. RegionCache detects reusable regions through semantic overlap between consecutive prompts and cross-attention localization, and adopts an adaptive reuse schedule based on prompt similarity and contextual consistency. Experiments on PixArt-alpha demonstrate that RegionCache achieves 1.43x--2.55x end-to-end speedup while maintaining comparable image quality. Code is available at https://github.com/hebutBryant/RegionCache.
Peizheng Li, Xin Ai, Hanyuan Liu +2
School of Computer Science and Engineering, Northeastern University, China · School of Computer Science
Recent advances in diffusion transformers (DiTs) have set new standards in image generation, yet remain impractical for on-device deployment due to their high computational and memory costs. In this work, we present an efficient DiT framework tailored for mobile and edge devices that achieves transformer-level generation quality under strict resource constraints. Our design combines three key components. First, we propose a compact DiT architecture with an adaptive global-local sparse attention mechanism that balances global context modeling and local detail preservation. Second, we propose an elastic training framework that jointly optimizes sub-DiTs of varying capacities within a unified supernetwork, allowing a single model to dynamically adjust for efficient inference across different hardware. Finally, we develop Knowledge-Guided Distribution Matching Distillation, a step-distillation pipeline that integrates the DMD objective with knowledge transfer from few-step teacher models, producing high-fidelity and low-latency generation (e.g., 4-step) suitable for real-time on-device use. Together, these contributions enable scalable, efficient, and high-quality diffusion models for deployment on diverse hardware.