mlsanityData to 14 Oct 2026

SRUM: Fine-Grained Self-Rewarding for Unified Multimodal Models

arXiv 2510.12784 ↗First posted 14 Oct 2025cs.CV cs.CLcited by 22

On the map: Synthetic data and self-improvement for LLMs → Self-improving MLLMs and synthetic data

Abstract (arXiv)

Recently, remarkable progress has been made in Unified Multimodal Models (UMMs), which integrate vision-language generation and understanding capabilities within a single framework. However, a model's strong visual understanding often fails to transfer to visual generation: it may correctly judge prompt-image alignment while failing to generate a faithful image from the same prompt. This raises a compelling question: Can a model improve itself by using its understanding module to reward its generation module? We introduce SRUM, a self-rewarding post-training framework directly applicable to existing UMMs of various designs. SRUM creates a feedback loop where the model's own understanding module acts as an internal ``evaluator'', providing corrective signals to improve generation without additional human-labeled data or external reward models. To provide comprehensive feedback, SRUM uses a global-local dual reward system: a global reward ensures overall visual semantics and layout, while a local reward refines fine-grained, object-level fidelity. SRUM shows strong generalization, boosting performance on T2I-CompBench from 82.18 to 88.37 and on T2I-ReasonBench from 43.82 to 46.75. Overall, our work establishes a powerful paradigm for enabling a UMM's understanding module to guide and enhance its own generation via self-rewarding.

Key facts. Cited by 22 papers in the map (data through 14 Oct 2026). Placed in Self-improving MLLMs and synthetic data, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 82 of its 91 references that are in the map. First posted on arXiv 14 Oct 2025.

Builds on

82 of 91 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025UniRL: Self-Improving Unified Multimodal Models via Supervised and Reinforcement Learning29 May
  2. 2024Calibrated Self-Rewarding Vision Language Models23 May

Other references · 80

  1. 2017Attention Is All You Needcited by 60,361
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  4. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  5. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  6. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  7. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  8. 2023Segment Anythingcited by 8,234
  9. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  10. 2023Visual Instruction Tuningcited by 7,869
  11. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  12. 2020Scaling Laws for Neural Language Modelscited by 6,586
38 more
  1. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
  2. 2022Classifier-Free Diffusion Guidancecited by 5,847
  3. 2022Flow Matching for Generative Modelingcited by 5,702
  4. 2025Qwen2.5-VL Technical Reportcited by 5,587
  5. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,703
  6. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,356
  7. 2021Zero-Shot Text-to-Image Generationcited by 4,288
  8. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,234
  9. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,181
  10. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
  11. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  12. 2022Training Compute-Optimal Large Language Modelscited by 3,019
  13. 2022InstructPix2Pix: Learning to Follow Image Editing Instructionscited by 2,611
  14. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,141
  15. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,080
  16. 2020GLU Variants Improve Transformercited by 1,804
  17. 2023MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Modelscited by 1,652
  18. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contextscited by 1,614
  19. 2017Accurate, Large Minibatch SGD: Training ImageNet in 1 Hourcited by 1,492
  20. 2019Root Mean Square Layer Normalizationcited by 1,480
  21. 2023GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpointscited by 1,346
  22. 2023MM-Vet: Evaluating Large Multimodal Models for Integrated Capabilitiescited by 1,239
  23. 2024How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suitescited by 1,050
  24. 2023ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generationcited by 1,045
  25. 2025Qwen-Image Technical Reportcited by 1,044
  26. 2023GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignmentcited by 972
  27. 2024Chameleon: Mixed-Modal Early-Fusion Foundation Modelscited by 927
  28. 2025Emerging Properties in Unified Multimodal Pretrainingcited by 897
  29. 2025Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scalingcited by 812
  30. 2024Emu3: Next-Token Prediction is All You Needcited by 781
  31. 2024Show-o: One Single Transformer to Unify Multimodal Understanding and Generationcited by 752
  32. 2024Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMscited by 735
  33. 2024MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategiescited by 723
  34. 2024Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusioncited by 681
  35. 2023MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editingcited by 624
  36. 2024Evaluating Text-to-Visual Generation with Image-to-Text Generationcited by 474
  37. 2024Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generationcited by 457
  38. 2025DanceGRPO: Unleashing GRPO on Visual Generationcited by 413

50 most cited of 80 All 80 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 19.

22 papers.

Sort:

By area:

  1. 2025Uniworld-V2: Reinforce Image Editing with Diffusion Negative-aware Finetuning and MLLM Implicit Feedbackcited by 79
  2. 2025Does Understanding Inform Generation in Unified Multimodal Models? From Analysis to Path Forwardcited by 10
  3. 2025Towards Reason-Informed Video Editing in Unified Models with Self-Reflective Learningcited by 7
  4. 2026UReason: Benchmarking Reasoning-to-Generation Alignment in Unified Multimodal Modelscited by 7
  5. 2026Ask, Solve, Generate: Self-Evolving Unified Multimodal Understanding and Generation via Self-Consistency Rewardscited by 7
  6. 2026Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Modelscited by 6
  7. 2026CoCo: Code as CoT for Text-to-Image Preview and Rare Concept Generationcited by 6
  8. 2026MetaPoint: Unlocking Precise Spatial Control in Agentic Visual Generationcited by 4
  9. 2026ReBind: Multi-Reference Video Editing via Structured Instructions with Explicit Reference Relationshipscited by 3
  10. 2026Dreaming in Flow: Generative Grounding Feedback for Self-Evolving Unified Multimodal Modelscited by 3
  11. 2026Seeing Touch from Motion: A Unified Modality-Aware Visuo-Tactile Policy with Tactile Motion Correlationcited by 2
  12. 2026Self-Corrected Image Generation with Explainable Latent Rewardscited by 1
10 more
  1. 2026LatentUMM: Dual Latent Alignment for Unified Multimodal Modelscited by 1
  2. 2026Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervisioncited by 0
  3. 2026Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generationcited by 0
  4. 2026TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Rewardcited by 0
  5. 2026Generative Residual Factorizationcited by 0
  6. 2026Mutually Adversarial Self-Training with Evolving Data for Unified Multimodal Modelscited by 0
  7. 2026UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvementcited by 0
  8. 2026Recursive Self-Improvement in Unified Multimodal Modelscited by 0
  9. 2026Self-Verifier Capture in Unified Visual Generators: Diagnosing and Repairing Closed-Loop Self-Rewardcited by 0
  10. 2026RSIgen: Reciprocal Self-Improvement for Agentic Visual Generationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025SUDER: Self-Improving Unified Large Multimodal Models for Understanding and Generation with Dual Self-Rewards
  2. 2025Unified Reward Model for Multimodal Understanding and Generation
  3. 2026A Model-Internal Protocol for Assessing Multimodal Models as Integrated Systems
  4. 2026Unified Personalized Reward Model for Vision Generation
  5. 2025Self-Rewarded Multimodal Coherent Reasoning Across Diverse Visual Domains
  6. 2025Multimodal RewardBench: Holistic Evaluation of Reward Models for Vision Language Models
  7. 2025InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward Model
  8. 2026SSL-R1: Self-Supervised Visual Reinforcement Post-Training for Multimodal Large Language Models