Qwen-Image-2.1 Guide: Model Files

September 28, 2026

About This Model

Qwen-Image-2.1 is an image generation model from Alibaba’s Qwen team. Its distinguishing feature is that text-to-image generation and image editing live in one model, and its image-generation component is compact for this class of model, at 7B parameters (32 DiT layers). It outputs images of about 4 megapixels, such as 2048×2048 at 1:1.

The publisher highlights four improvements:

  • Native support for transparent (RGBA) images. The same model can generate images with transparent backgrounds from text, edit transparent layers, and cut subjects out of photos. For logos, icons and compositing assets, no separate background-removal step is needed.
  • Up to 10 reference images. You can place a person or product in a new scene while preserving their appearance. Areas to edit can be marked with circles, painted annotations, or a separate mask.
  • Better typography, portrait lighting and fine texture, according to the publisher.
  • A lightweight design (mixed-granularity attention and prefix KV cache reuse) keeps compute costs down.

About the “PE" companion models: Two helper models that rewrite short requests into detailed image prompts were released alongside it. PE-T2I (for text-to-image) turns a brief request in any language into a detailed English prompt plus a recommended aspect ratio; PE-I2I (for image editing) takes a vague editing instruction and the input image(s) and produces a precise, actionable instruction. Both are fine-tuned from Qwen3.5-VL 9B. This site’s article on the official models and the “Distributed Files" table below cover PE-I2I, not the image generator itself.

What Makes It Stand Out

The model card does not include a numeric comparison with other models, so this page does not discuss benchmark numbers and focuses on capabilities instead.

  • Generation, editing and transparency handled by one model is the headline. Work that used to require a generation model, an editing model and a background-removal tool can run in a single pipeline.
  • The small 7B generation component helps when running on local GPUs, though total memory use—including the text-understanding part—depends on the format you use.

Running It Locally

  • The official route is diffusers’ QwenImage21Pipeline (at the time of writing, you need to install diffusers from GitHub).
  • Community GGUF builds are also available. Articles on this site include an “Uncensored" build from a third party with the safety tuning removed; its weights differ from the official model, so it is a separate model.
  • The license is the Qwen Research License, which limits use to research and evaluation. Commercial use requires a separate license from the publisher, and redistributions must include the required attribution notice.

Sources: model cards and license for Qwen/Qwen-Image-2.1, Qwen/Qwen-Image-2.1-PE-T2I and Qwen/Qwen-Image-2.1-PE-I2I, as of 2026-09-25.

Our Coverage and Data

Everything Local Model Watch has published about the Qwen-Image-2.1 family: 3 article(s) covering the base model and its fine-tunes, plus converted builds we tracked after publication. Memory requirements below are computed by this site from file sizes, not quoted from model cards. Part of our model family index.

At a Glance

Item Value
Base model(s) Qwen/Qwen-Image-2.1, Qwen/Qwen-Image-2.1-PE-T2I
Publisher Alibaba (Qwen)
License (model card) other
Articles 3

Can You Run It Locally?

The publisher distributes this model as safetensors.

License — other: A custom license from the publisher. Check the original terms directly, including whether commercial use is permitted.

Compiled by this site’s code from the published formats and the license field. License summaries are not legal advice — check the publisher’s original terms before relying on them.

This assessment is for Qwen/Qwen-Image-2.1-PE-I2I.

Distributed Files

Weight files published in Qwen/Qwen-Image-2.1-PE-I2I, listed by this site from the Hugging Face API. Sizes are the actual file sizes.

File Size
model-00001.safetensors 5.07GB
model-00002.safetensors 5.09GB
model-00003.safetensors 5.09GB
model-00004.safetensors 3.57GB

Quantized and Converted Variants

→ Scroll horizontally to see all columns

Added Publisher Format Repository Smallest VRAM tier (build, est. memory)
2026-09-23 unsloth GGUF unsloth/Qwen-Image-2.1-GGUF Q3_K_XL 4.0GB (fits in 4GB VRAM)
2026-09-23 unsloth FP8 unsloth/Qwen-Image-2.1-FP8 FP8 8.0GB (fits in 8GB VRAM)

File sizes of each build:

  • Available builds in unsloth/Qwen-Image-2.1-GGUF: Q2_K 2.3GB / Q3_K_S 2.5GB / Q3_K_M 3.0GB / Q3_K_XL 3.4GB / Q4_K_S 3.6GB / Q4_K_M 3.9GB / Q5_K_S 4.2GB / Q5_K_M 5.0GB / Q6_K 5.8GB / Q6_K_XL 6.3GB / Q8_0 7.1GB / F16 13.3GB
  • Available builds in unsloth/Qwen-Image-2.1-FP8: FP8 6.6GB / INT8 6.8GB / FP8(Qwen-Image-2.1-text_encoder-FP8) 8.7GB

This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files.

Articles (the family’s own models first, then newest)

Published Model Type Article
2026-09-20 Qwen/Qwen-Image-2.1-PE-I2I Image, Video and Audio Qwen-Image-2.1 Released: Open-Weight Image Gen & Editing
2026-09-23 Viggle/Qwen-Image-2.1-viggle-turbo Image, Video and Audio Qwen-Image-2.1-viggle-turbo Image Generation Model: 48GB+ VRAM
2026-09-21 abenzerps/Qwen-Image-2.1-Uncensored-GGUF Image, Video and Audio Qwen-Image-2.1-Uncensored-GGUF Image Generation Model: 16GB+ VRAM

Repositories

Last updated 2026-09-28 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log.