Show HN: Ideogram 4.0 – open-weight 9.3B text-to-image model (github.com)

Ideogram has launched Ideogram 4.0, its first open-source text-to-image model, featuring an innovative architecture trained from scratch rather than fine-tuning existing models. This state-of-the-art 9.3 billion parameter model introduces a structured JSON prompting interface that facilitates enhanced multilingual text rendering and allows unprecedented control over layout and color palettes. Users can generate high-resolution images (up to 2048 x 2048 pixels) online at ideogram.ai, making it accessible for designers and developers alike. Significantly, Ideogram 4 is positioned as the leading open-weight image model, surpassing its competitors in various performance benchmarks related to image generation and design usability. Its unique design features, including explicit bounding-box for subject placement and color palette specification, cater directly to design needs, demonstrating robust cross-modal interaction and high fidelity in text rendering. The model’s capacity to deliver tailored visual content efficiently positions it at the forefront of visual intelligence, inviting the AI/ML community to explore its capabilities for creative applications.