AI Influencer Master Class - Qwen Image 2.1 [EDIT] (Patreon)

Load a photo, say what you want changed in plain English, and get the edit back in seconds. New outfits, new locations, new lighting, products in her hand, your brand on her clothes, and brand new shots of the same girl, all from one photo.

What you can do with it

Virtual try-on. Put any outfit from any shop on your character.

Teleport her anywhere. Send her anywhere in the world.

Relight the scene. Same photo, a completely different mood.

Brand deal content. Put your client's product in her hand.

Merch and text. Your brand on her clothes, printed for real.

Mirror selfies. One photo turns into a brand new shot.

Members get the workflow, a full gallery of before and after examples, the exact prompts behind every one, and my tips for getting results like these on the first try.

Everything you need is inside the workflow. Every model download link, which folder each file goes in, and step by step instructions are in the How To Use note on the left side of the workflow. Open the workflow, read that note top to bottom, and you're set up. The only custom node is rgthree-comfy, and you can install that from the Manager.

Want to zoom in? The full-size 4x screenshot is attached to this post as well.

What's inside

Turbo LoRA from Comfy-Org with the official 8-step schedule, so edits come back fast.

Qwen's official prompt enhancer. It looks at your photo, thinks it through, and turns your one-liner into a detailed edit prompt. One red switch turns it on or off.

Consistency LoRA keeps your edit on the original frame, so the pose, crop and layout don't drift.

Results match your photo's shape, whatever size or aspect ratio you load.

Optional second image for outfits, products, people or backgrounds.

Before / After viewer right in the workflow. Hover over it and move your mouse left and right.

How I made each example

Default settings on every one. No inpainting, no masks, no Photoshop. The box under each card is exactly what I typed.

1. Virtual try-on

This is the one people ask me about the most. Grab a product photo of any outfit straight off a shop website, load it as Image 2, and tell Qwen to put it on your character. Qwen copies the cut, the satin shine and the lace trim, and fits it to her body while everything else in the photo stays exactly where it was.

Tip: Turn on the two Image 2 nodes first (select both, Ctrl+B). In the prompt, the photo you're editing is always <image1> and the outfit is <image2>.

2. Teleport her anywhere

One boring selfie becomes a travel post. Qwen swaps the whole location and, more importantly, relights her so she actually looks like she's standing in the sunset instead of pasted on top of it.

Tip: Name the light you want in the new place (golden hour, harsh noon sun, club lighting). That's what sells it.

3. Relight the scene

Same photo, same pose, completely different mood. You can take one shoot and spin it into a whole set: daylight, golden hour, neon, candlelight, phone flash.

Tip: Start the prompt with "Relight" and end it with what has to stay. The Consistency LoRA keeps the frame locked so only the light changes.

4. Brand deal content

If you run an AI influencer, this one pays the bills. Load the client's product photo as Image 2 and put it in her hand. Qwen keeps the label readable, and that's the part brands actually check.

Tip: When the edit needs a new hand or body position, bypass the Consistency LoRA (select it, Ctrl+B). It's built to stop things moving, so it fights pose changes.

5. Merch and text

Put your brand, your name or your handle on her clothes. Ask for a real print style (faded screen print, embroidery, puff print) and say it sits under the same light and wrinkles as the shirt. That's what stops it looking pasted on.

Tip: Anything inside double quotes gets spelled out letter for letter. Keep it to one to three words and say where it goes (across the chest, on the back, on the cap).

6. Turn it into a mirror selfie

One photo, two posts. Qwen keeps her face, body and outfit and rebuilds the shot as a full-length mirror selfie with her phone up. Perfect for filling a carousel or a story without another render from scratch.

Tip: This is a new pose, so bypass the Consistency LoRA (select it, Ctrl+B). Re-roll the seed a couple of times and keep the best one, poses vary a lot between seeds.

What the prompt enhancer actually does

Here's the merch example. This is what I typed:

Give the white tank top worn by the woman in <image1> a vintage screen-printed "LoRAtech" logo across the chest, in soft faded charcoal ink with a slightly worn, cracked texture, stretched and curved by the fabric over her chest. The print sits under the same sunlight, shadows and wrinkles as the rest of the shirt, so it looks like it was always part of the shirt. Keep her face, hair, pose and the car interior from <image1>.

And this is part of what the enhancer handed to Qwen:

Apply a vintage screen-printed logo reading "LoRAtech" across the chest of the white tank top worn by the woman in the input image, rendered in soft faded charcoal ink with a slightly worn, cracked texture, the lettering stretched and curved to follow the contour of the fabric as it drapes over her chest, and integrate the print seamlessly into the garment by seating it under the same directional sunlight, the same diagonal shadow band, and the same wrinkles and folds as the surrounding fabric so it appears to have always been part of the shirt. While adding this print, keep the woman's facial features and...

It works out exactly where the print sits, how it should look on the fabric, and lists everything in the photo that has to stay the same. That last part is why the edits stay so clean. You can see the full prompt in the Final prompt sent to Qwen box every time you run it, so if you get a result you love, copy that prompt and reuse it with the enhancer switched off.

Quick rules that save you re-rolls

Say the change first, then what has to stay: "Change X. Keep her face, hair and pose from <image1>."

Always call the photos <image1> and <image2>. "Picture 1" means nothing to Qwen.

Consistency LoRA on for outfits, backgrounds, lighting and text. Off for new poses, head turns, new camera angles and full restyles.

Exact words go in double quotes.

Set the seed to fixed while you tweak a prompt, so you only see what your change did.

The workflow file is attached below. Drop it into ComfyUI, open the How To Use note, and have fun with it.