Qwen Image Edit 2509 · Face Swap

ABOUT THE WORKFLOW

Swap a Face Between Two Photos Upload a main photo and a reference face. Paint a mask over each face so the crop knows where to look. The model reads the pose from the main photo, paints in the reference face, matches the lighting and colour, and stitches the result back into the full scene.

Model

• Qwen Image Edit 2509 by the Qwen team at Alibaba. The September 2025 update to the 20 billion parameter Qwen Image Edit model. Added multi-image editing and stronger identity preservation across poses and styles. Ranked third overall and first among open-weight models on the Artificial Analysis Image Editing Arena at launch.

• Qwen Image Lightning LoRA. An 8-step speed adapter that cuts sampling down from the full schedule.

HOW IT WORKS

Step 1. Upload the main image The photo with the body and the scene. Paint a mask over the face and hair you want replaced. Accuracy at the hairline matters. Works great with: portraits · headshots · group photos · editorial shots

Step 2. Upload the reference face The photo with the face to swap in. Paint a rough mask over the face so the crop finds it. This mask can be loose.

Step 3. Hit run The model reads the pose from the main photo, swaps in the reference face, blends colour and lighting across two sampling passes, and stitches the result back into the full scene.

Step 4. Compare and download The swapped photo is saved under FaceSwapQwen/FaceSwap. A comparison grid of the original and the result is saved under FaceSwapQwen/Demo. Ready for: Photoshop · Figma · Canva · any editor

First time? Leave every setting as-is. The defaults (8 steps · colour match 0.3 · LoRA strength 1 · fixed seed) are the right starting point for almost everyone.

RECOMMENDED SETTINGS

Quick-start guide. Find the goal that matches yours and copy the settings.

• Standard swap (most people) — 8 steps · colour match 0.3 · LoRA strength 1 · fixed seed. The right starting point for almost everyone.

• The face does not look like the reference — Try a different seed first. If it still drifts, check that the main mask covers the full face and hairline accurately.

• The swapped face's colour is off — Raise the colour match value above 0.3. It controls how strongly the swapped face is recoloured to match the lighting of the main photo.

• The colour match is too aggressive — Lower the value toward 0. At 0 the model keeps the reference face's original tones.

• The background shifts after the swap — The second sampling pass blends the face into the scene. If it changes too much outside the mask, try a different seed or check that the mask does not spill past the hairline.

• Want to preserve more character detail — The ReferenceLatent node inside the Image Gen group is available but can cause artifacts. Try enabling it for simple faces and bypassing it for complex ones.

• Reproduce a result you liked — Both seeds ship on fixed. The same photos, masks, and seed numbers give you the same swap back.

Prompt: The default prompt merges the reference head onto the main body and asks the model to preserve the lighting, environment, and background. Leave it as-is for most swaps. Change it only when you want something specific, like keeping the reference person's hair colour rather than matching the main photo.

LEARN

📹 Videos

• Intro to Floyo

• ComfyUI 101 Free Course ft. Sebastian Kamph

• Floyo 101 for Team Collaboration

✨ Quick links

• Help docs

• LoRA Training guide

• Discover more workflows

USE CASES

📸 Headshot and Portrait Work Replace one face with another in a studio-lit portrait while the pose, wardrobe, and background hold.

🎭 Creative and Concept Work Place a character's face into a different scene, style, or period for mood boards and pitch decks.

🛍️ Model Stand-ins Swap a placeholder face for the final model's face without reshooting the set.

🎞️ Continuity Fixes Correct a face in a frame where the wrong take was used, without re-rendering the full shot.

WHAT WORKS BEST / WHAT TO AVOID

✅ Works great

• Front-facing or three-quarter angles on both photos

• Accurate masks on the main image, especially around the hairline

• Similar lighting directions between the two photos

• One face per swap

⚠️ May produce softer results

• Extreme angle differences between main and reference

• Rough or spilling masks on the main image

• Fine hair detail when the mask edge is imprecise

• Multiple faces in one pass

FAQ

What is Qwen Image Edit 2509? Qwen Image Edit 2509 is the September 2025 update to Alibaba's 20 billion parameter Qwen Image Edit model. It added multi-image editing, supporting combinations like person plus person and person plus scene, and improved identity preservation across poses and styles. It ranked third overall and first among open-weight models on the Artificial Analysis Image Editing Arena at launch.

How does the face swap keep identity? Qwen Image Edit 2509 was built to preserve facial identity across different poses and styles, which is what makes it a strong base for face swaps. The workflow adds to that by reading the pose from the main photo via DWPose, matching colour and lighting with a colour-match step, and blending with a two-pass sampler, so the swapped face sits in the scene rather than looking dropped in.

Why do both photos need painted masks? The masks tell the crop where to look. Without them the workflow has no anchor for the face region. The main mask needs to be accurate around the hairline because that is the blending boundary. The reference mask can be rough because it only needs to find the face for cropping.

What is the Lightning LoRA doing? It is a speed adapter that compresses the sampling schedule to 8 steps. Without it the model would need a full schedule of 30 or more passes. It is loaded at strength 1 by default and can be lowered if you want more of the base model's undistilled behaviour.

Is Qwen Image Edit 2509 free for commercial use? Yes. It is released under Apache 2.0, which allows commercial use, modification, and self-hosted deployment with no revenue threshold and no territory restrictions. You are responsible for having the right to use the faces and photos you upload.

What if the face drifts or looks wrong? Try a different seed first. If it still drifts, check the main mask: an inaccurate hairline mask is the most common cause. If the colour is off, raise the colour match value. If the background shifts, the second sampling pass may be running too many steps, and reducing them can help preserve the surroundings.

How to run Qwen Image Edit 2509 face swap online? You can run Qwen Image Edit 2509 face swap online through Floyo. No installation, no setup, no model downloads. Open the workflow in your browser, upload your two photos, paint the masks, and hit run. Free to try.

WHY FLOYO?

Floyo is the only platform with team collaboration for ComfyUI in the browser. You run workflows with no install. You share run history, assets, and models across your team. You pay only when you generate. Floyo supports open-source and closed-source models.

A designer runs an edit and likes the result. A teammate opens that exact run from shared history and keeps going. No file handoffs. No version confusion.

For studios and enterprise teams, Floyo adds private workspaces, pooled resources, and a team usage dashboard. Other ComfyUI cloud tools run for one person at a time. Floyo runs for the whole team, with transparent per-generation costs.

Ready to try it? Upload your two photos, paint the masks, and run it. The settings are already set.

→ Launch Workflow, Free

Questions? Watch the free course or check the FAQ above.