Ideogram released open weights for Ideogram 4 on June 3, making it the lab’s first publicly available image model and the only open-weight release trained natively on structured JSON captions rather than free text.
The architectural bet is that free-text prompting has been the binding constraint for production image generation. Ideogram 4 accepts a JSON object specifying subject placement, bounding-box coordinates for each element, a hex color palette, and text content per region. The model was trained exclusively on this format, so plain-text inputs are internally rewritten by a “magic prompt” LLM before inference.
At 9.3B parameters, it outputs natively at 2K resolution without upscaling. Multilingual text rendering covers Latin, Cyrillic, CJK, Arabic, and Devanagari scripts. In a blind typography evaluation by ContraLabs, ten professional designers picked Ideogram 4 as the best of four models 47.9 percent of the time, ahead of Gemini 3.1 Flash Image Preview at 30.0 percent. The license is non-commercial only.
Teams building design-automation pipelines should note that JSON-structured prompting aligns directly with programmatic workflows, making Ideogram 4 a candidate for any product that generates branded visuals at scale.
Ideogram on GitHub (github.com/ideogram-oss/ideogram4), 2026-06-03.
