This is the direct GGUF conversion of Qwen/Qwen Image.

The model files can be used in ComfyUI with ComfyUI GGUF custom nodes. Place the required models in the following folders:

Sample Workflow

Sample Output  The sample size is 1 and may not be fully representative.

Notes

Q5_K_M, Q4_K_M, and most importantly, low-bit quantization (Q3_K_M, Q3_K_S, Q2_K) use a new dynamic logic where the first/last layer retains high precision.

For comparison, see this imgsli page. Using this method, even Q2_K can still be partially utilized.

Since this is a quantized model rather than a fine-tuned model, all the same limitations/original licensing terms still apply.

The main features of Qwen Image include:

• Text rendering capability: Qwen Image excels in complex text rendering, supporting multi-line layouts, paragraph-level text generation, and fine-grained detail presentation. It can achieve high-fidelity output in both English and Chinese.

• Consistent image editing capability: Through enhanced multi-task training paradigms, Qwen Image maintains consistency during the editing process.

• Cross-benchmark performance: Evaluations on multiple public benchmarks indicate that Qwen Image achieves SOTA across various generation and editing tasks.

The Qwen team conducted comprehensive evaluations of Qwen Image on multiple public benchmarks, including GenEval, DPG, and OneIG Bench for general image generation, and GEdit, ImgEdit, and GSO for image editing.

Qwen Image achieved state-of-the-art performance across all benchmarks. Moreover, results on LongText Bench, ChineseWord, and TextCraft for text rendering demonstrate that Qwen Image excels particularly in Chinese text rendering, significantly outperforming existing state-of-the-art models.