mlsanityData to 14 Oct 2026

LegoDiffusion: Micro-Serving Text-to-Image Diffusion Workflows

arXiv 2604.08123 ↗First posted 9 Apr 2026cs.DC cs.AIcited by 0

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Text-to-image generation executes a diffusion workflow comprising multiple models centered on a base diffusion model. Existing serving systems treat each workflow as an opaque monolith, provisioning, placing, and scaling all constituent models together, which obscures internal dataflow, prevents model sharing, and enforces coarse-grained resource management. In this paper, we make a case for micro-serving diffusion workflows with LegoDiffusion, a system that decomposes a workflow into loosely coupled model-execution nodes that can be independently managed and scheduled. By explicitly managing individual model inference, LegoDiffusion unlocks cluster-scale optimizations, including per-model scaling, model sharing, and adaptive model parallelism. Collectively, LegoDiffusion outperforms existing diffusion workflow serving systems, sustaining up to 3x higher request rates and tolerating up to 8x higher burst traffic.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 26 of its 75 references that are in the map. First posted on arXiv 9 Apr 2026.

Builds on

26 of 75 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025TetriServe: Efficiently Serving Mixed DiT Workloads2 Oct
  2. 2024xDiT: an Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism4 Nov
  3. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  4. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  5. 2022Classifier-Free Diffusion Guidance26 Jul

Other references · 21

  1. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,141
  2. 2017Ray: A Distributed Framework for Emerging AI Applicationscited by 527
  3. 2024Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Servecited by 526
  4. 2024CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusioncited by 223
  5. 2024BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusioncited by 202
  6. 2016Clipper: A Low-Latency Online Prediction Serving Systemcited by 191
  7. 2024OOTDiffusion: Outfitting Fusion based Latent Diffusion for Controllable Virtual Try-oncited by 151
  8. 2020Serving DNNs like Clockwork: Performance Predictability from the Bottom Upcited by 120
  9. 2024LoongServe: Efficiently Serving Long-Context Large Language Models with Elastic Sequence Parallelismcited by 112
  10. 2024Parrot: Efficient Serving of LLM-based Applications with Semantic Variablecited by 99
  11. 2024Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flowcited by 55
  12. 2024MuxServe: Flexible Spatial-Temporal Multiplexing for Multiple LLM Servingcited by 52
9 more
  1. 2024Teola: Towards End-to-End Optimization of LLM-based Applicationscited by 32
  2. 2025Characterizing Mobile SoC for Accelerating Heterogeneous LLM Inferencecited by 22
  3. 2023DeltaZip: Efficient Serving of Multiple Full-Model-Tuned LLMscited by 18
  4. 2025JITServe: SLO-aware LLM Serving with Imprecise Request Informationcited by 17
  5. 2025IC-Cache: Efficient Large Language Model Serving via In-context Cachingcited by 13
  6. 2022Autothrottle: A Practical Bi-Level Approach to Resource Management for SLO-Targeted Microservicescited by 12
  7. 2021Following the Data, Not the Function: Rethinking Function Orchestration in Serverless Computingcited by 7
  8. 2024Towards Pareto Optimal Throughput in Small Language Model Servingcited by 5
  9. 2025TridentServe: A Stage-level Serving System for Diffusion Pipelinescited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024SwiftDiffusion: Efficient Diffusion Model Serving with Add-on Modules
  2. 2024Not All Prompts Are Made Equal: Prompt-based Pruning of Text-to-Image Diffusion Models
  3. 2026DiffBench Meets DiffAgent: End-to-End LLM-Driven Diffusion Acceleration Code Generation
  4. 2025Paris: A Decentralized Trained Open-Weight Diffusion Model
  5. 2025Reusing Computation in Text-to-Image Diffusion for Efficient Generation of Image Sets
  6. 2025Accelerating Parallel Diffusion Model Serving with Residual Compression
  7. 2026Accelerating Diffusion via Hybrid Data-Pipeline Parallelism Based on Conditional Guidance Scheduling
  8. 2026FlashDiff: Efficient Regional Execution and Scheduling for Diffusion Model Serving