mlsanityData to 14 Oct 2026

From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning

arXiv 2504.16080 ↗First posted 22 Apr 2025cs.CVcited by 56

On the map: Diffusion models → Image generation, text-to-image and editing

Abstract (arXiv)

Recent text-to-image diffusion models achieve impressive visual quality through extensive scaling of training data and model parameters, yet they often struggle with complex scenes and fine-grained details. Inspired by the self-reflection capabilities emergent in large language models, we propose ReflectionFlow, an inference-time framework enabling diffusion models to iteratively reflect upon and refine their outputs. ReflectionFlow introduces three complementary inference-time scaling axes: (1) noise-level scaling to optimize latent initialization; (2) prompt-level scaling for precise semantic guidance; and most notably, (3) reflection-level scaling, which explicitly provides actionable reflections to iteratively assess and correct previous generations. To facilitate reflection-level scaling, we construct GenRef, a large-scale dataset comprising 1 million triplets, each containing a reflection, a flawed image, and an enhanced image. Leveraging this dataset, we efficiently perform reflection tuning on state-of-the-art diffusion transformer, FLUX.1-dev, by jointly modeling multimodal inputs within a unified framework. Experimental results show that ReflectionFlow significantly outperforms naive noise-level scaling methods, offering a scalable and compute-efficient solution toward higher-quality image synthesis on challenging tasks.

Key facts. Cited by 56 papers in the map (data through 14 Oct 2026). Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models. Builds on 55 of its 69 references that are in the map. First posted on arXiv 22 Apr 2025.

Builds on

55 of 69 references.

Closest ancestors · 11

Same direction, newest first

  1. 2025VisualCloze: A Universal Image Generation Framework via Visual In-Context Learning10 Apr
  2. 2025Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflection15 Mar
  3. 2024OmniEdit: Building Image Editing Generalist Models Through Specialist Supervision11 Nov
  4. 2024OmniGen: Unified Image Generation17 Sep
  5. 2024Hunyuan-DiT: A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding14 May
  6. 2023ToddlerDiffusion: Interactive Structured Image Generation with Cascaded Schr"odinger Bridge24 Nov
  7. 2023A Picture is Worth a Thousand Words: Principled Recaptioning Improves Image Generation25 Oct
  8. 2023GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment17 Oct
  9. 2023IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models13 Aug
  10. 2023Adding Conditional Control to Text-to-Image Diffusion Models10 Feb
  11. 2022InstructPix2Pix: Learning to Follow Image Editing Instructions17 Nov

Other references · 44

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  3. 2020Denoising Diffusion Implicit Modelscited by 9,781
  4. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  5. 2020Scaling Laws for Neural Language Modelscited by 6,586
  6. 2022Scalable Diffusion Models with Transformerscited by 6,054
  7. 2022Flow Matching for Generative Modelingcited by 5,702
  8. 2025Qwen2.5-VL Technical Reportcited by 5,587
  9. 2024GPT-4o System Cardcited by 4,588
  10. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,356
  11. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  12. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,181
32 more
  1. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  2. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  3. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,211
  4. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  5. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  6. 2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,160
  7. 2023Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generationcited by 977
  8. 2023Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesiscited by 890
  9. 2023Diffusion Model Alignment Using Direct Preference Optimizationcited by 790
  10. 2024Training Language Models to Self-Correct via Reinforcement Learningcited by 426
  11. 2023Language Models can Solve Computer Taskscited by 372
  12. 2024SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformerscited by 340
  13. 2024OminiControl: Minimal and Universal Control for Diffusion Transformercited by 340
  14. 2024PixArt-Sigma: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generationcited by 329
  15. 2023Alphazero-like Tree-Search can Guide Large Language Model Decoding and Trainingcited by 295
  16. 2025Improving Video Generation with Human Feedbackcited by 284
  17. 2022Generating Sequences by Learning to Self-Correctcited by 277
  18. 2025Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Stepscited by 234
  19. 2025A General Framework for Inference-time Scaling and Steering of Diffusion Modelscited by 224
  20. 2024Recursive Introspection: Teaching Language Model Agents How to Self-Improvecited by 165
  21. 2025SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformercited by 165
  22. 2023Prodigy: An Expeditiously Adaptive Parameter-Free Learnercited by 133
  23. 2024Lumina-Next: Making Lumina-T2X Stronger and Faster with Next-DiTcited by 132
  24. 2025Can We Generate Images with CoT? Let's Verify and Reinforce Image Generation Step by Stepcited by 131
  25. 2024Playground v3: Improving Text-to-Image Alignment with Deep-Fusion Large Language Modelscited by 116
  26. 2024TFG: Unified Training-Free Guidance for Diffusion Modelscited by 107
  27. 2024GLoRe: When, Where, and How to Improve LLM Reasoning via Global and Local Refinementscited by 103
  28. 2025Lumina-Image 2.0: A Unified and Efficient Image Generative Frameworkcited by 96
  29. 2024Golden Noise for Diffusion Models: A Learning Frameworkcited by 92
  30. 2024Not All Noises Are Created Equally:Diffusion Noise Selection and Optimizationcited by 71
  31. 2024A Noise is Worth Diffusion Guidancecited by 49
  32. 2024Getting it Right: Improving Spatial Consistency in Text-to-Image Modelscited by 30

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 20; 2026: 36.

56 papers.

Sort:

By area:

  1. 2025Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understandingcited by 88
  2. 2025MINT-CoT: Enabling Interleaved Visual Tokens in Mathematical Chain-of-Thought Reasoningcited by 49
  3. 2025Interleaving Reasoning for Better Text-to-Image Generationcited by 41
  4. 2025Lumina-mGPT 2.0: Stand-Alone AutoRegressive Image Modelingcited by 37
  5. 2026Mind-Brush: Integrating Agentic Cognitive Search and Reasoning into Image Generationcited by 23
  6. 2026GEMS: Agent-Native Multimodal Generation with Memory and Skillscited by 21
  7. 2025Factuality Matters: When Image Generation and Editing Meet Structured Visualscited by 20
  8. 2025Thinking-while-Generating: Interleaving Textual Reasoning throughout Visual Generationcited by 18
  9. 2025Lavida-O: Elastic Large Masked Diffusion Models for Unified Multimodal Understanding and Generationcited by 15
  10. 2026RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Timecited by 14
  11. 2025DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generationcited by 13
  12. 2026LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Modelcited by 10
38 more
  1. 2025Resurrect Mask AutoRegressive Modeling for Efficient and Scalable Image Generationcited by 9
  2. 2026CoF-T2I: Video Models as Pure Visual Reasoners for Text-to-Image Generationcited by 9
  3. 2025MILR: Improving Multimodal Image Generation via Test-Time Latent Reasoningcited by 8
  4. 2026UniT: Unified Multimodal Chain-of-Thought Test-time Scalingcited by 8
  5. 2026From Scale to Speed: Adaptive Test-Time Scaling for Image Editingcited by 8
  6. 2026Agentic Retoucher for Text-To-Image Generationcited by 7
  7. 2026RAISE: Requirement-Adaptive Evolutionary Refinement for Training-Free Text-to-Image Alignmentcited by 7
  8. 2026Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoningcited by 7
  9. 2026SCOPE: Structured Decomposition and Conditional Skill Orchestration for Complex Image Generationcited by 7
  10. 2025IRG-MotionLLM: Interleaving Motion Generation, Assessment and Refinement for Text-to-Motion Generationcited by 6
  11. 2026SPIRAL: Self-Evolving Action-Conditioned Video Generation via Reflective Planning Agentscited by 5
  12. 2026Tiny Inference-Time Scaling with Latent Verifierscited by 4
  13. 2026Stream-T1: Test-Time Scaling for Streaming Video Generationcited by 4
  14. 2026From Statics to Dynamics: Physics-Aware Image Editing with Latent Transition Priorscited by 3
  15. 2026Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasonerscited by 3
  16. 2025dMLLM-TTS: Self-Verified and Efficient Test-Time Scaling for Diffusion Multi-Modal Large Language Modelscited by 2
  17. 2026EPIC: Efficient Predicate-Guided Inference-Time Control for Compositional Text-to-Image Generationcited by 2
  18. 2026OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibrationcited by 2
  19. 2026Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedbackcited by 2
  20. 2026InterleaveThinker: Reinforcing Agentic Interleaved Generationcited by 2
  21. 2025Visual Autoregressive Models Beat Diffusion Models on Inference Time Scalingcited by 1
  22. 2025Compositional Image Synthesis with Inference-Time Scalingcited by 1
  23. 2025Diversifying Long Prompt Image Generation through Structured Prompt Embedding Space Samplingcited by 1
  24. 2025ShapeGen: Towards High-Quality 3D Shape Synthesiscited by 1
  25. 2025Rethinking Test Time Scaling for Flow-Matching Generative Modelscited by 1
  26. 2025Verifier Threshold: An Efficient Test-Time Scaling Approach for Image Generationcited by 1
  27. 2026Exploiting the Final Component of Generator Architectures for AI-Generated Image Detectioncited by 1
  28. 2026Unlocking Complex Visual Generation via Closed-Loop Verified Reasoningcited by 1
  29. 2026Don't Settle at the Mode! Mitigating Diversity Collapse in Pretrained Flow Models via Feature Self-Guidancecited by 1
  30. 2026Calibrate What You SHIP: Post-Selection Risk Control for Verifier-Guided Text-to-Image Generationcited by 1
  31. 2026Step Back to Move Forward: Reflection-Aware Preference Optimization for Visual Generationcited by 1
  32. 2026On-Policy Self-Distillation for Multi-Turn Image Editingcited by 1
  33. 2025Performance Plateaus in Inference-Time Scaling for Text-to-Image Diffusion Without External Modelscited by 0
  34. 2025EvoMail: Self-Evolving Cognitive Agents for Adaptive Spam and Phishing Email Defensecited by 0
  35. 2026Inline Critic Steers Image Editingcited by 0
  36. 2026Compositional Video Generation via Inference-Time Guidancecited by 0
  37. 2026RT-Lynx: Putting GEMM Sparsity in the Right Place for Diffusion Modelscited by 0
  38. 2026MathVis-Fine: Aligning Visual Supervision with Necessity via Progressive Dependency-Guided Training for Multimodal Mathematical Reasoningcited by 0

50 most cited of 56 All 56 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Abra: Scaling Diffusion Image Training
  2. 2026i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models
  3. 2024Zigzag Diffusion Sampling: Diffusion Models Can Self-Improve via Self-Reflection
  4. 2026Scaling Text-to-Image Diffusion Transformers with Representation Autoencoders
  5. 2024Bigger is not Always Better: Scaling Properties of Latent Diffusion Models
  6. 2024Enhancing Semantic Fidelity in Text-to-Image Synthesis: Attention Regulation in Diffusion Models
  7. 2023DiffFit: Unlocking Transferability of Large Diffusion Models via Simple Parameter-Efficient Fine-Tuning
  8. 2023Semantic Guidance Tuning for Text-To-Image Diffusion Models