Qwen Edit is a powerful image editing model, but even the latest 2511 version still suffers from a well-known problem called pixel drift. Pixel drift means the edited output does not perfectly align with the input image. It may shift slightly, zoom in or out, or move by a few pixels. While the drift is small, it is noticeable and problematic when you need pixel-perfect editing.

Even though version 2511 improves performance, one major issue remains: image quality degradation after editing. For example, when relighting a photo, the lighting result may look acceptable, but the final image quality is noticeably worse than the original.

Video Tutorial:

This tutorial shows how to:

1. Reduce or eliminate pixel drift
2. Restore lost details
3. Handle both global and local editing scenarios
4. Use ComfyUI workflow techniques to maintain alignment

---

Key Concept: Two Types of Editing

This guide covers two major use cases:

1. Global Editing Editing the entire image (for example changing the whole background)
2. Local Editing Editing only a selected part of the image (for example changing clothing color)

Different strategies are required for each scenario.

---

Part 1: Global Editing Without Pixel Drift

Global editing means the model modifies the whole image. A typical example would be changing the background to a sunset beach while keeping the subject intact.

Step 1 — Pad the Image to a Square

Use the Image Pad KJ node.

• Convert the image to a square format
• Example: original becomes 3000 × 3000
• Gray padding appears on unused sides

This ensures a uniform canvas for Qwen Edit.

---

Step 2 — Resize to Qwen-Friendly Resolution

Use Image Resize to scale the square image to something like 1024 × 1024.

Then:

• Send this resized image to the text encoder
• Include a prompt instruction telling Qwen Edit to keep the gray padding unchanged

This prevents unwanted distortion.

---

Step 3 — Generate the Edit

Run Qwen Edit. Result:

• No pixel shift
• Perfect alignment with original
• However, visible detail loss (face, clothing, textures degrade)

So the drift is fixed, but quality is not yet acceptable.

---

Step 4 — Restore Original Resolution

Resize the edited image back to 3000 × 3000 to match the padded image.

---

Step 5 — Remove the Padded Borders

Use Image Crop By Mask to remove gray padding.

Now you have a full-resolution edited image, aligned correctly. But detail quality is still weak, especially on the subject.

---

Step 6 — Restore Original Subject Detail

Since only the subject needs restored detail:

1. Use Remove Background to isolate the subject
2. Use Image Detail Transfer to transfer original detail back onto the edited output

Result:

• Background remains edited
• Subject recovers original high quality detail
• Pixel alignment remains perfect

This produces significantly better results than other tools such as Nano Banana Pro in this scenario.

---

Part 2: Local Editing Without Pixel Drift

Local editing means modifying only a specific area, such as changing dress color while keeping the rest untouched.

You will generally work with:

• Manual mask painting
• Cropping
• Special stitching workflows

---

Method 1 — Using Qwen Edit Utils (Jason’s Pack)

This method uses specialized nodes to try to reduce drift. However, testing shows:

• Pixel drift still occurs
• Not ideal for precision editing

So while useful, it is not the best solution.

---

Method 2 — The Recommended Drift-Free Approach

Step 1 — Crop Using Inpaint Crop Node

Use Inpaint Crop to extract a 1024 × 1024 section based on your painted mask.

Important notes:

• The node adds some margin padding
• The final crop becomes exactly 1024 × 1024
• This fixed resolution prevents pixel drift during editing

---

Step 2 — Prepare Mask for Conditioning

Preview the cropped mask. Then feed this mask into Inpaint Model Conditioning to ensure:

• Only masked region is edited
• Areas outside remain untouched
• Edits stay contained

---

Step 3 — Run Qwen Edit on Cropped Image

Apply your edit (for example change dress color to red).

Result:

• No pixel drift
• The face and surrounding areas remain unaffected
• Alignment remains perfect

---

Step 4 — Stitch Back to Original Image

Use Inpaint Stitch to merge the edited region back into the full-resolution image.

This completes the drift-free local edit.

---

Issue: Loss of Fine Detail

Because:

• Original image might be 2000 × 3000
• Cropped edit is only 1024 × 1024

This downscaling removes detail. After stitching, the edited section may look lower quality than the original.

---

Fixing Lost Detail After Local Editing

To restore high quality:

• Add a high-resolution fix sampler or
• Upscale using SeedVR2 nodes then
• Feed the enhanced result into inpainted_image input in Inpaint Stitch

This restores detailed texture quality in the edited region.