Nucleus-Image: Sparse MoE for Image Generation

• URL: http://arxiv.org/abs/2604.12163v1
• Date: Tue, 14 Apr 2026 00:43:23 GMT
• Title: Nucleus-Image: Sparse MoE for Image Generation
• Abstract summary: We present a text-to-image generation model that exceeds leading models on GenEval, DPG-Bench, and OneIG-Bench.<n>Nucleus-Image employs a sparse mixture-of-experts (MoE) diffusion transformer architecture.<n>We construct a large-scale training corpus of 1.5B high-quality training pairs spanning 700M images.
• Score: 5.769753912757775
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We present Nucleus-Image, a text-to-image generation model that establishes a new Pareto frontier in quality-versus-efficiency by matching or exceeding leading models on GenEval, DPG-Bench, and OneIG-Bench while activating only approximately 2B parameters per forward pass. Nucleus-Image employs a sparse mixture-of-experts (MoE) diffusion transformer architecture with Expert-Choice Routing that scales total model capacity to 17B parameters across 64 routed experts per layer. We adopt a streamlined architecture optimized for inference efficiency by excluding text tokens from the transformer backbone entirely and using joint attention that enables text KV sharing across timesteps. To improve routing stability when using timestep modulation, we introduce a decoupled routing design that separates timestep-aware expert assignment from timestep-conditioned expert computation. We construct a large-scale training corpus of 1.5B high-quality training pairs spanning 700M unique images through multi-stage filtering, deduplication, aesthetic tiering, and caption curation. Training follows a progressive resolution curriculum (256 to 512 to 1024) with multi-aspect-ratio bucketing at every stage, coupled with progressive sparsification of the expert capacity factor. We adopt the Muon optimizer and share our parameter grouping recipe tailored for diffusion models with timestep modulation. Nucleus-Image demonstrates that sparse MoE scaling is a highly effective path to high-quality image generation, reaching the performance of models with significantly larger active parameter budgets at a fraction of the inference cost. These results are achieved without post-training optimization of any kind: no reinforcement learning, no direct preference optimization, and no human preference tuning. We release the training recipe, making Nucleus-Image the first fully open-source MoE diffusion model at this quality.

Related papers

• Exploring the Performance Frontier of Compact Unified Image Generation Models [24.485980902368706]
  Swift-Image is a compact unified model for text-to-image generation, single-image editing, and multi-image editing.<n>Swift-Image achieves leading aggregate performance among evaluated open-source models with only 6B parameters and 243K GPU training hours.
  arXiv  Detail & Related papers  (2026-08-20T17:59:52Z)
• SeFi-Image: A Text-to-Image Foundation Model with Semantic-First Diffusion [0.0]
  SeFi-Image is a text-to-image foundation model built upon semantic-first diffusion.<n>We instantiate SeFi-Image at three model scales, 1B, 2B, and 5B parameters.<n>SeFi-Image achieves strong performance on a wide range of benchmarks, including GenEval, DPG, LongTextBench, OneIG, and DMDG-2K.
  arXiv  Detail & Related papers  (2026-06-21T16:10:22Z)
• Efficient and Training-Free Single-Image Diffusion Models [17.578119446864132]
  We consider the problem of generating images whose internal structure matches that of a single reference image.<n>We integrate a patch-based denoiser into an efficient, training-free image diffusion model.<n>Our approach achieves state-of-the-art generation quality and diversity compared to trained single-image diffusion models.
  arXiv  Detail & Related papers  (2026-06-03T00:05:36Z)
• Diffusion Image Generation with Explicit Modeling of Data Manifold Geometry [64.61931485161833]
  Image generative models aim to sample data points from the underlying data manifold.<n>We propose the Data Manifold-aware Image diffusioN moDel (MIND)<n>MIND integrates discrete patch tokenization into the score function of a continuous diffusion model.
  arXiv  Detail & Related papers  (2026-05-25T08:43:14Z)
• DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editing [67.77471070868852]
  DeepGen 1.0 is a lightweight 5B unified model for image generation and editing.<n>It is trained on only 50M samples, surpassing the 80B HunyuanImage by 28% on WISE and the 27B Qwen-Image-Edit by 37% on UniREditBench.<n>By open-sourcing our training code, weights, and datasets, we provide an efficient, high-performance alternative to democratize unified multimodal research.
  arXiv  Detail & Related papers  (2026-02-12T17:44:24Z)
• Home-made Diffusion Model from Scratch to Hatch [0.9383683724544296]
  Home-made Diffusion Model (HDM) is an efficient yet powerful text-to-image diffusion model optimized for training on consumer-grade hardware.<n>HDM achieves competitive 1024x1024 generation quality while maintaining a remarkably low training cost of $535-620.
  arXiv  Detail & Related papers  (2025-09-07T14:21:57Z)
• Policy Optimized Text-to-Image Pipeline Design [73.9633527029941]
  We introduce a novel reinforcement learning-based framework for text-to-image generation.<n>Our approach first trains an ensemble of reward models capable of predicting image quality scores directly from prompt-workflow combinations.<n>We then implement a two-phase training strategy: initial vocabulary training followed by GRPO-based optimization.
  arXiv  Detail & Related papers  (2025-05-27T17:50:47Z)
• Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models [33.519892081718716]
  We propose aligning the latent space with pre-trained vision foundation models when training the visual tokenizers.<n>Our proposed VA-VAE significantly expands the reconstruction-generation frontier of latent diffusion models.<n>We build an enhanced DiT baseline with improved training strategies and architecture designs, termed LightningDiT.
  arXiv  Detail & Related papers  (2025-01-02T18:59:40Z)
• Make a Cheap Scaling: A Self-Cascade Diffusion Model for Higher-Resolution Adaptation [112.08287900261898]
  This paper proposes a novel self-cascade diffusion model for rapid adaptation to higher-resolution image and video generation. Our approach achieves a 5X training speed-up and requires only an additional 0.002M tuning parameters. Experiments demonstrate that our approach can quickly adapt to higher resolution image and video synthesis by fine-tuning for just 10k steps, with virtually no additional inference time.
  arXiv  Detail & Related papers  (2024-02-16T07:48:35Z)
• A-SDM: Accelerating Stable Diffusion through Redundancy Removal and Performance Optimization [54.113083217869516]
  In this work, we first explore the computational redundancy part of the network. We then prune the redundancy blocks of the model and maintain the network performance. Thirdly, we propose a global-regional interactive (GRI) attention to speed up the computationally intensive attention part.
  arXiv  Detail & Related papers  (2023-12-24T15:37:47Z)
• Dual-former: Hybrid Self-attention Transformer for Efficient Image Restoration [6.611849560359801]
  We present Dual-former, which combines the powerful global modeling ability of self-attention modules and the local modeling ability of convolutions in an overall architecture. Experiments demonstrate that Dual-former achieves a 1.91dB gain over the state-of-the-art MAXIM method on the Indoor dataset for single image dehazing. For single image deraining, it exceeds the SOTA method by 0.1dB PSNR on the average results of five datasets with only 21.5% GFLOPs.
  arXiv  Detail & Related papers  (2022-10-03T16:39:21Z)
• Image-specific Convolutional Kernel Modulation for Single Image Super-resolution [85.09413241502209]
  In this issue, we propose a novel image-specific convolutional modulation kernel (IKM) We exploit the global contextual information of image or feature to generate an attention weight for adaptively modulating the convolutional kernels. Experiments on single image super-resolution show that the proposed methods achieve superior performances over state-of-the-art methods.
  arXiv  Detail & Related papers  (2021-11-16T11:05:10Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.