Boogu-Image 0.1

Boogu-Image 0.1 is an Apache-2.0 open-source image generation and editing family that aims to bring near closed-source quality to fully self-hostable weights. It ships as several 10B-parameter variants — Base for diverse, controllable text-to-image and ultra-dense text rendering, Turbo for photorealistic generation in roughly 3–4 distilled steps, and Edit for image-to-image transformations — all released on Hugging Face and ModelScope (including FP8 builds).

Its standout strength is bilingual text rendering: posters, stamps, documents, interfaces and brand layouts with stable typography across both Chinese and English. The team reports competitive results in their own Boogu Arena preference evaluation despite training on roughly an order of magnitude less data than comparable open models — a deliberate bet that better understanding, data quality and training pipelines matter more than raw scale.

The honest trade-offs are spelled out by the authors: it’s a research-oriented release without an official hosted API, image-to-image consistency still trails systems like Seedream 5.0 and Nano Banana Pro, and world knowledge (real brands, people, landmarks) lags the strongest closed models. Small faces, limbs and very dense text can also show artifacts.

For teams that want a commercial-friendly, self-hosted image model — especially for bilingual, text-heavy design work — Boogu-Image 0.1 is a compelling open option. If you need a turnkey hosted API or the absolute frontier in photorealism and editing fidelity, a closed model like Nano Banana 2 or an open API-first pick like FLUX.2 may fit better.

Best for

• Self-hosted text-to-image for privacy or cost control
• Bilingual posters, documents and text-heavy designs
• Fine-tuning and downstream research on an open base

Pros & cons

Sources

• Boogu-Image GitHub repository
• Boogu on Hugging Face
• Boogu project page