Boogu-Image-0.1 - Boogu's open-source unified image generation and editing model - AiBoss

What is Boogu-Image-0.1?

Boogu-Image-0.1 is an open-source family of unified image generation and editing models from the Boogu team. The models support text-to-image generation, imperative image editing, and bilingual (Chinese and English) text rendering within the same architecture. The family includes three variants: Base, Edit, and Turbo. In the self-built Boogu Arena ELO benchmark for thousands of prompt words, the series ranked among the top of all participating open-source and closed-source systems, and ranked first among open-source models on Qwen-Image-Bench.

Main functions of Boogu-Image-0.1

Text-based image generationIt accurately understands photography-related prompts and generates high-quality images with natural lighting, harmonious composition, and faithful details, supporting complex and realistic scenes.
Instructional Image EditingIt supports object insertion, replacement and removal, attribute and material modification, background and scene replacement, and cross-art style migration, while maintaining consistency with the original subject and composition.
Chinese and English bilingual text renderingSupports text-intensive visual designs such as posters, stamps, documents, interfaces, and brand guides, achieving readable structure and stable layout.
Stylization generationIt covers a variety of styles, including traditional Chinese style, pixel art, picture book, and product-level rendering, and is sensitive to prompts and provides stable output.
Poster and product image optimizationGenerate personalized posters and product visualizations with a consistent brand style, exquisite typography, and professional lighting composition.
Image text editingIt supports replacing, adding, or deleting Chinese and English characters, and allows for flexible adjustment of font, weight, color, and layout.

Technical Principles of Boogu-Image-0.1

Unified Multimodal ArchitectureBoogu-Image-0.1 adopts a unified multimodal understanding and generation architecture, integrating visual understanding and image generation capabilities into a single model framework. This enables the model to simultaneously understand and generate images, providing a semantic-level control foundation for editing tasks.
Distillation accelerates reasoningThe Turbo variant is built on distillation technology and compresses the inference path with the same number of parameters as Base. It typically only requires 3-4 steps to output high-quality, realistic photographic images while retaining bilingual text rendering capabilities and cue word compliance.
Instruction alignment editingThe Edit variant (10B parameter) is specifically optimized for image editing tasks on a unified architecture. It achieves fine-grained local modifications through an instruction alignment mechanism, and completes cross-style migration and content replacement while maintaining consistency with the original subject and composition.
Bilingual text renderingThe model integrates the ability to understand and generate structured Chinese and English characters within a unified architecture, and ensures readable layout and stable rendering in text-intensive scenarios such as posters and brand materials through a layout awareness mechanism.

How to use Boogu-Image-0.1

Online experienceVisit ModelScope (https://modelscope.cn/studios/Boogu/boogu-image-edit-gradio) to directly upload images and enter editing commands; no local deployment is required.
Model DownloadObtain the weights of each variant of Base, Edit, and Turbo through the ModelScope organization page https://modelscope.cn/organization/Boogu.
Local deploymentClone the GitHub repository to obtain the training and inference code, configure the environment according to the README, and then load the model to generate or edit it.

Boogu-Image-0.1's core advantages

Unified architectureThe same model family covers the three major tasks of generation, editing, and text rendering, eliminating the need to switch between independent models for different purposes.
Speed ReasoningThe Turbo variant can generate high-quality data in just 3-4 steps, and its inference efficiency is comparable to or even surpasses that of some closed-source solutions.
Advantages of bilingual textsIt performs stably in rendering Chinese and English text, making it especially suitable for local design scenarios such as Chinese posters and brand materials.
Edit consistencyThe Edit variant maintains the original subject and composition height consistency when modifying content, avoiding the overall redraw distortion common in traditional editing models.

The project address for Boogu-Image-0.1

Project official websitehttps://boogu.org/
GitHub repository: https://github.com/boogu-project/Boogu-Image
HuggingFace model libraryhttps://huggingface.co/Boogu

Comparison of Boogu-Image-0.1 with similar competing products

Dimension	Boogu-Image-0.1	Qwen-Image-2.0
Publisher	Boogu Team	Alibaba Tongyi Team
Release time	June 2026	February 2026
Open source license	Apache-2.0	Apache-2.0
Model parameters	Edit variant 10B; the number of Base/Turbo parameters is the same as Base (specific values not disclosed).	The diffusion decoder has 7B bits, plus the Qwen3-VL encoder has 8B bits, totaling approximately 15B levels.
Core Architecture	A unified multimodal understanding and generation architecture that integrates visual understanding and image generation into a single framework.	Dual-component architecture: 8B Qwen3-VL visual language encoder → 7B MMDiT diffusion decoder
Unified capabilities	A three-in-one system for text-based image editing, instruction-based image editing, and bilingual (Chinese and English) text rendering.	Image processing, image editing, and text rendering are all unified under a single 7B model.
Maximum resolution	The examples are mostly 1024×1024 (higher resolutions are supported, but the upper limit is not specified).	Native 2048×2048 (2K), no upsampling required
Prompt word length	No explicit upper limit for disclosure	Up to 1000 tokens
Text rendering	Bilingual in Chinese and English, supporting complex layouts such as ultra-dense text, posters, and brand materials.	Bilingual in Chinese and English, supporting infographics, PowerPoint presentations, posters, calligraphy, etc., emphasizing material adaptation and structured alignment.
Reasoning efficiency	The Turbo variant requires only 3-4 steps to generate a high-quality product.	The number of optimization steps is not specified; the 7B lightweight architecture reduces memory requirements.
Local deployment	Supported. Obtain weights and code via GitHub and ModelScope.	Supported, open source on GitHub, DiffSynth-Studio supports 4GB VRAM layered unloading.

Application scenarios of Boogu-Image-0.1

e-commerce designQuickly generate product main images, detail page posters, and multilingual promotional materials, supporting style unification and partial modifications based on reference images.
Advertising and MarketingGenerates bilingual (Chinese and English) posters and social media images based on brand guidelines, and supports text replacement and layout adjustments for existing materials.
Content creationIt provides stylized illustrations, comic storyboards, and meme generation for short videos and social media, and supports precise editing of text within images.
Publishing and PrintingGenerates book covers, magazine layouts, stamps, and document interface designs, ensuring text readability and layout stability.