The project studies model design, text and noise conditioning, backbone architecture, synthetic captions, prompt rewriting, dataset mixing, and training/evaluation choices. It positions openness as a core contribution by releasing the model, code, and data to support controlled follow-up research.
i1 is useful for researchers who want to understand which concrete design choices move text-to-image quality rather than only use a finished model. The project page links to paper, code, model, and data resources.
