README.md

metadata

base_model: Qwen/Qwen-Image-2.1
tags:
  - fp8
  - int8
  - unsloth
  - quantized
  - diffusers
  - qwen
  - image-generation
  - image-editing
  - rgba
  - text-to-image
license: other
license_name: qwen-research
license_link: https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE
language:
  - en
  - zh
library_name: diffusers
pipeline_tag: text-to-image
widget:
  - text: >-
      Photorealistic photograph of an old movie theater with faded red seats,
      peeling walls, and a torn screen.
    output:
      url: samples/int8_2.png
  - text: >-
      A rally car speeding across vast desert dunes, throwing a dramatic trail
      of sand behind it.
    output:
      url: samples/fp8_1.png
  - text: >-
      Photorealistic editorial photograph of a woman barista making a latte in a
      modern, minimalist café on a sunny tropical morning.
    output:
      url: samples/int8_1.png
  - text: >-
      A lone astronaut crossing a dark, frozen lake beneath enormous rings
      stretching across an alien sky. Fine cracks visible under the translucent
      ice, distant mountains, soft blue twilight, cinematic scale,
      photorealistic detail.
    output:
      url: samples/fp8_2.png

Read our How to Run Qwen-Image-2.1 Guide! 💜

This is an FP8 / INT8 quantized version of Qwen-Image-2.1, plus a pre-cast FP8 copy of its Qwen3-VL text encoder.

• For higher quality, you can also run Dynamic FP8 on just 6GB of VRAM via offloading (<2x slower).
• Optimized for efficient inference with reduced memory footprint. Same-seed LPIPS vs the bf16 model (lower is better): 0.064 INT8, 0.112 FP8. INT8 is the shipped scheme.
• In Unsloth Desktop these are used automatically: pick Qwen-Image-2.1 and an INT8 or FP8 precision.
  

file	size	replaces
Qwen-Image-2.1-INT8.safetensors	7.26 GB	the 14.23 GB bf16 transformer
Qwen-Image-2.1-FP8.safetensors	7.12 GB	the 14.23 GB bf16 transformer
Qwen-Image-2.1-text_encoder-FP8.safetensors	9.39 GB	the 17.5 GB bf16 text_encoder
vae/qwen_image_2.1_vae_bf16.safetensors	0.63 GB	the fp32 vae, for unsloth/Qwen-Image-2.1-GGUF

Everything here is safetensors rather than a pickle, so loading runs no arbitrary code. Needs a diffusers main build: Qwen-Image-2.1 support postdates the 0.40 release.

---

Samples

INT8	INT8
FP8	FP8

---

🤖 ModelScope  |   🤗 HuggingFace  |   📑 Blog  |   🖥️ Demo  |   🫨 Discord  |   💬 WeChat

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.

For more details, see the GitHub repo and Blog.

Quick Start

Installation

pip install torch>=2.4.0
pip install transformers>=5.17
pip install git+https://github.com/huggingface/diffusers
pip install accelerate pillow

Text-to-Image

import torch
from diffusers import QwenImage21Pipeline

pipe = QwenImage21Pipeline.from_pretrained(
    "Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16
).to("cuda")

image = pipe(
    prompt="A neon shop sign that reads \"QWEN IMAGE 2.1\", rainy night, reflections on wet pavement",
    width=2048, height=2048,
    num_inference_steps=40,
    generator=torch.Generator("cuda").manual_seed(42),
).images[0]

image.save("t2i_example.png")

Image Editing

import torch
from PIL import Image
from diffusers import QwenImage21Pipeline

pipe = QwenImage21Pipeline.from_pretrained(
    "Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16
).to("cuda")

input_image = Image.open("input.png")

image = pipe(
    prompt="Change the background to a sunset beach",
    image=input_image,
    num_inference_steps=40,
    generator=torch.Generator("cuda").manual_seed(42),
).images[0]

image.save("edit_example.png")

Transparent Image Generation (RGBA)

Use the recommended prompt format for transparent images:

image = pipe(
    prompt="This is an RGBA image with transparency. A cute cartoon dragon sticker. The image has alpha channel and the background is transparent.",
    width=2048, height=2048,
    num_inference_steps=40,
    generator=torch.Generator("cuda").manual_seed(42),
).images[0]

image.save("transparent_example.png")

Supported Aspect Ratios

aspect_ratios = {
    "1:1":  (2048, 2048),
    "4:3":  (2400, 1792),
    "3:4":  (1792, 2400),
    "3:2":  (2528, 1696),
    "2:3":  (1696, 2528),
    "16:9": (2752, 1536),
    "9:16": (1536, 2752),
}

Memory Optimization

pipe = QwenImage21Pipeline.from_pretrained(
    "Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()

Showcase

Native transparent image generation

Group photograph generated from six portrait references

Text rendering

License

This model is licensed under the Qwen Research License Agreement.