---
title: Qwen-Image-2.1-PE-I2I
canonical_url: "https://www.modelscope.cn/models/qwen/Qwen-Image-2.1-PE-I2I"
md_url: "https://www.modelscope.cn/models/qwen/Qwen-Image-2.1-PE-I2I.md"
repository: qwen/Qwen-Image-2.1-PE-I2I
last_updated: 2026-09-21
license: other
model_type:
- qwen3_5
architectures:
- Qwen3_5ForConditionalGeneration
parameters: 9.4B
tensor_type:
- BF16
library_name:
- pytorch
- transformer
- safetensors
frameworks:
- pytorch
downloads: 1178
stars: 9
tags:
- qwen
- prompt-rewriting
- image-editing
---
# Qwen-Image-2.1-PE-I2I
> Qwen-Image-2.1-PE-I2I - qwen 在 ModelScope 开源的模型。🤖 ModelScope | 🤗 HuggingFace | 📑 Blog | 🖥️ Demo | 🫨 Discord
qwen/Qwen-Image-2.1-PE-I2I 是 ModelScope 魔搭社区上的 9.4B 参数机器学习模型，采用 other 许可。
- **Repository**: qwen/Qwen-Image-2.1-PE-I2I
- **License**: other
- **Parameters**: 9.4B
- **Tags**: qwen, prompt-rewriting, image-editing
- **Downloads**: 1178
- **Stars**: 9
- **Last updated**: 2026-09-21
Source: https://www.modelscope.cn/models/qwen/Qwen-Image-2.1-PE-I2I
---
🤖 ModelScope |
🤗 HuggingFace |
📑 Blog |
🖥️ Demo |
🫨 Discord
## Introduction
We are excited to open-source **Qwen-Image-2.1**, a unified text-to-image generation and image editing model in the Qwen family. With just **7B parameters in its visual generation component** (32 Single-Stream DiT layers), Qwen-Image-2.1 balances generation quality, inference efficiency, and versatility.
Four key improvements define this release:
- **Compact and Efficient** — A lightweight architecture with mixed-granularity attention and prefix KV cache reuse delivers strong image quality at low computational cost.
- **Native Transparency, Unified Creation and Editing** — Generate regular or transparent (RGBA) images from text, edit transparent layers, and extract subjects from photographs—all in one model.
- **Versatile Editing** — Support up to **10 reference images**, specify local edits via circles, painted annotations, or separate masks, and preserve identity for people and products.
- **Realistic Textures and Refined Aesthetics** — Improved typography, portrait lighting, and fine details for more visually compelling results.
# Qwen-Image-2.1-PE-I2I
Image editing **prompt rewriting model** for [Qwen-Image-2.1](https://huggingface.co/Qwen/Qwen-Image-2.1). A fine-tuned Qwen3.5-VL 9B that takes a vague editing instruction plus input image(s) and produces a precise, actionable prompt suitable for downstream image editing.
For more details, see the [GitHub repo](https://github.com/QwenLM/Qwen-Image-2.1) and [Blog](https://qwen.ai/blog?id=qwen-image-2.1).
## Quick Start
### Installation
```bash
pip install transformers>=5.4.0 torch>=2.4.0 accelerate pillow
```
### Usage with Transformers
```python
import json
import torch
from PIL import Image
from transformers import AutoModelForImageTextToText, AutoProcessor
model_id = "Qwen/Qwen-Image-2.1-PE-I2I"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
model_id, dtype=torch.bfloat16, device_map="auto"
).eval()
# Load the system prompt shipped with the model
import huggingface_hub
sys_prompt_path = huggingface_hub.hf_hub_download(model_id, "system_prompt.txt")
system_prompt = open(sys_prompt_path).read().strip()
input_image = Image.open("input.png").convert("RGB")
user_prompt = "make the sky sunset"
messages = [
{"role": "system", "content": [{"type": "text", "text": system_prompt}]},
{"role": "user", "content": [
{"type": "image", "image": input_image},
{"type": "text", "text": user_prompt},
]},
]
inputs = processor.apply_chat_template(
messages, add_generation_prompt=True, tokenize=True,
return_dict=True, return_tensors="pt", enable_thinking=True,
).to(model.device)
with torch.no_grad():
out = model.generate(
**inputs, max_new_tokens=24000,
do_sample=True, temperature=1.0, top_p=0.95, top_k=20,
)
gen = processor.tokenizer.decode(
out[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True
)
# Split thinking from the answer
thinking, _, answer = gen.partition("")
result = json.loads(answer.strip())
print(result)
# {"rewritten_prompt": "...", "wh_ratio": "", "ratio_follow": ""}
```
### Multi-Image Editing
The model supports multiple input images — referred to as ``, ``, etc.:
```python
images = [Image.open("portrait.png").convert("RGB"),
Image.open("scene.png").convert("RGB")]
messages = [
{"role": "system", "content": [{"type": "text", "text": system_prompt}]},
{"role": "user", "content": [
{"type": "image", "image": images[0]},
{"type": "image", "image": images[1]},
{"type": "text", "text": "Place 's subject into 's scene"},
]},
]
```
### Integration with Diffusers
```python
import torch
from PIL import Image
from diffusers import QwenImage21Pipeline
# Assuming `result` and `input_image` from above
prompt = result["rewritten_prompt"]
pipe = QwenImage21Pipeline.from_pretrained(
"Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16
).to("cuda")
image = pipe(
prompt=prompt,
image=input_image,
num_inference_steps=40,
generator=torch.Generator("cuda").manual_seed(42),
).images[0]
image.save("rewritten_edit.png")
```
### Output Format
The model outputs a JSON object after a `` reasoning block:
```json
{
"rewritten_prompt": "",
"wh_ratio": "",
"ratio_follow": ""
}
```
- `rewritten_prompt` — the expanded prompt to pass to the image editing model
- `wh_ratio` — aspect ratio chosen by the model (e.g. `"16:9"`), when the task creates a new composition
- `ratio_follow` — inherit aspect ratio from an input image (e.g. `""`), when editing in-place
`wh_ratio` and `ratio_follow` are mutually exclusive — exactly one carries a value.
## License
This model is licensed under the [Qwen Research License Agreement](./LICENSE).