Qwen Image 2.1 Edit — Simple, with A/B Compare

The plain Qwen Image 2.1 edit graph: a source, an instruction, a matching canvas, and a Compare slider before you save. Core Qwen nodes plus my free AusBoss nodes, nothing else.

The first run makes it snow on the train-platform portrait from the sample inputs: soft flakes through the flash light and a thin layer of snow on the platform, with her face, pose and outfit unchanged, at 1 MP in about 7 seconds on an RTX 5090.

What's inside

• Working image: Image Resize sets a 1 MP working image and snaps its edges to 32 pixels (736 × 1344 for the sample).

• Instruction: a Text node. Name what should change, name what to keep, and refer to the source as <image1>.

• Encode: Text Encode Qwen Image 2.1 reads the source with its VAE and returns the conditioning plus an empty latent that matches it. Resolution 0 keeps the already resized picture as it is.

• Reference cache: the core Qwen Image 2.1 cache node, left on auto.

• Sample: 25 steps, CFG 1, Euler / simple, fixed seed.

• Compare + save: the result slides against the working image; Save Image writes a PNG with the workflow embedded.

• More references: connect another Load Image to the encoder's next image socket and call it <image2> in the instruction. Qwen Image 2.1 takes up to 10 reference images; the first one sets the output size.

Quick start

1. Update ComfyUI. You need a build with Text Encode Qwen Image 2.1 (mid-September 2026 or newer; I tested on 0.37.0).

2. Install AusBoss nodes from ComfyUI-Manager (search "AusBoss", registry id ausboss-nodes) and restart. You need 2.2.0 or newer, which is what Manager installs.

3. Unzip sample_inputs.zip into ComfyUI/input/ (it holds ausboss_night_train_platform.png), or load your own picture.

4. Drag the workflow in, download the three files from the card and queue.

5. For your own picture, rewrite the instruction: the change first, then what must stay.

Models

• qwen_image_2.1_int8_convrot.safetensors → models/diffusion_models · 7.26 GB

• qwen3vl_8b_int8_convrot.safetensors → models/text_encoders · 9.35 GB

• qwen_image_2.1_vae_bf16.safetensors → models/vae · 676 MB

BF16 versions are on the same page: Comfy-Org/Qwen-Image-2.1

Custom nodes

Only ComfyUI-AusBoss, 2.2.0 or newer (Manager installs it): free and open source. Text Encode Qwen Image 2.1, the reference cache and the sampler are core ComfyUI.

Manager can't find AusBoss? In the classic Manager window, if the Channel box is empty, pick default, then click Install Missing Custom Nodes again. See what to click.

Settings that worked, and what didn't

• Sampling: 25 steps, CFG 1, Euler / simple, denoise 1: the starting values of the official template. 40 steps gives a more complete render. The negative prompt has no effect at CFG 1.

• Say what to keep: the sample instruction ends with "Keep her face, hair, pose, outfit and the passing train exactly the same. Keep the original photo's look exactly: the same soft film rendering, colors, exposure, contrast and gentle grain." Naming what stays, including the photo's look, is what holds her identity and the film grain.

Speed

Measured on an RTX 5090 (32 GB) with the INT8 files and the models loaded: 6.7 to 7.0 s at 1 MP.

I haven't tested smaller cards yet. ComfyUI offloads what doesn't fit, so expect it to run slower rather than fail.

About the gallery

The snow, rain and red-knit frames are straight outputs of this workflow (seed 6202), each from the same source with a different one-line instruction; drag a PNG into ComfyUI to load its exact settings. The source portrait was made with Qwen Image 2.1. The cover is the source photo with the three edits and a screenshot of the workflow's own Edit instruction card.

License & credits

Qwen Image 2.1 is © Alibaba Qwen, released under the Qwen Research License: non-commercial use only; commercial use needs a separate license from Qwen. The workflow and the AusBoss nodes are free.

Changelog

• v1.0 (2026-09-24): first release.

Questions or bugs: GitHub issues or the comments here. Post what you make with it; I read everything. I post new workflows on X @Zanzibased and GitHub.