用纯开源训练方案打造强图像生成模型，支持精细编辑。
LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes
- 从零训练60亿参数扩散Transformer，先做图像无文本预训练
- 220万样本中含98万真实图像，生成结果在双语评测中达领先水平
- 模型可蒸馏为快速版，2-4步采样即完成生成，适合高效应用
我们提出LLaDA-Image，一个统一框架，将60亿参数的扩散Transformer（DiT）与基于LLaDA2.0-Mini扩散语言模型骨干的冻结视觉-语言理解模块结合。不同于依赖大量图文配对数据，我们通过纯图像预训练和中段训练建立强大生成先验。生成流程使用220万样本，其中98万为真实图像。为实现高效可扩展优化，采用无参数的RMSNorm和Muon优化器。最终模型生成高度逼真的图像，并准确遵循细粒度编辑指令。我们进一步将模型蒸馏为LLaDA-Image-Turbo，实现2-4步采样下的快速推理。在Qwen-Image-Bench上，该模型在英文和中文赛道分别取得53.53和53.38分，成为开源模型中的新标杆。为推动高效生成模型研究，我们开源了模型权重、训练代码及详细训练方案。
原文摘要 · Abstract (English)
We introduce LLaDA-Image, a unified framework that pairs a 6B Diffusion Transformer (DiT) trained from scratch with a frozen vision-language understanding module built on the LLaDA2.0-Mini diffusion language model backbone. Instead of relying heavily on paired image-text data from the beginning, we first build a strong visual generative prior through image-only pre-training and mid-training. The generation pipeline comprises 220M samples, 98 of which are real images. For efficient and scalable optimization, we use parameter-free RMSNorm throughout the DiT together with the Muon optimizer. The resulting unified model produces highly photorealistic images while accurately following fine-grained editing instructions. We further distill LLaDA-Image into LLaDA-Image-Turbo, enabling fast inference in 2-4 sampling steps. On Qwen-Image-Bench, LLaDA-Image achieves overall scores of 53.53 and 53.38 on the English and Chinese tracks, respectively, setting a new state-of-the-art among open-source models on both tracks. To support further research on capable and efficient generative models, we release our model weights, training code, and detailed recipes.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。