Paper 2508.02324
Qwen-Image Technical Report
- Published
- Aug 2025
- Research lab
- Qwen
- Citations
- 875
- GitHub
- 8.2K stars
01 In brief
Summary
Qwen-Image is a new image generation foundation model from the Qwen team, designed to excel in complex text rendering and precise image editing.
It uses a comprehensive data pipeline and a progressive curriculum learning strategy to enhance text rendering, especially for Chinese.
For editing, it employs a multi-task training paradigm with dual-encoding (semantic features from Qwen2.5-VL and reconstructive features from a VAE) to balance semantic consistency and visual fidelity.
The architecture includes a frozen Qwen2.5-VL as condition encoder, a VAE with a single-encoder dual-decoder design, and an MMDiT backbone with a novel MSRoPE positional encoding.
Training uses flow matching, a Producer-Consumer framework, and hybrid parallelism.
Post-training includes SFT and RL (DPO and GRPO).
Evaluations show state-of-the-art performance on benchmarks like GenEval, DPG, OneIG-Bench, GEdit, ImgEdit, and GSO, and significant gains in text rendering, particularly for Chinese, on LongText-Bench, ChineseWord, and CVTG-2K.
The model also demonstrates strong capabilities in novel view synthesis and depth estimation, positioning it as a versatile multimodal foundation model.
02 From the paper
Abstract
We present Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. To address the challenges of complex text rendering, we design a comprehensive data pipeline that includes large-scale data collection, filtering, annotation, synthesis, and balancing. Moreover, we adopt a progressive training strategy that starts with non-text-to-text rendering, evolves from simple to complex textual inputs, and gradually scales up to paragraph-level descriptions. This curriculum learning approach substantially enhances the model's native text rendering capabilities. As a result, Qwen-Image not only performs exceptionally well in alphabetic languages such as English, but also achieves remarkable progress on more challenging logographic languages like Chinese. To enhance image editing consistency, we introduce an improved multi-task training paradigm that incorporates not only traditional text-to-image (T2I) and text-image-to-image (TI2I) tasks but also image-to-image (I2I) reconstruction, effectively aligning the latent representations between Qwen2.5-VL and MMDiT. Furthermore, we separately feed the original image into Qwen2.5-VL and the VAE encoder to obtain semantic and reconstructive representations, respectively. This dual-encoding mechanism enables the editing module to strike a balance between preserving semantic consistency and maintaining visual fidelity. Qwen-Image achieves state-of-the-art performance, demonstrating its strong capabilities in both image generation and editing across multiple benchmarks.