Click a thing in your picture and say what to change. Qwen Image 2.1 repaints only that part. The rest of your picture stays exactly as it was.
Watch it used, start to finish: <a target="_blank" href="https://youtu.be/iMgaiXoB0v4" rel="ugc">the video on YouTube</a>.
<h3>Why use this workflow</h3>
<ul>
<li>Only the part you pick changes. A plain Qwen edit repaints the whole picture, so other things can change with it. Here everything outside the picked part stays your own pixels.</li>
<li>You click, you don't paint a mask. The thing lights up under your pointer, and one click picks it. SAM 3 finds its outline.</li>
<li>Small things get full detail. The picked part is worked on at about 1.5 megapixels and put back into your picture at its own size.</li>
</ul>
<h3>How to use it</h3>
<ol>
<li>Load your image. Optional: drag the blue handles to crop it first.</li>
<li>On Mask by Touch, click the thing you want to change. It lights up before you click.</li>
<li>In What to change, say it the way you would ask a person: "Replace the cup with a glass of orange juice."</li>
<li>Run.</li>
</ol>
Before and after slides between your picture and the result.
<h3>Faster, optional</h3>
Download the Viggle Turbo file below and switch Fast on. It runs 8 steps instead of 25, and a run takes about half the time.
<h3>What you need</h3>
<ul>
<li>AusBoss nodes 2.11.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first. If Manager still shows an older version, pick 2.11.2 in its version list.</li>
<li>ComfyUI 0.38 or newer.</li>
<li>The Qwen Image 2.1 models:<ul>
<li><a target="_blank" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_int8_convrot.safetensors" rel="ugc">qwen_image_2.1_int8_convrot.safetensors</a> → <code>models/diffusion_models</code> · 7.26 GB</li>
<li><a target="_blank" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors" rel="ugc">qwen3vl_8b_int8_convrot.safetensors</a> → <code>models/text_encoders</code> · 9.35 GB</li>
<li><a target="_blank" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors" rel="ugc">qwen_image_2.1_vae_bf16.safetensors</a> → <code>models/vae</code> · 676 MB</li>
</ul>
</li>
<li>SAM 3, which finds the thing you click: <a target="_blank" href="https://huggingface.co/Comfy-Org/sam3.1/resolve/main/checkpoints/sam3.1_multiplex_fp16.safetensors" rel="ugc">sam3.1_multiplex_fp16.safetensors</a> → <code>models/checkpoints</code> · 1.75 GB</li>
</ul>
Optional, for Fast: <a target="_blank" href="https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo/resolve/main/Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r128.safetensors" rel="ugc">Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r128.safetensors</a> → <code>models/loras</code> · 680 MB. The workflow runs without it as long as Fast stays off.
<h3>Tips</h3>
<ul>
<li>Wrong thing picked? The arrows under the picture swap it for the next thing it could mean. Draw paints an area by hand.</li>
<li>Pick all of it. A second click adds a piece: the other sleeve, a strap. What you leave out stays.</li>
<li>The new thing is painted inside the pick. Ask for something much bigger and it is cut at the edge. Raise Grow under the picture to give it room.</li>
<li>To remove something, pick it and write "Remove the cup." Pick its shadow too, or the shadow stays.</li>
<li>To add something, switch to Draw, paint where it should go with some room, and say how big: "Add small black sunglasses on her eyes."</li>
<li>A hand that was under the old thing can come out odd. Run it again. The seed changes each run.</li>
<li>About 12 to 15 seconds a run on an RTX 5090. A new picture needs about 20 seconds before things light up.</li>
</ul>
On Krea 2 instead: <a target="_blank" href="https://civarchive.com/models/2971849" rel="ugc">Krea 2 Inpaint Masked</a>. To take a thing out and keep it as a transparent PNG: <a target="_blank" href="https://civarchive.com/models/2987326" rel="ugc">Qwen Image 2.1 Cut Out + Remove</a>.
<h3>Credits</h3>
Qwen Image 2.1 is by Alibaba Qwen, under the Qwen Research License: non-commercial use only. SAM 3 is by Meta, under the SAM License. The optional speed LoRA is <a target="_blank" href="https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo" rel="ugc">Viggle Turbo v0.3</a> by Viggle, under the Qwen Research License: research use only.
v1.0 (2026-10-13): first release.
Questions or bugs: <a target="_blank" href="https://github.com/ausboss/ComfyUI-AusBoss/issues" rel="ugc">GitHub issues</a> or the comments here. Videos on <a target="_blank" href="https://www.youtube.com/@ausbossdev" rel="ugc">YouTube @ausbossdev</a>. More from me on <a target="_blank" href="https://x.com/Zanzibased" rel="ugc">X @Zanzibased</a> and <a target="_blank" href="https://github.com/ausboss" rel="ugc">GitHub</a>.

Qwen Image 2.1 Inpaint

First release. Needs ComfyUI 0.38 or newer and AusBoss nodes 2.11.0 or newer: already have them? Update them in ComfyUI-Manager first, then restart ComfyUI. If Manager still shows an older version, pick 2.11.2 in its version list.

Files with SHA256: e55b9f94e59b77b02c6044de93eabee4ef65b828297788ca3483e3d3daa5e0b6