Stable-Layers: Fine-Tuning Image Layer Decomposition Models with VLM-Scored Reinforcement Learning
On the map: Multimodal LLMs / vision-language models → Embodied and 3D vision-language models
Abstract (arXiv)
We present Stable-Layers, a reinforcement learning framework that eliminates the need for paired supervision by fine-tuning a pretrained layer decomposition model using only feedback from a vision-language model (VLM). Starting from Qwen-Image-Layered, we apply Flow-GRPO with LoRA adaptation, sampling multiple candidate decompositions per image, scoring them with a VLM, and optimising the policy from group-relative advantages. The key challenge lies in designing a reliable reward signal: VLMs scoring samples in isolation tend to compress their judgements into a narrow band, leaving GRPO with little within-group variance to learn from. We address this with a two-stage evaluation pipeline that pairs structured per-sample scoring across five edit-centric criteria with a grid-based calibration step in which the VLM re-scores all candidates side-by-side. Stable-Layers produces decompositions with stronger layer separation, fewer blank or artifact-heavy layers, and lower per-layer reconstruction error on the Crello dataset compared to the base model.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 37 of its 39 references that are in the map. First posted on arXiv 28 May 2026.
Builds on
37 of 39 references.
Closest ancestors · 1
Same direction, newest first
Other references · 36
- 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
- 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
- 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
- 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
- 2023ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generationcited by 1,045
- 2023Training Diffusion Models with Reinforcement Learningcited by 982
- 2023Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesiscited by 890
- 2023Diffusion Model Alignment Using Direct Preference Optimizationcited by 790
- 2025Flow-GRPO: Training Flow Matching Models via Online RLcited by 661
- 2023DPOK: Reinforcement Learning for Fine-tuning Text-to-Image Diffusion Modelscited by 450
- 2023Directly Fine-Tuning Diffusion Models on Differentiable Rewardscited by 448
- 2025DanceGRPO: Unleashing GRPO on Visual Generationcited by 413
24 more
- 2023Aligning Text-to-Image Diffusion Models with Reward Backpropagationcited by 288
- 2022Rectified Flow: A Marginal Preserving Approach to Optimal Transportcited by 283
- 2024VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generationcited by 156
- 2024Direct Preference Optimization of Video Large Multimodal Models from Language Model Rewardcited by 147
- 2024Transparent Image Layer Diffusion using Latent Transparencycited by 114
- 2021CanvasVAE: Learning to Generate Vector Graphic Documentscited by 95
- 2024MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?cited by 66
- 2025GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clippingcited by 49
- 2024Generative Image Layer Decomposition with Visual Effectscited by 34
- 2025Qwen-Image-Layered: Towards Inherent Editability via Layer Decompositioncited by 34
- 2026TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Roboticscited by 33
- 2024LayerDiff: Exploring Text-guided Multi-layered Composable Image Synthesis via Layer-Collaborative Diffusion Modelcited by 26
- 2025LayerD: Decomposing Raster Graphic Designs into Layerscited by 21
- 2024LayerFusion: Harmonized Multi-Layer Text-to-Image Generation with Generative Priorscited by 19
- 2025DreamLayer: Simultaneous Multi-Layer Generation via Diffusion Modecited by 17
- 2025LayeringDiff: Layered Image Synthesis via Generation, then Disassembly with Generative Knowledgecited by 16
- 2025PSDiffusion: Harmonized Multi-Layer Image Generation via Layout and Appearance Alignmentcited by 12
- 2025Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Rankingcited by 10
- 2025Controllable Layer Decomposition for Reversible Multi-Layer Image Generationcited by 9
- 2026Fine-T2I: An Open, Large-Scale, and Diverse Dataset for High-Quality T2I Fine-Tuningcited by 5
- 2025From Inpainting to Layer Decomposition: Repurposing Generative Inpainting Models for Image Layer Decompositioncited by 4
- 2026See-through: Single-image Layer Decomposition for Anime Characterscited by 3
- 2026Referring Layer Decompositioncited by 3
- 2026Workflow-Aware Structured Layer Decomposition for Illustration Productioncited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2026Harnessing Image Question Dependence for Better VLM Test-time Reinforcement Learning
- 2026Enhancing VLM Reward Models Through Structure-Aware Fine-Tuning
- 2025Stabilizing Reinforcement Learning with LLMs: Formulation and Practices
- 2026λ-Controlled GRPO: Turning Flow-Matching Ratio Instability into a Budgeted Resource
- 2025Reinforcement Fine-Tuning of Flow-Matching Policies for Vision-Language-Action Models
- 2024Code as Reward: Empowering Reinforcement Learning with VLMs
- 2025Offline RLAIF: Piloting VLM Feedback for RL via SFO
- 2026Rethinking Reinforcement Fine-Tuning in LVLM: Convergence, Reward Decomposition, and Generalization