Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
hub
High- resolutionimagesynthesiswithlatentdiffusionmodels
26 Pith papers cite this work. Polarity classification is still indexing.
26
Pith papers citing it
abstract
This paper presents SANA-1.5, a linear Diffusion Transformer for efficient scaling in text-to-image generation. Building upon SANA-1.0, we introduce three key innovations: (1) Efficient Training Scaling: A depth-growth paradigm that enables scaling from 1.6B to 4.8B parameters with significantly reduced computational resources, combined with a memory-efficient 8-bit optimizer. (2) Model Depth Pruning: A block importance analysis technique for efficient model compression to arbitrary sizes with minimal quality loss. (3) Inference-time Scaling: A repeated sampling strategy that trades computation for model capacity, enabling smaller models to match larger model quality at inference time. Through these strategies, SANA-1.5 achieves a text-image alignment score of 0.81 on GenEval, which can be further improved to 0.96 through inference scaling with VILA-Judge, establishing a new SoTA on GenEval benchmark. These innovations enable efficient model scaling across different compute budgets while maintaining high quality, making high-quality image generation more accessible. Our code and pre-trained models are released.
hub tools
citation-role summary
background 2
baseline 2
citation-polarity summary
representative citing papers
OctoT2I uses a no-supervision PSEL loop to discover model capability frontiers and route T2I tasks, reaching 0.96 GenEval score with 90.3% speedup over Flow-GRPO.
Art Arena evaluates how artistic styles from training data leak into AI-generated images without explicit prompts, revealing asymmetric blending due to differences in representational strength and interaction dynamics across models like Stable Diffusion.
ICEdit achieves state-of-the-art instructional image editing in Diffusion Transformers via in-context generation, requiring only 0.1% of prior training data and 1% trainable parameters.
UDM-GRPO integrates reinforcement learning with Uniform Discrete Diffusion Models by treating the final clean sample as the action and reconstructing trajectories via the forward process, achieving state-of-the-art text-to-image performance.
Introduces SciIR-82k dataset and SciIR-Bench for scientific image reasoning generation organized by Peirce's semiotic triad, with fine-tuning raising model score from 35% to 43%.
Feature self-guidance disperses internal features of flow models during batch generation and applies manifold regularization to increase output diversity while preserving condition alignment.
IV-CoT introduces an implicit chain-of-thought framework that decomposes visual queries into a structural-to-semantic cascade with training-only sketch supervision to improve structure-aware text-to-image generation.
FlowCodec decouples latent compression from one-step flow-based transport to plug pretrained text-to-image models into ultra-low-bitrate codecs with under 0.54% trainable parameters.
Moebius introduces a compressed diffusion inpainting model using Local-λ Mix Interaction blocks and latent-space multi-granularity distillation to reach 10B-level quality with 0.22B parameters.
Switching a diffusion editor from image-conditioned to caption-only mode for a mid-trajectory interval and back improves edit fidelity and naturalness on FLUX2-Klein and BAGEL, while predictably reducing source-image preservation.
SANA-Streaming delivers 1280x704 streaming video editing at 24 FPS end-to-end on an RTX 5090 using hybrid DiT blocks, cycle-reverse training, and mixed-precision quantization.
Dense scene composition and instance fidelity in synthetic diffusion images drive better segmentation performance; SENSE framework exploits this to improve models on Cityscapes, COCO, and ADE20K.
Policy entropy remains constant in flow-matching models during RLHF due to fixed noise schedules while perceptual diversity collapses from mode-seeking policy gradients, so perceptual entropy constraints are introduced to preserve diversity and improve quality.
EPIC introduces predicate-guided inference-time search that lifts compositional T2I prompt accuracy from 34% to 71% on GenEval2 with 31-81% lower execution costs.
A 17B-parameter sparse MoE diffusion transformer activates 2B parameters per pass and reaches competitive quality on image generation benchmarks without post-training.
TC-AE improves reconstruction and generative performance in deep compression by decomposing token-to-latent compression into two stages and using joint self-supervised training.
RAPO++ is a three-stage prompt optimization framework combining retrieval-augmented refinement, closed-loop test-time scaling, and LLM fine-tuning to enhance text-to-video generation quality.
The work introduces rCM, a score-regularized continuous-time consistency model that matches DMD2 quality on large models up to 14B parameters while improving diversity and enabling 1-4 step sampling.
YOLOv12 is a new attention-based real-time object detector that reports higher accuracy than YOLOv10, YOLOv11, and RT-DETR variants at comparable or better speed and efficiency.
A flow-matching speech separator with biometric best-of-N candidate selection and chunk-wise channel alignment achieves competitive separation metrics and the best downstream ASR/SV error rates among evaluated systems on Libri2Mix.
TMP is a pruning framework that reduces HunyuanImage-3.0 from 80B to 20B parameters (75% reduction) and Z-Image turbo from 6B to 4B with limited quality degradation.
A survey of test-time scaling for multimodal foundation models that introduces a three-way taxonomy of sampling, feedback, and search approaches along with applications and benchmarks.
WinTok is a hybrid visual tokenizer that supplements pixel tokens with learnable semantic tokens distilled asymmetrically from foundation models to improve reconstruction, understanding, and generation.
citing papers explorer
-
Flow-GRPO: Training Flow Matching Models via Online RL
Flow-GRPO is the first online RL method for flow matching models, raising GenEval accuracy from 63% to 95% and text-rendering accuracy from 59% to 92% with little reward hacking.
-
OctoT2I: A Self-Evolving Agentic Text-to-Image Router
OctoT2I uses a no-supervision PSEL loop to discover model capability frontiers and route T2I tasks, reaching 0.96 GenEval score with 90.3% speedup over Flow-GRPO.
-
The Silent Brush: Evaluating Artistic Style Leakage in AI Art Generation
Art Arena evaluates how artistic styles from training data leak into AI-generated images without explicit prompts, revealing asymmetric blending due to differences in representational strength and interaction dynamics across models like Stable Diffusion.
-
In-Context Edit: Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer
ICEdit achieves state-of-the-art instructional image editing in Diffusion Transformers via in-context generation, requiring only 0.1% of prior training data and 1% trainable parameters.
-
UDM-GRPO: Stable and Efficient Group Relative Policy Optimization for Uniform Discrete Diffusion Models
UDM-GRPO integrates reinforcement learning with Uniform Discrete Diffusion Models by treating the final clean sample as the action and reconstructing trajectories via the forward process, achieving state-of-the-art text-to-image performance.
-
SciIR: A Large-scale Training Dataset and Benchmark for Scientific Image Reasoning Generation
Introduces SciIR-82k dataset and SciIR-Bench for scientific image reasoning generation organized by Peirce's semiotic triad, with fine-tuning raising model score from 35% to 43%.
-
Don't Settle at the Mode! Mitigating Diversity Collapse in Pretrained Flow Models via Feature Self-Guidance
Feature self-guidance disperses internal features of flow models during batch generation and applies manifold regularization to increase output diversity while preserving condition alignment.
-
IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation
IV-CoT introduces an implicit chain-of-thought framework that decomposes visual queries into a structural-to-semantic cascade with training-only sketch supervision to improve structure-aware text-to-image generation.
-
FlowCodec: One-Step Flow Prior for Generative Image Compression
FlowCodec decouples latent compression from one-step flow-based transport to plug pretrained text-to-image models into ultra-low-bitrate codecs with under 0.54% trainable parameters.
-
Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance
Moebius introduces a compressed diffusion inpainting model using Local-λ Mix Interaction blocks and latent-space multi-granularity distillation to reach 10B-level quality with 0.22B parameters.
-
DuET: Dual Expert Trajectories for Diffusion Image Editing
Switching a diffusion editor from image-conditioned to caption-only mode for a mid-trajectory interval and back improves edit fidelity and naturalness on FLUX2-Klein and BAGEL, while predictably reducing source-image preservation.
-
SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer
SANA-Streaming delivers 1280x704 streaming video editing at 24 FPS end-to-end on an RTX 5090 using hybrid DiT blocks, cycle-reverse training, and mixed-precision quantization.
-
What Makes Synthetic Data Effective in Image Segmentation
Dense scene composition and instance fidelity in synthetic diffusion images drive better segmentation performance; SENSE framework exploits this to improve models on Cityscapes, COCO, and ADE20K.
-
When Policy Entropy Constraint Fails: Preserving Diversity in Flow-based RLHF via Perceptual Entropy
Policy entropy remains constant in flow-matching models during RLHF due to fixed noise schedules while perceptual diversity collapses from mode-seeking policy gradients, so perceptual entropy constraints are introduced to preserve diversity and improve quality.
-
EPIC: Efficient Predicate-Guided Inference-Time Control for Compositional Text-to-Image Generation
EPIC introduces predicate-guided inference-time search that lifts compositional T2I prompt accuracy from 34% to 71% on GenEval2 with 31-81% lower execution costs.
-
Nucleus-Image: Sparse MoE for Image Generation
A 17B-parameter sparse MoE diffusion transformer activates 2B parameters per pass and reaches competitive quality on image generation benchmarks without post-training.
-
TC-AE: Unlocking Token Capacity for Deep Compression Autoencoders
TC-AE improves reconstruction and generative performance in deep compression by decomposing token-to-latent compression into two stages and using joint self-supervised training.
-
RAPO++: Cross-Stage Prompt Optimization for Text-to-Video Generation via Data Alignment and Test-Time Scaling
RAPO++ is a three-stage prompt optimization framework combining retrieval-augmented refinement, closed-loop test-time scaling, and LLM fine-tuning to enhance text-to-video generation quality.
-
Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency
The work introduces rCM, a score-regularized continuous-time consistency model that matches DMD2 quality on large models up to 14B parameters while improving diversity and enabling 1-4 step sampling.
-
YOLOv12: Attention-Centric Real-Time Object Detectors
YOLOv12 is a new attention-based real-time object detector that reports higher accuracy than YOLOv10, YOLOv11, and RT-DETR variants at comparable or better speed and efficiency.
-
Flow Matching-Based Speech Source Separation with Best-of-N Biometric Sampling
A flow-matching speech separator with biometric best-of-N candidate selection and chunk-wise channel alignment achieves competitive separation metrics and the best downstream ASR/SV error rates among evaluated systems on Libri2Mix.
-
TMP: Tree-structured Mixed-policy Pruning for Large-scale Image Generation and Editing
TMP is a pruning framework that reduces HunyuanImage-3.0 from 80B to 20B parameters (75% reduction) and Z-Image turbo from 6B to 4B with limited quality degradation.
-
Test-Time Scaling in Multimodal Foundation Models: A Comprehensive Survey of Generation and Reasoning
A survey of test-time scaling for multimodal foundation models that introduces a three-way taxonomy of sampling, feedback, and search approaches along with applications and benchmarks.
-
WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens
WinTok is a hybrid visual tokenizer that supplements pixel tokens with learnable semantic tokens distilled asymmetrically from foundation models to improve reconstruction, understanding, and generation.
-
Qwen-Image Technical Report
Qwen-Image is a foundation model that reaches state-of-the-art results in image generation and editing by combining a large-scale text-focused data pipeline with curriculum learning and dual semantic-reconstructive encoding for editing consistency.
-
Breaking the Lock-in: Diversifying Text-to-Image Generation via Representation Modulation
Early DC component convergence in text-to-image Transformer features causes output homogeneity; selective early attenuation via DAVE improves diversity without retraining or extra cost.