Qwen Image 2.1 Generate + Edit | Full Model, Auto Prompt

Say what to change in plain words, and the full Qwen Image 2.1 model edits your picture. Switches write the prompt for you, upscale the result to 8 megapixels, make it about three times faster, or make a new picture instead.

Want the 8-step Turbo checkpoint instead? That's my Turbo version of this workflow.

How to use it

1. Load your picture.

2. Write the change, like "make it snow". With Auto prompt on, a few words are enough.

3. Run. A before and after slider shows your picture against the edit.

4. To make a new picture instead, switch Edit a picture off, write a prompt and set the size in Latent Size.

The four switches

• Edit a picture. On as the file opens: the prompt changes the picture you load, at 2 megapixels in your picture's own shape. It also turns my Consistency LoRA on at 0.5, which keeps the edit lined up with your picture. Off: the prompt makes a new picture.

• Auto prompt. Qwen writes the full prompt from a few words of yours. For an edit it looks at your picture and writes the instruction. What it wrote shows in Show Text.

• Upscale the result. Redraws the result bigger in tiles with my Restore LoRA, to the Megapixels you set. It opens at 8.

• Fast. The Viggle Turbo LoRA with its own 6 steps instead of 25. About three times faster, a little softer.

What you need

• AusBoss nodes 2.15.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.

• ComfyUI 0.38 or newer.

• The 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

• For Edit a picture (on as the file opens), my Consistency LoRA: qwen-image-2.1-consistency.safetensors → models/loras · 159 MB

• Only for Upscale the result, my Restore LoRA: qwen-image-2.1-restore.safetensors → models/loras · 159 MB

• Only for Fast, the Viggle Turbo LoRA: Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r128.safetensors → models/loras · 680 MB

Tips

• The steps are in the Steps box, under the switches. 25 is what Qwen's own template uses. With Fast on, the picture runs Viggle's own 6 steps and the box goes to 8 for the upscale.

• For an edit, say what changes and what stays: "Change the black t-shirt to a red sweater. Keep everything else the same."

• For a restyle, raise the Consistency row to 1. A painting or anime look can come back zoomed in or shifted. At 1 it brought my restyles from about 24 px of movement down to 1.6 px. 0.5 is enough for snow or new clothes and keeps a little more skin detail.

• Removing a person? Say what they're holding goes too. Otherwise a cup or a phone can stay behind.

• Don't like the result? Press New seed on the Seed node.

• Times on an RTX 5090 at 2 megapixels: about 15 seconds at 25 steps, about 6 with Fast. The upscale to 8 megapixels adds about 85 seconds, or about 55 with Fast.

• If the run stops at a red LoRA Loader, a LoRA that is switched on is missing. Download it from the list above.

What it does not do

• It takes one picture. For an outfit from another photo, see my Turbo Multi-Image Edit page. To change only an area you paint, see my Qwen Image 2.1 Inpaint page.

• The upscale is a redraw. Fine detail is made up to fit, and small lettering can come back misspelled.

• Want it with core nodes only? Pick v1.1 in the version list and download its second file, the one ending in 3295395.

Credits

Qwen Image 2.1 is by Alibaba Qwen, under the Qwen Research License: non-commercial use only. The ComfyUI files are from Comfy-Org. The Viggle Turbo LoRA is by Viggle, for research use under the same license.

v2.0 (2026-10-20): rebuilt. Change a picture or make a new one, with Auto prompt, Upscale, Fast and a Steps box. v1.1 (2026-09-30): new layout, an empty LoRA Loader for the Consistency LoRA, and transparent PNGs no longer come back on black. v1.0 (2026-09-24): first release.

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