First Open-Source Text-to-Image Model Launched by Tongyige: Breaking Down Barriers in AI Image Generation, Stunning Complex Text Rendering and Editing Effects
Phoenix Net Technology News August 5, Tongyige officially announced the open-sourcing of its 20B MMDiT model Qwen-Image, which is the first image generation foundation model in the Tongyige series, showing significant progress in complex text rendering and precise image editing.
This model boastsexcellent text rendering capabilities,consistent image editing abilities, andoutstanding cross-benchmark performance.
In multiple open-benchmark tests, including GenEval, DPG, OneIG - Bench (general image generation), GEdit, ImgEdit, GSO (image editing), LongText - Bench, ChineseWord, TextCraft (text rendering), and others, Qwen-Image achieved SOTA in various generation and editing tasks.
In terms of performance, Qwen-Image can achieve high-fidelity text rendering in various scenarios. For example, in poster creation, it not only accurately displays the poster style but also retains the pose and expression of characters, generating specified Chinese and English text; in modular cases, it can complete layout and generate icons, titles, and introduction texts for each part; and even when the paper area is small and the text is long, it can accurately generate text, allowing for flexible switching between languages.
Meanwhile, Qwen-Image supports various artistic styles in general image generation, from photorealistic to impressionist painting, from cartoon-style to minimalist design, and can respond flexibly to creative prompts.
Tongyige expresses the hope that Qwen-Image will further drive the development of the image generation field, lower the technical barrier for visual content creation, and inspire more innovative applications, while looking forward to community participation and feedback to build an open, transparent, and sustainable generative AI ecosystem.
The model is currently available in the Magic Camp community and Hugging Face's open-source platform.