在一项突破性的公告中，Ideogram AI 揭开了其最新力作 —— Ideogram 4，这是一款开源图像生成模型，有望革新视觉智能…

In a groundbreaking announcement, Ideogram AI has unveiled its latest masterpiece, Ideogram 4, an open-source image generation model that promises to revolutionize the field of visual intelligence. This state-of-the-art foundation model, trained entirely from scratch and available for public use, marks a significant leap forward in text-to-image synthesis, offering unparalleled control and quality. **Background and History:** Ideogram AI, founded with the mission to drive innovation through openness, has been at the forefront of developing advanced models that bridge the gap between human creativity and machine learning. Ideogram 4 builds upon this legacy by introducing a new structured JSON prompting interface, which provides users unprecedented control over composition, style, lighting, color palette, typography, and spatial layout. The model is integrated with both Diffusers and Flow-Matching architectures, ensuring compatibility across various platforms. **Key Features:** - **Structured JSON Prompting:** Ideogram 4 supports a new structured JSON caption format for enhanced user control. - **Best-in-Class Text Rendering:** With 9.3 billion parameters, the model delivers superior text rendering quality compared to larger models like Qwen-Image (20B), FLUX.2 [dev] (32B), and HunyuanImage 3.0 (80B MoE). - **Dual-Branch Classifier-Free Guidance:** The model offers independent refinement of conditional (positive) and unconditional (negative) branches, enabling separate control over prompt adherence and image quality. - **Flexible Resolution Support:** Native support for resolutions from 256 to 2048 pixels with aspect ratios up to 6:1. **Performance Metrics:** Across various benchmarks, Ideogram 4 has demonstrated exceptional performance: - In the Design Arena, it ranks as the top open-weight model. - On LMArena, it is a top-5 image generation lab overall. - For internal human-preference evaluations focused on graphic design and photography, Bradley-Terry scores rank Ideogram 4 #2. **Technical Details:** The inference code for Ideogram 4 lives in the ideogram4 GitHub repository. Users can install the model via pip or use it through a hosted magic-prompt API pr…

在一项突破性的公告中，Ideogram AI 揭开了其最新力作 —— Ideogram 4，这是一款开源图像生成模型，有望革新视觉智能领域。这个前沿的基础模型完全从头训练，并可供公众使用，标志着文本到图像合成领域的重大飞跃，提供前所未有的控制力和质量。 **背景与历史：** Ideogram AI 以通过开放性推动创新为使命创立，一直走在开发弥合人类创造力与机器学习之间鸿沟的高级模型的最前沿。Ideogram 4 在这一传承的基础上更进一步，引入了全新的结构化 JSON 提示（prompting）接口，为用户在构图、风格、光照、色彩调色板、排版和空间布局方面提供前所未有的控制力。该模型同时集成了 Diffusers 和 Flow-Matching 架构，确保在各种平台上的兼容性。 **主要特性：** - **结构化 JSON 提示：** Ideogram 4 支持全新的结构化 JSON 描述（caption）格式，以增强用户的控制力。 - **业界领先的文本渲染能力：** 该模型拥有 93 亿参数，与更大的模型（如 Qwen-Image (20B)、FLUX.2 [dev] (32B) 和 HunyuanImage 3.0 (80B MoE)）相比，能提供更出色的文本渲染质量。 - **双分支无分类器引导（Classifier-Free Guidance）：** 该模型可对条件（正向）和无条件（负向）分支进行独立优化，从而分别控制提示遵循度和图像质量。 - **灵活的分辨率支持：** 原生支持 256 至 2048 像素的分辨率，长宽比最高可达 6:1。 **性能指标：** 在各种基准测试中，Ideogram 4 展示了卓越的性能： - 在 Design Arena 上，它被评为顶级开放权重模型。 - 在 LMArena 上，它是总体排名前 5 的图像生成实验室。 - 在专注于平面设计和摄影的内部人类偏好评估中，Bradley-Terry 分数将 Ideogram 4 排在第 #2 位。 **技术细节：** Ideogram 4 的推理代码托管在 ideogram4 GitHub 仓库中。用户可以通过 pip 安装该模型，或通过托管的 magic-prompt API 使用……