Qwen Image 2.1 Edit + Realign

Say what to change in plain words, and Qwen Image 2.1 edits your picture without shifting it. Qwen often draws an edit slightly zoomed in or moved. Realign to Source puts it back on top of your original.

How to use it

1. Load your image.

2. Say the change in plain words, like "make it a soft watercolor painting".

3. Run.

Qwen3-VL writes the full edit instruction from your short request.

What you need

• AusBoss nodes 2.3.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.

• The Qwen Image 2.1 models, on a recent ComfyUI (I tested 0.37.0):

  • 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

• Optional: qwen-image-2.1-consistency.safetensors → models/loras · 159 MB. Its row is off, so it runs without it.

Tips

• Say where a moved arm or hand ends up. "put her arm on the table" kept her clasped hands and added a third one. "put her arm on the table, her hands aren't clasped anymore" worked.

• Check what Qwen got when an edit misses. Double-click Write the instruction to read the full instruction, then say the missing part more plainly.

• The Consistency LoRA holds shapes on restyles. Turn its row on for those, and set Load Image + Pad's four pads to 0 inside Edit with Qwen Image 2.1. Leave it off for pose changes: it holds the old pose.

• Realign fixes the frame, not redrawn details. Something the edit redrew in a new spot stays there.

• About 9 seconds a run on an RTX 5090, once the models are loaded.

Credits

Qwen Image 2.1 is by Alibaba's Qwen team, under the Qwen Research License: non-commercial use only. Comfy-Org repackaged the files. The Consistency LoRA is mine, under the same license.

v1.1 (2026-09-30): the same edit in a smaller, tidier workflow, and the prompt writer changes only what you ask for. v1.0 (2026-09-28): first release.

Questions or bugs: GitHub issues or the comments here. More from me on X @Zanzibased and GitHub.