MOSS-Image pairs a 5B diffusion model with prompt planning to improve text-heavy, layout-coherent image generation.
Key results
The model is pretrained from scratch for text-rich, layout-controlled image generation.
Bilingual long-form text-rendering benchmark tasks.
Average score among the open-weight models evaluated.
Elo rating; second place behind GPT Image 2.
Elo rating, highest among the evaluated generators given shared enhanced prompts.
What the paper found
MOSS-Image targets a persistent weakness in open-weight image generation: producing dense, readable text inside complex layouts. Its core is a 5B single-stream diffusion transformer, pretrained from scratch to follow natural-language prompts and structured layout specifications through one conditioning pathway. Training combines broad image-text data, multilevel captions, explicit text and bounding-box annotations, and progressive resolution; a GPT-6 prompt-enhancement agent expands brief requests into validated content and layout plans. The report introduces LongLongTextBench (LLTB), a 240-task bilingual benchmark for long-form text, and RealLayoutArena for judging complete systems on posters, infographics, and interfaces. MOSS-Image reaches 96.3% average text coverage on LLTB, leading the tested open-weight models. On RealLayoutArena it scores 1698.70 Elo, placing second behind GPT Image 2 and ahead of Seedream 5.0 Pro; with shared enhanced prompts, its generator leads the evaluated backbones at 1807.34 Elo. The results suggest that strong rendering and explicit content-and-layout planning work best together, while the system still depends on an external language model and does not yet support image-conditioned editing.
Original abstract
Recent closed-source image generation systems can produce diverse visual content from simple compositions to information-dense graphics, yet open-weight text-to-image models still struggle with dense text and intricate layouts. To narrow this gap, we introduce MOSS-Image, an open generation system that combines high-fidelity visual synthesis with intent-aware composition. First, we pretrain a 5B single-stream diffusion transformer (DiT) from scratch with a carefully designed recipe, enabling bilingual long-form text rendering and precise layout control. Second, we develop an intention-aware agentic prompt enhancement (PE) module that translates diverse user requests into detailed content and layout specifications. Together, they enable MOSS-Image to generate text-rich and layout-coherent images. We further introduce LongLongTextBench (LLTB) for evaluating long-form English and Chinese text rendering, and RealLayoutArena (RLA) for end-to-end layout-rich generation. MOSS-Image achieves state-of-the-art performance among open-weight models on LLTB, with an average text coverage of 96.3%. On RLA, MOSS-Image ranks second in the end-to-end system comparison, while its backbone ranks first among generators evaluated with shared enhanced prompts. We will release the model, training recipe, and evaluation framework to support further research and promote the text-to-image community.
Read the original paperMore in Diffusion Models
Browse all 73 papers →On the SPOT: Enhancing Speculative Decoding for Diffusion Language Models via Commit-Order Prediction
Eliezer Shlomi, Eliron Rahimi, Avi Mendelson, Chaim Baskin
SPOT speeds up diffusion language models by predicting which positions to commit next, then checking multiple proposed tokens in one pass.
PhysLDM: Latent Diffusion for High-Fidelity Deformable Simulation
Yu Zhang, Xudong Xu, Xingang Pan
PhysLDM uses latent diffusion to predict plausible long-term motion of complex deformable objects more effectively than deterministic averaging.
SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
Yanjin He, Cuifeng Shen, Wanzhou Lei, Maohua Li, Tao Lan, Kan Liu, Hanlin Tang, Qing Qu
SWIFT uses a weaker diffusion model to help train a stronger one, aiming for better image generation without extra inference-time cost.