| MPO: Single-Stream Policy Optimization for Efficient Text-to-Image Alignment | | 0.98 | | | | | | | FLUX.1 Dev + MPO | 2026-09-08 |
| On-Policy Diffusion Reinforcement Learning Meets Off-Policy Quality Anchoring | ✓ Link | 0.97 | | | | | | | DiffusionCompass | 2026-09-08 |
| UDM-GRPO: Stable and Efficient Group Relative Policy Optimization for Uniform Discrete Diffusion Models | ✓ Link | 0.96 | | | | | | | URSA (1.7B, UDM-GRPO) | 2026-04-20 |
| DiffusionOPD: A Unified Perspective of On-Policy Distillation in Diffusion Models | ✓ Link | 0.96 | | | | | | | DiffusionOPD | 2026-05-14 |
| Flow-GRPO: Training Flow Matching Models via Online RL | ✓ Link | 0.95 | | | | | | | SD3.5-Medium+Flow-GRPO | 2025-05-08 |
| GARDO: Reinforcing Diffusion Models without Reward Hacking | ✓ Link | 0.95 | | | | | | | GARDO (SD3.5-M, β=0.04, 2000 steps) | 2025-12-30 |
| VisionDirector: Vision-Language Guided Closed-Loop Refinement for Generative Image Synthesis | | 0.94 | | | | | | | VisionDirector | 2025-12-22 |
| PromptRL: Prompt Matters in RL for Flow-Based Image Generation | ✓ Link | 0.94 | | | | | | | PromptRL (without prompt enhancement) | 2026-02-01 |
| Reinforcing Few-step Generators via Reward-Tilted Distribution Matching | ✓ Link | 0.94 | | | | | | | RTDMD (SD3.5-M) | 2026-05-25 |
| TDM-R1: Reinforcing Few-Step Diffusion Models with Non-Differentiable Reward | ✓ Link | 0.92 | | | | | | | TDM-R1 | 2026-03-08 |
| Flow-OPD: On-Policy Distillation for Flow Matching Models | ✓ Link | 0.92 | | | | | | | Flow-OPD | 2026-05-08 |
| HiDream-O1-Image: A Natively Unified Image Generative Foundation Model with Pixel-level Unified Transformer | ✓ Link | 0.92 | | | | | | | HiDream-O1-Image-Pro | 2026-05-11 |
| GR-GRPO: Graph-Diffused Credit for Autoregressive Image RL Alignment | ✓ Link | 0.92 | | | | | | | GR-GRPO-7B | 2026-09-08 |
| SenseNova-U1.5: Towards Native Unified Visual Intelligence | ✓ Link | 0.92 | | | | | | | SenseNova-U1.5 | 2026-09-10 |
| Qwen-Image Technical Report | ✓ Link | 0.91 | | | | | | | Qwen-Image-RL | 2025-08-04 |
| Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understanding | ✓ Link | 0.91 | | | | | | | Lumina-DiMOO w/ Self-GRPO | 2025-10-07 |
| BLIP3o-NEXT: Next Frontier of Native Image Generation | ✓ Link | 0.91 | | | | | | | BLIP3o-NEXT-GRPO-GenEval (3B) | 2025-10-17 |
| LightFusion: A Light-weighted, Double Fusion Framework for Unified Multimodal Understanding and Generation | | 0.91 | | | | | | | LightFusion | 2025-10-27 |
| EMMA: Efficient Multimodal Understanding, Generation, and Editing with a Unified Architecture | ✓ Link | 0.91 | | | | | | | EMMA (4B, original prompts) | 2025-12-04 |
| STAR: STacked AutoRegressive Scheme for Unified Multimodal Learning | | 0.91 | | | | | | | STAR-7B | 2025-12-15 |
| LVRPO: Language-Visual Alignment with GRPO for Multimodal Understanding and Generation | | 0.91 | | | | | | | LVRPO | 2026-03-29 |
| RewardFlow: Generate Images by Optimizing What You Reward | | 0.91 | | | | | | | Qwen + RewardFlow | 2026-04-09 |
| Refinement via Regeneration: Enlarging Modification Space Boosts Image Refinement in Unified Multimodal Models | ✓ Link | 0.91 | | | | | | | BAGEL-RvR | 2026-04-28 |
| SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture | ✓ Link | 0.91 | | | | | | | SenseNova-U1-A3B-MoT | 2026-05-12 |
| SPAR: Semantic-Pixel Self-Alignment and Adaptive Routing for Unified Multimodal Models | | 0.91 | | | | | | | SPAR-3B (extra data) | 2026-06-22 |
| Intrinsic Distribution Matching | ✓ Link | 0.91 | | | | | | | IDM (Qwen-Image) (Full) (2 NFE) (extra data) | 2026-10-29 |
| UniLiP: Adapting CLIP for Unified Multimodal Understanding, Generation and Editing | ✓ Link | 0.90 | | | | | | | UniLIP-3B | 2025-07-31 |
| OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation | | 0.90 | | | | | | | OneCAT-3B | 2025-09-03 |
| Reconstruction Alignment Improves Unified Multimodal Models | ✓ Link | 0.90 | | | | | | | Harmon-1.5B-RecA-plus (extra data) | 2025-09-08 |
| TUNA: Taming Unified Visual Representations for Native Unified Multimodal Models | | 0.90 | | | | | | | Tuna (7B) | 2025-12-01 |
| Understanding and Harnessing Sparsity in Unified Multimodal Models | ✓ Link | 0.9 | | | | | | | Qwen-Image with Neuron Partition (50%) | 2025-12-02 |
| TwinFlow: Realizing One-step Generation on Large Models with Self-adversarial Flows | ✓ Link | 0.9 | | | | | | | Qwen-Image-TwinFlow (full, longer training) | 2025-12-03 |
| UniReason 1.0: A Unified Reasoning Framework for World Knowledge Aligned Image Generation and Editing | | 0.90 | | 0.96 | 0.82 | | 0.82 | 0.88 | UniReason 1.0 | 2026-02-02 |
| UniGRPO: Unified Policy Optimization for Reasoning-Driven Visual Generation | | 0.9 | | | | | | | UniGRPO | 2026-03-24 |
| Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation | ✓ Link | 0.9 | | | | | | | EMF (3B) | 2026-04-20 |
| HiDream-O1-Image: A Natively Unified Image Generative Foundation Model with Pixel-level Unified Transformer | ✓ Link | 0.90 | | | | | | | HiDream-O1-Image | 2026-05-11 |
| Lance: Unified Multimodal Modeling by Multi-Task Synergy | ✓ Link | 0.9 | | | | | | | Lance (3B) | 2026-05-18 |
| Nemotron-Labs-Diffusion-Image: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis | | 0.90 | | | | | | | Nemotron-Labs-Diffusion-Image | 2026-06-29 |
| Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing | ✓ Link | 0.90 | | | | | | | Mage-Flow | 2026-07-21 |
| GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling | | 0.9 | | | | | | | GenFirst EiT1.3B CLIPtextencoder | 2026-08-29 |
| Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation | ✓ Link | 0.898 | | | | | | | Self-SpectraReward | 2026-07-13 |
| Echo-4o: Harnessing the Power of GPT-4o Synthetic Images for Improved Image Generation | ✓ Link | 0.895 | | | | | | | Echo-4o | 2025-08-13 |
| Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training | ✓ Link | 0.89 | | | | | | | IOMM-B 512 | 2026-03-17 |
| Self-Adversarial One Step Generation via Condition Shifting | ✓ Link | 0.89 | | | | | | | APEX 20B (LoRA, r=64) | 2026-04-14 |
| LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Model | ✓ Link | 0.89 | | | | | | | LLaDA2.0-Uni (16B) | 2026-04-22 |
| Uni-Edit: Intelligent Editing Is A General Task For Unified Model Tuning | ✓ Link | 0.89 | | | | | | | BAGEL-Uni-Edit (Ours) | 2026-05-20 |
| Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization | ✓ Link | 0.89 | | | | | | | Janus-Pro-R1-7B + GCPO | 2026-05-28 |
| Learning Representations to Guide Pixel Diffusion Transformers | ✓ Link | 0.89 | | | | | | | RG-T2I-L/16 | 2026-10-09 |
| Intern Lumina U2: A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding and Image Generation | ✓ Link | 0.89 | | | | | | | Intern Lumina U2 (Ascend) | 2026-10-14 |
| Nemotron-Labs-Diffusion-Uni: Elastic Self-Refining Unified Diffusion LLM with Cross-Paradigm On-Policy Distillation | | 0.89 | | | | | | | NLD-Uni | 2026-10-27 |
| Enhancing Alignment for Unified Multimodal Models via Semantically-Grounded Supervision | | 0.8866 | | | | | | | SeGroS-Harmon-1.5B (extra data) | 2026-03-20 |
| IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation | | 0.885 | | | | | | | Full IV-CoT (OpenUni-L-1024, 512 queries) | 2026-06-23 |
| MiniT2I: A Minimalist Baseline for Text-to-Image Generation | ✓ Link | 0.883 | | | | | | | MiniT2I-L/16 | 2026-06-14 |
| UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement | | 0.882 | | | | | | | UniEvo-VL (GPT5.6-Luna) | 2026-09-30 |
| UniLiP: Adapting CLIP for Unified Multimodal Understanding, Generation and Editing | ✓ Link | 0.88 | | | | | | | UniLIP-1B | 2025-07-31 |
| Query-Kontext: An Unified Multimodal Model for Image Generation and Editing | | 0.88 | | | | | | | Query-Kontext | 2025-09-30 |
| TUNA: Taming Unified Visual Representations for Native Unified Multimodal Models | | 0.88 | | | | | | | Tuna (1.5B) | 2025-12-01 |
| Chain-of-Trajectories: Unlocking the Intrinsic Generative Optimality of Diffusion Models via Graph-Theoretic Planning | ✓ Link | 0.88 | | | | | | | CoTj + Qwen-Image (50 steps) | 2026-03-16 |
| Self-Adversarial One Step Generation via Condition Shifting | ✓ Link | 0.88 | | | | | | | APEX 20B (LoRA, r=32) | 2026-04-14 |
| Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning | | 0.88 | | | | | | | CLVR (9B) | 2026-05-14 |
| DuoGen: Towards Autonomous Interleaved Multimodal Generation | | 0.88 | | | | | | | DuoGen | 2026-06-03 |
| HYDRA-X: Native Unified Multimodal Models with Holistic Visual Tokenizers | | 0.88 | | | | | | | Hydra-X | 2026-06-11 |
| JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces | ✓ Link | 0.88 | | | | | | | JUICE (DINO) | 2026-10-20 |
| HiRAE: Hierarchical Representation Autoencoding with Residual Budgets | | 0.877 | | | | | | | HiRAE-24 | 2026-09-29 |
| Aligned but Stereotypical? The Hidden Influence of System Prompts on Social Bias in LVLM-Based Text-to-Image Models | ✓ Link | 0.874 | | | | | | | Qwen-Image + FairPro | 2025-12-04 |
| BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image Diffusion | ✓ Link | 0.874 | | | | | | | BudgetPix (MiniT2I-L/16, 25% budget) | 2026-10-08 |
| Simplifying and Improving On-Policy Distillation for Few-Step Diffusion | | 0.873 | | | | | | | SMD (Qwen-Image, 4-NFE) | 2026-10-13 |
| ParaUni: Enhance Generation in Unified Multimodal Model with Reinforcement-driven Hierarchical Parallel Information Interaction | ✓ Link | 0.87 | | | | | | | ParaUni | 2025-12-05 |
| LongCat-Image Technical Report | ✓ Link | 0.87 | | | | | | | LongCat-Image | 2025-12-08 |
| DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editing | ✓ Link | 0.87 | | | | | | | DeepGen-1.0 (RL) | 2026-02-12 |
| Dynin-Omni: Omnimodal Unified Large Diffusion Language Model | ✓ Link | 0.87 | | | | | | | Dynin-Omni | 2026-03-09 |
| UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations | | 0.87 | | | | | | | UniCom | 2026-03-11 |
| Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training | ✓ Link | 0.87 | | | | | | | IOMM-B 1024 | 2026-03-17 |
| Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training | ✓ Link | 0.87 | | | | | | | IOMM-L 512 | 2026-03-17 |
| LatentUM: Unleashing the Potential of Interleaved Cross-Modal Reasoning via a Latent-Space Unified Model | | 0.87 | | | | | | | LatentUM Vis-Gen | 2026-04-02 |
| Nucleus-Image: Sparse MoE for Image Generation | ✓ Link | 0.87 | | | | | | | Nucleus-Image | 2026-04-14 |
| Tuna-2: Pixel Embeddings Beat Vision Encoders for Multimodal Understanding and Generation | ✓ Link | 0.87 | | | | | | | Tuna-2 7B | 2026-04-27 |
| Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning | | 0.87 | | | | | | | FLUX.2 4B + CLVR + PPRL + DSWM | 2026-05-14 |
| RepFusion: Leveraging Multimodal Priors for Denoising in Representation Space | | 0.87 | | | | | | | RepFusion-SFT w/ Diffusion Decoder | 2026-06-12 |
| UniDDT: Unifying Multimodal Understanding and Generation with Decoupled Diffusion Transformer | | 0.87 | | | | | | | VLM-UniDDT | 2026-06-15 |
| Multi-Resolution Flow Matching: Training-Free Diffusion Acceleration via Staged Sampling | ✓ Link | 0.87 | | | | | | | MrFlow + Qwen-Image (20+1 steps) | 2026-07-02 |
| JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces | ✓ Link | 0.87 | | | | | | | JUICE (SigLIP2) | 2026-10-20 |
| Skywork UniPic: Unified Autoregressive Modeling for Visual Understanding and Generation | ✓ Link | 0.86 | | | | | | | Skywork UniPic | 2025-08-05 |
| Emu3.5: Native Multimodal Models are World Learners | ✓ Link | 0.86 | | | | | | | Emu3.5 (34B, AR) | 2025-10-30 |
| DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generation | ✓ Link | 0.86 | | | | | | | DeCo-XXL/16 | 2025-11-24 |
| DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generation | ✓ Link | 0.86 | | | | | | | DraCo | 2025-12-04 |
| STAR: STacked AutoRegressive Scheme for Unified Multimodal Learning | | 0.86 | | | | | | | STAR-3B | 2025-12-15 |
| CoF-T2I: Video Models as Pure Visual Reasoners for Text-to-Image Generation | ✓ Link | 0.86 | | | | | | | CoF-T2I (14B) | 2026-01-15 |
| BitDance: Scaling Autoregressive Generative Models with Binary Tokens | ✓ Link | 0.86 | | | | | | | BitDance-14B-16x | 2026-02-15 |
| RaPD: Resolution-Agnostic Pixel Diffusion via Semantics-Enriched Implicit Representations | | 0.86 | | | | | | | RaPD (256→512) | 2026-05-15 |
| RebRL: Reinforcing Discrete Visual Diffusion Models with Rebalanced Timestep Credits | | 0.86 | | | | | | | MMaDA-8B-Base w/ RebRL (ours) (extra data) | 2026-06-03 |
| ARM: An AutoRegressive Large Multimodal Model with Unified Discrete Representations | ✓ Link | 0.86 | | | | | | | ARM-RL-7B | 2026-06-09 |
| Parallel Decoding Distillation for Fast Image and Video Generation | ✓ Link | 0.86 | | | | | | | PDD Midpoint (Qwen-Image, 4 NFE) | 2026-07-28 |
| Beyond the Estimate: Wiener-Informed Decoding for Easier Training of Pixel-Space Flow Models | | 0.86 | | | | | | | DeCo-XXL + WiD | 2026-10-26 |
| APT: Feedback-Driven Agentic Post-Training for Text-to-Image Generation | ✓ Link | 0.859 | | | | | | | Lens (w/ APT) | 2026-10-13 |
| OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation | | 0.85 | | | | | | | OneCAT-1.5B | 2025-09-03 |
| Interleaving Reasoning for Better Text-to-Image Generation | ✓ Link | 0.85 | | | | | | | IRG | 2025-09-08 |
| Generating an Image From 1,000 Words: Enhancing Text-to-Image With Structured Captions | ✓ Link | 0.85 | | | | | | | FIBO | 2025-11-10 |
| ChatUMM: Robust Context Tracking for Conversational Interleaved Generation | | 0.85 | | | | | | | ChatUMM | 2026-02-06 |
| InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing | ✓ Link | 0.85 | | | | | | | InternVL-U (4B) | 2026-03-10 |
| Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training | ✓ Link | 0.85 | | | | | | | IOMM-L 1024 | 2026-03-17 |
| LatentUM: Unleashing the Potential of Interleaved Cross-Modal Reasoning via a Latent-Space Unified Model | | 0.85 | | | | | | | LatentUM Base | 2026-04-02 |
| Beyond Text Prompts: Visual-to-Visual Generation as A Unified Paradigm | | 0.85 | | | | | | | Qwen-Image-V2V-Zero | 2026-05-12 |
| RaPD: Resolution-Agnostic Pixel Diffusion via Semantics-Enriched Implicit Representations | | 0.85 | | | | | | | RaPD | 2026-05-15 |
| Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification | ✓ Link | 0.85 | | | | | | | UniAR (original prompts) | 2026-06-16 |
| Illuminating Unified Multimodal Model for Free-form Interleaved Text-Image Generation | ✓ Link | 0.85 | | | | | | | ILLUME-X | 2026-06-29 |
| Boogu-Image-0.1: Boosting Open-Source Unified Multimodal Understanding and Generation | ✓ Link | 0.85 | | | | | | | Boogu-Image-0.1-Base | 2026-07-14 |
| Energy-Guided Flow Matching | ✓ Link | 0.85 | | | | | | | EG-FM-T2I | 2026-08-06 |
| LinCa: Accelerating Diffusion Models via Learnable Decomposed Feature Caching | ✓ Link | 0.85 | | | | | | | LinCa (Qwen-Image-Lightning, N=3) | 2026-08-18 |
| LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes | ✓ Link | 0.85 | | | | | | | LLaDA-Image | 2026-09-03 |
| DanceOPD: On-Policy Generative Field Distillation | ✓ Link | 0.849 | | | | | | | DanceOPD (Ours) | 2026-06-25 |
| Pluggable Pruning with Contiguous Layer Distillation for Diffusion Transformers | ✓ Link | 0.847 | | | | | | | PPCL (Qwen-Image, 14B) | 2025-11-20 |
| RubricRL: Simple Generalizable Rewards for Text-to-Image Generation | | 0.8468 | | | | | | | RubricRL (Phi3-3.8B) (extra data) | 2025-11-25 |
| Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models | ✓ Link | 0.843 | | | | | | | Lens (20-step, no reasoner) | 2026-05-20 |
| BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset | ✓ Link | 0.84 | | | | | | | BLIP3-o (8B) | 2025-05-14 |
| UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation | ✓ Link | 0.84 | 0.98 | 0.93 | 0.71 | 0.90 | 0.81 | 0.74 | UniWorld-V1 (Rewrite) | 2025-06-03 |
| WEAVE: Unleashing and Benchmarking the In-context Interleaved Comprehension and Generation | ✓ Link | 0.84 | | | | | | | Bagel + WEAVE-100k | 2025-11-14 |
| Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer | ✓ Link | 0.84 | | | | | | | Z-Image | 2025-11-27 |
| Ovis-Image Technical Report | ✓ Link | 0.84 | | | | | | | Ovis-Image (2B+7B) | 2025-11-28 |
| Visual-Aware CoT: Achieving High-Fidelity Visual Consistency in Unified Models | | 0.84 | | | | | | | Visual-Aware CoT | 2025-12-22 |
| LongCat-Next: Lexicalizing Modalities as Discrete Tokens | ✓ Link | 0.84 | | | | | | | LongCat-Next | 2026-03-29 |
| Representation Forcing for Bottleneck-Free Unified Multimodal Models | | 0.84 | | | | | | | RF-Pixel (original prompts) | 2026-05-29 |
| i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models | ✓ Link | 0.84 | | | | | | | i1 | 2026-06-09 |
| UniSpace: Unified Visual Representation and Scalable Multimodal Modeling | ✓ Link | 0.84 | | | | | | | UniSpace | 2026-08-09 |
| Learning Native Reflection in Unified Models with Interleaved Reinforcement Learning | ✓ Link | 0.84 | | | | | | | UMM-Reflection | 2026-09-28 |
| DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergence | ✓ Link | 0.84 | | | | | | | 1.6B-DiT (extra data) | 2026-09-30 |
| FLAT: Resampling Image and Text into 1D Flexible-Length Aligned Transmodal Tokens for Retrieval and Generation | | 0.831 | | | | | | | FLAT | 2026-09-15 |
| MindOmni: Unleashing Reasoning Generation in Vision Language Models with RGPO | ✓ Link | 0.83 | 0.99 | 0.94 | 0.71 | 0.90 | 0.71 | 0.71 | MindOmni | 2025-05-19 |
| X-Omni: Reinforcement Learning Makes Discrete Autoregressive Image Generative Models Great Again | | 0.83 | | | | | | | X-Omni | 2025-07-29 |
| Uni-cot: Towards Unified Chain-of-Thought Reasoning Across Text and Vision | | 0.83 | | | | | | | Uni-CoT | 2025-08-07 |
| Uni-X: Mitigating Modality Conflict with a Two-End-Separated Architecture for Unified Multimodal Models | ✓ Link | 0.83 | | | | | | | Uni-X (240B) | 2025-09-29 |
| Think in Strokes, Not Pixels: Process-Driven Image Generation via Interleaved Reasoning | | 0.83 | | | | | | | Process-driven BAGEL-7B | 2026-04-08 |
| Representation Distribution Matching for One-Step Visual Generation | ✓ Link | 0.826 | | | | | | | iRDM (1-step) | 2026-07-02 |
| RxBrain: Embodied Cognition Foundation Model with Joint Language-Visual Reasoning and Imagination | ✓ Link | 0.824 | | | | | | | RxBrain | 2026-07-15 |
| Emerging Properties in Unified Multimodal Pretraining | ✓ Link | 0.82 | | | | | | | BAGEL (7B MoT) | 2025-05-20 |
| Uni-X: Mitigating Modality Conflict with a Two-End-Separated Architecture for Unified Multimodal Models | ✓ Link | 0.82 | | | | | | | Uni-X (140B) | 2025-09-29 |
| UniCorn: Towards Self-Improving Unified Multimodal Models through Self-Generated Supervision | ✓ Link | 0.82 | | | | | | | UniCorn | 2026-01-06 |
| LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model | ✓ Link | 0.82 | | | | | | | LLaDA-o | 2026-03-01 |
| STARFlow2: Bridging Language Models and Normalizing Flows for Unified Multimodal Generation | | 0.82 | | | | | | | STARFlow2 (10.6B) | 2026-05-08 |
| Normalizing Trajectory Models | | 0.82 | | | | | | | NTM (from scratch, 256×256) | 2026-05-08 |
| Asymmetric Flow Models | ✓ Link | 0.82 | | | | | | | AsymFLUX.2 klein | 2026-05-13 |
| InsightTok: Improving Text and Face Fidelity in Discrete Tokenization for Autoregressive Image Generation | ✓ Link | 0.82 | | | | | | | InsightAR | 2026-05-14 |
| C3-UniMM: Causal Cycle-Consistent Unified Multimodal Modeling via Super Alignment and Shared Decoding Space | | 0.82 | | | | | | | C3-UniMM (7B) | 2026-07-04 |
| Chimera: Designing and Chinchilla-Scaling Hybrid Visual Diffusion Transformers | | 0.82 | | | | | | | Chimera | 2026-07-30 |
| MANGO: Unleashing Image Generation Capability of Unified Multimodal Models | | 0.82 | | | | | | | MANGO-4B | 2026-09-08 |
| Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding Spaces | ✓ Link | 0.82 | | | | | | | MF-1 | 2026-09-30 |
| OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World Modeling | ✓ Link | 0.82 | | | | | | | Qwen-Image-2512 + RL(Decomposed Graded Verifier) | 2026-10-15 |
| An Empirical Study of Training Pixel-Space Text-to-Image Diffusion Models | | 0.8185 | | | | | | | Ours-Turbo (FLUX2-klein pixel-space) (extra data) | 2026-08-17 |
| Self-Evaluation Unlocks Any-Step Text-to-Image Generation | | 0.8151 | | | | | | | Self-E | 2025-12-26 |
| Internalizing Agent Experience into Diffusion Model Weights via On-Policy Context Distillation | ✓ Link | 0.8128 | | | | | | | D-OPCD generator | 2026-10-05 |
| Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents | | 0.81 | | | | | | | Bifrost-1 (Qwen2.5-VL 7B, 62M) | 2025-08-08 |
| AIA: Rethinking Architecture Decoupling Strategy In Unified Multimodal Model | ✓ Link | 0.81 | | | | | | | Janus-Pro + AIA | 2025-11-27 |
| Imagine Before You Draw: Visual Prompt Engineering for Image Generation | | 0.81 | | | | | | | VPE (External) | 2026-06-03 |
| MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities | ✓ Link | 0.81 | | | | | | | MODUS (original prompts, no self-verification) | 2026-07-28 |
| Semantic Generative Tuning for Unified Multimodal Models | ✓ Link | 0.8095 | | | | | | | SGT-BAGEL (original prompts) | 2026-05-18 |
| PixelDense: Dense Prediction as Representation Alignment for Pixel Diffusion | ✓ Link | 0.8093 | | | | | | | PixelDense | 2026-09-30 |
| Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling | ✓ Link | 0.80 | | | | | | | Janus-Pro-7B | 2025-01-29 |
| SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer | ✓ Link | 0.80 | | | | | | | SANA-1.5 4.8B (+ Inference Scaling) | 2025-01-30 |
| Transfer between Modalities with MetaQueries | | 0.80 | | | | | | | MetaQuery-XL (Rewrite) | 2025-04-08 |
| UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation | ✓ Link | 0.80 | 0.99 | 0.93 | 0.70 | 0.89 | 0.79 | 0.49 | UniWorld-V1 | 2025-06-03 |
| Autoregressive Image Generation Needs Only a Few Lines of Cached Tokens | ✓ Link | 0.8 | | | | | | | LineAR (Janus-Pro-7B, 1/4 cache) | 2025-12-04 |
| Auto-Rubric as Reward: From Implicit Preferences to Explicit Multimodal Generative Criteria | ✓ Link | 0.8 | | | | | | | FLUX.1-dev + ARR-RPO (Gemini 3.1 Pro judge) | 2026-05-08 |
| UniPath: Adaptive Coordination of Understanding and Generation for Unified Multimodal Reasoning | | 0.8 | | | | | | | UniPath (BAGEL backbone) | 2026-05-12 |
| Aligning One-Step Generative Models with Reward-Weighted Transport Distillation | ✓ Link | 0.8 | | | | | | | RWTD (SANA Sprint 1.6B) | 2026-09-25 |
| Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning | ✓ Link | 0.798 | | | | | | | Iris-3B | 2026-10-07 |
| On the Diffusibility of High-Dimensional Latents | | 0.7969 | | | | | | | On the Diffusibility of High-Dimensional Latents (Ours, 512) | 2026-09-23 |
| Lumos-1: On Autoregressive Video Generation with Discrete Diffusion from a Unified Model Perspective | ✓ Link | 0.791 | | | | | | | Lumos-1 (512x512) (3.6B) | 2025-07-11 |
| PixelGen: Pixel Diffusion Beats Latent Diffusion with Perceptual Loss | ✓ Link | 0.79 | | | | | | | PixelGen-XXL/16 | 2026-02-02 |
| Channel-wise Vector Quantization | ✓ Link | 0.79 | | | | | | | CAR-8B | 2026-05-25 |
| Imagine Before You Draw: Visual Prompt Engineering for Image Generation | | 0.79 | | | | | | | VPE (Internal) | 2026-06-03 |
| Spectral Progressive Diffusion for Efficient Image and Video Generation | ✓ Link | 0.782 | | | | | | | Spectral Progressive Diffusion (PixelGen, TF) | 2026-05-18 |
| PixelDiT: Pixel Diffusion Transformers for Image Generation | ✓ Link | 0.78 | | | | | | | PixelDiT-T2I | 2025-11-25 |
| Sparse-LaViDa: Sparse Multimodal Discrete Diffusion Language Models | | 0.78 | | | | | | | Sparse-LaViDa | 2025-12-16 |
| StageVAR: Stage-Aware Acceleration for Visual Autoregressive Models | ✓ Link | 0.78 | | | | | | | Infinity-8B + StageVAR | 2025-12-18 |
| Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation | ✓ Link | 0.78 | | | | | | | Cheers | 2026-03-13 |
| SNCE: Geometry-Aware Supervision for Scalable Discrete Image Generation | | 0.78 | | | | | | | LaViDa-O (10B, Adaptation, SNCE) | 2026-03-16 |
| DRM: Diffusion-based Reward Model With Step-wise Guidance | ✓ Link | 0.78 | | | | | | | SD3.5-M + DRM Step-GRPO | 2026-05-25 |
| C3-UniMM: Causal Cycle-Consistent Unified Multimodal Modeling via Super Alignment and Shared Decoding Space | | 0.78 | | | | | | | C3-UniMM (1.5B) | 2026-07-04 |
| Both Semantics and Reconstruction Matter: Making Representation Encoders Ready for Text-to-Image Generation and Editing | | 0.7714 | | | | | | | PS-VAE (96c, SigLIP2-so400m/14) | 2025-12-19 |
| Can We Generate Images with CoT? Let's Verify and Reinforce Image Generation Step by Step | ✓ Link | 0.77 | | | | | | | Show-o [xie2024show] PARM It. DPO PARM | 2025-01-23 |
| Draw-In-Mind: Rebalancing Designer-Painter Roles in Unified Multimodal Models Benefits Image Editing | ✓ Link | 0.77 | | | | | | | DIM-4.6B-T2I | 2025-09-02 |
| Visual Generation Tuning | ✓ Link | 0.77 | | | | | | | VGT (Qwen2.5-VL) | 2025-11-28 |
| Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation | ✓ Link | 0.77 | | | | | | | FIRM-SD3.5 | 2026-03-12 |
| PixelUMM: Encoder-Free Unified Image and Video Understanding and Generation | ✓ Link | 0.77 | | | | | | | PixelUMM 8B MoT | 2026-09-29 |
| Show-o2: Improved Native Unified Multimodal Models | ✓ Link | 0.76 | | | | | | | Show-o2 (7B) | 2025-06-18 |
| VQRAE: Representation Quantization Autoencoders for Multimodal Understanding, Generation and Reconstruction | | 0.76 | | | | | | | VQRAE | 2025-11-28 |
| SnapGen++: Unleashing Diffusion Transformers for Efficient High-Fidelity Image Generation on Edge Devices | | 0.76 | | | | | | | SnapGen++ (full, 1.6B) | 2026-01-13 |
| L2P: Unlocking Latent Potential for Pixel Generation | ✓ Link | 0.76 | | | | | | | L2P | 2026-05-12 |
| Text Template Tokens Are Implicit Semantic Registers in Diffusion Transformers | ✓ Link | 0.758 | | | | | | | Qwen-Image-2512 with training-free head pruning (rank by image-to-semantic attention m_S, K=216) | 2026-07-21 |
| High-Fidelity Two-Step Image Generation via Teacher-Aligned End-to-End Distillation | | 0.757 | | | | | | | Z-Image Turbo++ | 2026-06-10 |
| Can We Generate Images with CoT? Let's Verify and Reinforce Image Generation Step by Step | ✓ Link | 0.75 | | | | | | | Show-o [xie2024show] Ft. ORM It. DPO Ft. ORM | 2025-01-23 |
| VA-π: Variational Policy Alignment for Pixel-Aware Autoregressive Generation | ✓ Link | 0.744 | | | | | | | VA-π (Janus-Pro 1B) | 2025-12-22 |
| LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories | | 0.742 | | | | | | | FLUX.1-dev + LeapAlign | 2026-04-16 |
| Mobile-O: Unified Multimodal Understanding and Generation on Mobile Device | ✓ Link | 0.74 | | | | | | | Mobile-O-0.5B | 2026-02-23 |
| MONET: A Massive, Open, Non-redundant and Enriched Text-to-image dataset | | 0.74 | | | | | | | MONET (4B) | 2026-05-20 |
| Xiaomi-Robotics-U0: Unified Embodied Synthesis with World Foundation Model | ✓ Link | 0.74 | | | | | | | Xiaomi-Robotics-U0-AR | 2026-07-13 |
| MEND: RL For Flow Models via Proximal Velocity Matching | ✓ Link | 0.74 | | | | | | | MEND (SD3.5-M, 100 updates) | 2026-10-05 |
| Poly-OPD: Heterogeneous Multi-Teacher On-Policy Distillation for Capability-Selectable Flow Models | | 0.733 | | | | | | | Poly-OPD (SD3.5-Medium, composition mode) | 2026-08-05 |
| Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling | ✓ Link | 0.73 | | | | | | | Janus-Pro-1B | 2025-01-29 |
| Lumina-Image 2.0: A Unified and Efficient Image Generative Framework | ✓ Link | 0.73 | | | | | | | Lumina-Image 2.0 | 2025-03-27 |
| Show-o2: Improved Native Unified Multimodal Models | ✓ Link | 0.73 | | | | | | | Show-o2 (1.5B) | 2025-06-18 |
| PixNerd: Pixel Neural Field Diffusion | ✓ Link | 0.73 | | | | | | | PixNerd-XXL/16 | 2025-07-31 |
| Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairs | | 0.73 | | | | | | | PNAPO (SD3-M) | 2026-05-10 |
| Lumos-1: On Autoregressive Video Generation with Discrete Diffusion from a Unified Model Perspective | ✓ Link | 0.725 | | | | | | | Lumos-1 (512x512) (1.5B) | 2025-07-11 |
| Anchoring and Steering Diffusion: Enhancing the Faithfulness of Text-to-Image Generation at Inference Time | | 0.7247 | | | | | | | AnchorSteer (HunyuanDiT-v1.2) | 2026-07-29 |
| SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer | ✓ Link | 0.72 | 0.99 | 0.85 | | | | | SANA-1.5 4.8B | 2025-01-30 |
| DreamLite: A Lightweight On-Device Unified Model for Image Generation and Editing | ✓ Link | 0.72 | | | | | | | DreamLite | 2026-03-30 |
| HYDRA-X: Native Unified Multimodal Models with Holistic Visual Tokenizers | | 0.72 | | | | | | | Hydra-XTok (img distill; Decompressor distilled with video teacher) | 2026-06-11 |
| TexTailor: Inference-Time Textual Guidance Tailoring for Multimodal Diffusion Transformers | | 0.7163 | | | | | | | TexTailor (SD3.5) | 2026-01-05 |
| Argus-Unified: Towards A Compact and Economical Unified Model for Image Understanding and Generation | | 0.71 | | | | | | | Argus-Unified (1.5B) | 2026-07-28 |
| SynVAR: Synergizing Spatial and Semantic Alignment in Visual Autoregressive Model | ✓ Link | 0.7079 | | | | | | | Infinity + SynVAR | 2026-08-08 |
| Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning | ✓ Link | 0.7053 | | | | | | | FLUX.1-dev + Pref-GRPO | 2025-08-28 |
| Fluid: Scaling Autoregressive Text-to-image Generative Models with Continuous Tokens | ✓ Link | 0.69 | 0.96 | 0.83 | 0.51 | 0.80 | 0.63 | 0.39 | Fluid (10.5B) | 2024-10-17 |
| The Image as Its Own Reward: Reinforcement Learning with Adversarial Reward for Image Generation | ✓ Link | 0.69 | | | | | | | Adv-GRPO (DINO) | 2025-11-25 |
| Linearizing Vision Transformer with Test-Time Training | ✓ Link | 0.69 | | | | | | | SD3.5-T 5 | 2026-05-28 |
| RADAR: VQ-VAE Decoder of VAR is a Good Student for Restoring Against Degradation by Acceleration | | 0.69 | | | | | | | Infinity + RADAR | 2026-06-03 |
| LazyVAR: Accelerating Visual Autoregressive Models via Scale-wise Token Pruning and Parallel Group Decoding | | 0.6864 | | | | | | | Infinity + LazyVAR | 2026-06-03 |
| Show-o: One Single Transformer to Unify Multimodal Understanding and Generation | ✓ Link | 0.68 | | | | | | | Und. and Gen. Show-o (Ours) | 2024-08-22 |
| Uniform Discrete Diffusion with Metric Path for Video Generation | ✓ Link | 0.68 | | | | | | | URSA-1.7B (1024x1024, original prompts) | 2025-10-28 |
| UltraGen: Efficient Ultra-High-Resolution Image Generation with Hierarchical Local Attention | | 0.67 | | | | | | | UltraImageGen (4K) | 2026-09-08 |
| Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models | ✓ Link | 0.67 | | | | | | | TORS (Flux.1-Dev) | 2026-09-08 |
| Lumos-1: On Autoregressive Video Generation with Discrete Diffusion from a Unified Model Perspective | ✓ Link | 0.664 | | | | | | | Lumos-1 (352x352) (3.6B) | 2025-07-11 |
| DiverseVAR: Balancing Diversity and Quality of Next-Scale Visual Autoregressive Models | | 0.662 | | | | | | | DIVERSEVAR | 2025-11-26 |
| Emu3: Next-Token Prediction is All You Need | ✓ Link | 0.66 | | | | | | | Emu3 | 2024-09-27 |
| SnapGen: Taming High-Resolution Text-to-Image Models for Mobile Devices with Efficient Architectures and Training | | 0.66 | | | | | | | SnapGen | 2024-12-12 |
| Drift-AR: Single-Step Visual Autoregressive Generation via Anti-Symmetric Drifting | ✓ Link | 0.66 | | | | | | | Drift-AR (NextStep-1) | 2026-03-30 |
| Argus-Unified: Towards A Compact and Economical Unified Model for Image Understanding and Generation | | 0.66 | | | | | | | Argus-Unified (0.5B) | 2026-07-28 |
| Importance-Aware Low-Rank Distillation of Diffusion Transformers | | 0.645 | | | | | | | SVDtrunc-m | 2026-09-08 |
| LogCD: Local-to-global Consistency Distillation for Few-step Image Generation | | 0.64 | | | | | | | LogCM (FLUX.1-dev, 4-step) | 2026-06-03 |
| JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation | ✓ Link | 0.63 | | | | | | | JanusFlow | 2024-11-12 |
| NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale | ✓ Link | 0.63 | | | | | | | NextStep-1 | 2025-08-14 |
| Denoising, Fast and Slow: Difficulty-Aware Adaptive Sampling for Image Generation | ✓ Link | 0.63 | | | | | | | PFT-1.2B + Look-Ahead | 2026-04-21 |
| RoAR-VAE: Improving Latent Diffusability through Robust Prediction | ✓ Link | 0.6222 | | | | | | | RoAR-VAE | 2026-10-19 |
| Compositional Text-to-Image Generation Via Region-aware Bimodal Direct Preference Optimization | ✓ Link | 0.62 | | | | | | | SDXL-BiDPO | 2026-05-27 |
| Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation | ✓ Link | 0.61 | | | | | | | Janus | 2024-10-17 |
| Bifrost-1: Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents | | 0.61 | | | | | | | Bifrost-1 (Qwen2.5-VL 3B, 9M) | 2025-08-08 |
| Lumos-1: On Autoregressive Video Generation with Discrete Diffusion from a Unified Model Perspective | ✓ Link | 0.601 | | | | | | | Lumos-1 (352x352) (1.5B) | 2025-07-11 |
| PixelFlow: Pixel-Space Generative Models with Flow | ✓ Link | 0.6 | | | | | | | PixelFlow (original prompts) | 2025-04-10 |
| VINO: A Unified Visual Generator with Interleaved OmniModal Context | ✓ Link | 0.59 | | | | | | | VINO (original prompts) | 2026-01-05 |
| Learning from Noisy Preferences: A Semi-Supervised Learning Approach to Direct Preference Optimization | ✓ Link | 0.5841 | | | | | | | Semi-DPO (SDXL) | 2026-04-27 |
| Optimizing Few-Step Generation with Adaptive Matching Distillation | | 0.57 | | | | | | | SDXL-AMD | 2026-02-07 |
| SCALE: Semantic-Calibrated Guidance Enhancement for Prompt-Faithful Diffusion | ✓ Link | 0.5641 | | | | | | | SCALE | 2026-09-08 |
| Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generation | ✓ Link | 0.558 | | | | | | | OpenUni-3.6B + UniMRG | 2026-01-29 |
| Scaling Text-to-Image Diffusion Transformers with Representation Autoencoders | ✓ Link | 0.555 | | | | | | | RAE (Qwen-2.5 7B + DiT 5.5B, without TTS) | 2026-01-22 |
| Text-to-Image Generative Modeling Without Language Pretraining | | 0.5348 | | | | | | | Halcyon-1.60B | 2026-10-19 |
| PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation | ✓ Link | 0.53 | | | | | | | PixArt-Σ | 2024-03-07 |
| Few-Step Distillation for Text-to-Image Generation: A Practical Guide | ✓ Link | 0.5281 | | | | | | | sCM (2 NFE) (extra data) | 2025-12-15 |
| DiffMoE: Dynamic Token Selection for Scalable Diffusion Transformers | | 0.51 | | | | | | | DiffMoE-E16-T2I-Flow (w SFT) | 2025-03-18 |
| Beyond Language Modeling: An Exploration of Multimodal Pretraining | | 0.367 | | | | | | | MoE (Per-Modality Shared Expert) | 2026-03-03 |
| Improved Distributional Diffusion Models | ✓ Link | 0.353 | | | | | | | iDDM T2I (1.6B) (extra data) | 2026-09-29 |
| Alchemist: Unlocking Efficiency in Text-to-Image Model Training via Meta-Gradient Data Selection | ✓ Link | 0.2645 | | | | | | | STAR-0.3B (trained with Alchemist selection, 15M) | 2025-12-18 |
| PIXART-δ: Fast and Controllable Image Generation with Latent Consistency Models | ✓ Link | 0 | | | | | | | PIXART-δ | 2024-01-10 |