Qwen Releases a Prompt-Rewriting Model Qwen-Image-2.1-PE-T2I

On September 20, 2026, the model Qwen/Qwen-Image-2.1-PE-T2I was published on HuggingFace. This is not an image generator, but an auxiliary model for preparing prompts: it takes a short description in any language and turns it into a detailed English prompt with a recommended aspect ratio for the main model Qwen-Image-2.1.

What it means

The model contains 9.4 billion parameters, the weights in the repository take up 17.5 GB in safetensors format, the largest file is 4.7 GB. The architecture is DiT, it is a fine-tune of the Qwen3.5-VL 9B vision-language model. The license is the Qwen RESEARCH LICENSE AGREEMENT: the model may be used, modified and distributed only for non-commercial purposes, for research or evaluation. For commercial use, a separate license must be requested from the rights holder. When distributing, a copy of the license must be included, the authorship notice must be retained and changes must be noted. If the model is used to train another AI model, the documentation must state "Built with Qwen" or "Improved using Qwen". The developer does not specify the minimum VRAM requirements or exactly what to run the model with. The code is released as open source, so if you have a suitable GPU, the model can be deployed locally. For those who already work with Qwen-Image-2.1, this model covers the task of preparing prompts without external services. You can follow the news about generation tools in the CPU3D news feed.