Boogu-Image-0.1: Open-Source Multimodal Model Approaches Closed-Source Performance

Jul 16, 2026

Boogu-Image-0.1 is an open-source family of unified multimodal understanding and generation models, including Base, Turbo, Edit, and Edit-Turbo variants. It delivers competitive performance in text-to-image generation, fast inference, instruction-based editing, and bilingual text rendering, consistently matching or surpassing other open-source models and achieving results that approach those of leading closed-source systems. The model was trained on 208.62 million unique images with a theoretical training cost of approximately $400K, and its weights, code, and recipes are released under Apache 2.0.

Why it matters: This work shows that targeted improvements and inference-time scaling can significantly boost multimodal generation performance under limited compute, advancing open-source capabilities in unified understanding and generation.

Full story at: arXiv Computer Vision ↗