# SAKURA: Serving Attention Key-Value State Under Rate Adaptation for Private Image Editing on Phones

DevFeed: [SAKURA: Serving Attention Key-Value State Under Rate Adaptation for Private Image Editing on Phones](<https://devfeed.tech/articles/sakura-serving-attention-key-value-state-under-rate-adaptation-for-private-image-editing-on-phones-94638.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29733>)

Author: Shuncheng Jie, Desen Sun, Ri Hong, Yuyang Tian, Hong Zhang, Sihang Liu

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [diffusion-transformers](<https://devfeed.tech/topics/diffusion-transformers.md>), [image editing](<https://devfeed.tech/topics/image-editing.md>), [Transformer](<https://devfeed.tech/topics/transformer.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [attention](<https://devfeed.tech/tags/attention.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [budget](<https://devfeed.tech/tags/budget.md>), [cache](<https://devfeed.tech/tags/cache.md>), [compression](<https://devfeed.tech/tags/compression.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [device](<https://devfeed.tech/tags/device.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [diffusion-transformers](<https://devfeed.tech/tags/diffusion-transformers.md>), [edge](<https://devfeed.tech/tags/edge.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [image](<https://devfeed.tech/tags/image.md>), [image-editing](<https://devfeed.tech/tags/image-editing.md>), [latency](<https://devfeed.tech/tags/latency.md>), [mobile](<https://devfeed.tech/tags/mobile.md>), [mobile-devices](<https://devfeed.tech/tags/mobile-devices.md>), [phones](<https://devfeed.tech/tags/phones.md>), [private](<https://devfeed.tech/tags/private.md>), [state](<https://devfeed.tech/tags/state.md>)

## AI overview

SAKURA keeps request-specific image inference on the phone while sending reusable attention KV state from the cloud, avoiding transfer of personal images and prompts. To address limited storage and variable network bandwidth, it compresses cache state using token importance, selects a representation for each upcoming shard based on bandwidth, and overlaps transfer and reconstruction with on-device execution. On three editing tasks and two diffusion models using a Snapdragon 8 Elite Gen 5, it reports 1.82x lower latency than CacheGen adapted to DiT and 3.27x lower latency than on-device generation, plus 3.06x less cache storage than an uncompressed cache while preserving image quality.

## Source excerpt

arXiv:2610.29733v1 Announce Type: new Abstract: 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.