How to Set Up AI Virtual Try-On in Your Fashion Brand’s Instagram DMs

Complete Apiway guide with examples for fashion brand owners, SMM specialists,
Facebook Ads experts, UGC creators, and fashion influencers

Anton Viborniy

Choose Strategies

Instagram Ads vs Post Comment Triggers

At the beginning of this guide, I want to explain the two different strategies you can use for AI Virtual Try-On.
Both strategies are widely used by Apiway customers, but they work in different ways and serve different goals.

Click-to-Message Instagram Ads
(Send Keyword in DMs)

Your audience sees your Instagram ad featuring a garment.

They tap the ad and see a pre-built message with a keyword, for example “Dress_12K”.

When they tap Send, the message is delivered to your Instagram DMs.

(video example)

From a technical point of view, the user is sending a keyword to your DMs.
That keyword triggers a specific AI Virtual Try-On flow.

You can think of it like this:
“Send me a DM with a specific keyword, and I will send you a specific try-on experience.”

Each keyword runs its own flow.

For example:

Dress_12K → Try-on flow for one garment
Dress_16G → Try-on flow for another garment

Even keywords sent through Story replies work the same way and can trigger try-on flows.

Because of this, each keyword must be unique when you use Instagram ads (Keyword in DMs).

Post Comment Trigger (Organic Strategy)

(video example)

Now let’s look at the second AI Virtual Try-On automation strategy.

In most cases, this strategy works best when an Instagram shop or influencer has more than 10K followers.

With this approach, your Instagram grid becomes a showcase of garments that users can try on.

Users comment under a post, and this comment triggers a try-on flow in DMs.

Important difference from Instagram Ads:

In this strategy, you can use one simple keyword (for example, “tryon”) across all posts.

Why?

Because Apiway links a specific try-on flow to a post, not to the keyword.

Keyword logic in Comment Strategy

The keyword is only a trigger
The post itself defines which try-on flow will run
That’s why the keyword can be the same for many posts

For example:

Post A → Try-on flow for Dress A
Post B → Try-on flow for Dress B
Keyword used in both posts: “tryon”

You can use any keyword or language you prefer.

“tryon”
“примерка”
“probar”
“測試”

The comment keyword simply serves as a trigger for the flow.

Connect Your Instagram Account to Apiway

Before you can create automations and launch AI Virtual Try-On, you need to connect your Instagram account to Apiway.

This is a one-time setup.

Important: Apiway never asks for your Instagram login or password.
The connection is done securely through the official Instagram / Meta API.
You authorize access directly on Instagram’s side, and you can revoke it at any time.

After you sign up on apiway.ai and log in, you will be redirected to your Apiway dashboard.

In the left menu, go to Automations - Connected accounts.
You will see a list of available apps.

Find and click: Instagram (DMs Auto-reply)

If the account is not connected yet, you will see a Connect button.
Click Connect.

You will be redirected to Instagram. Instagram will show a permission screen asking if you want to allow Apiway to access your account.

Click: Allow.

After that, you will be redirected back to Apiway.

You should see: Instagram (DMs Auto-reply) — already connected.

Now your Instagram account is successfully linked to Apiway.

Step-by-step customization of
AI Virtual Try-On in Instagram DMs

Let’s open Apiway and build automation step by step.

Create a New Automation (Way)

On the main Apiway dashboard, click “Create a Way” in the left menu.

A Way is an automation scenario that:

Receives users from Instagram
Launches the AI Try-On flow
Sends messages and processes images automatically

After clicking, you will enter the Way editor.

Setup a trigger block:

In Trigger app, select: Instagram (DMs Auto-reply)
In Trigger event, select: AI Virtual Try-On (Clicking Instagram Ads / Stories)

What actually happens:
When a user clicks your ad or story, Instagram opens a DM and automatically sends a predefined keyword to your account. This keyword is what Apiway listens to and uses to start the automation.

So technically:

The user clicks the ad or story
Instagram opens Direct Messages
A keyword is sent to your DM
Apiway receives this keyword and launches the AI Try-On flow

In Additional features → Keyword, enter a unique identifier, for example:
Dress_12j

This is an internal product ID.
It allows you to:

Create many try-on flows
Each one for a different product

Setup an action block:

In Action app, select: Instagram (DMs Auto-reply)
In Action event, select: AI Virtual Try-On

And click: Update available fields. This will load all AI Try-On settings.

In the field: Garment Image for AI processing upload the product image:

Dress / jacket / suit / shoes
Preferably on a clean background

This is the image that will be placed on the user.

In the field: Comment text to send enter the message that will be sent under the post comment, for example:
👉 Check your DMs — we just sent you a message!

This tells the user to go check their Direct Messages.

In Welcome message image you can upload:

A product banner
A lifestyle image
Or a product preview

In Offer a Try On Image upload an image that:

Shows the product
Or shows a try-on example
Or visually explains what will happen

This is the main conversion screen before photo upload.

In the Prompt field, you will see that Apiway already provides a pre-filled default prompt.

We have prepared this prompt for you in advance, and all other important fields in this flow are also pre-configured by default to make the setup as fast and easy as possible.

This default prompt:

Is tested by our team
Works well for most fashion and e-commerce use cases
Preserves the person’s identity, pose, lighting, and scene
Replaces only the clothing with the selected garment

In most cases, you don’t need to change anything here — you can use it as-is.

However, if you want, you can:

Modify the prompt
Adapt it to your specific product or use case
Experiment with different instructions and styles

The prompt controls how the AI behaves, so advanced users can fine-tune it, but for 90% of cases, the default version works perfectly out of the box.

After all, in the top-right corner:

Click Save & exit
Turn the toggle ON
Click Test to check the flow

⚠️ While the automation is ON, you cannot edit it. You must turn it OFF first.

If you want to launch this AI Virtual Try-On flow using Instagram Ads, there is one important thing you need to do in Ads Manager.

When creating a new campaign in Ads Manager, you must select:
Campaign objective: Engagement

This is required because this flow works through messages and interactions, not through website clicks or conversions.

After you select the Engagement campaign objective and go to the Ad level, scroll down to the Chat builder section.

This is where you configure what the user will see inside Instagram right after they click your ad — and this is exactly where you set up the automatic sending of the keyword that will trigger the Apiway automation.

When you create your ad, you will need to set up the message template.

In the message text, you can write something like:
Do you want to virtually try on Dress_12j?

Then, in the Questions / Quick Replies section, you must add a question like:
Yes, I want to do an AI virtual try-on of Dress_12j.

⚠️ This is very important:
The keyword you use in your ad message (Dress_12j in this example) must be exactly the same as the keyword you specified in Apiway when setting up the automation.

In Apiway:
Keyword = Dress_12j

In Instagram Ad:
The quick reply / question must contain the same keyword: Dress_12j

Influencers collabs

One of the most popular use cases is collaborations where influencers offer AI Virtual Try-On to their followers inside DMs.

People already trust these creators and want to try on the garments that the creator recommends.

For creators, this is a unique use case that drives engagement.
AI Try-On is fun. People generate images, then send them to friends saying:
“Look what this creator is offering. You should try it too.”

Now, it’s important to understand one technical detail about how Instagram collaborations work.

In every Instagram collaboration, there is an author of the post and a collaborator. The automation always works from the author’s account, not from the collaborator’s account.

Because of this, the correct setup looks like this: the influencer creates the content, sends it to the brand, and the brand publishes the Reel on its own account and invites the influencer as a collaborator. The post appears in both feeds, but technically the brand remains the author.

This is important because the brand already has Apiway connected and the AI Try-On automation configured. The influencer does not need to connect any software, set up automations, or touch any technical systems at all. They just create content and approve the collaboration.

When a user interacts with such a post, the DM opens with the brand’s account, and Apiway launches the AI Virtual Try-On flow from the brand’s side.

If the influencer publishes the post on their own account and only tags the brand, the situation is reversed: the influencer becomes the author, and then the automation would have to be set up on the influencer’s account.

That’s why the right business model is simple: the influencer creates content and brings attention, and the brand owns the funnel, the automation, and the conversion.

This is also why paid influencer collaborations are the perfect distribution channel for AI Virtual Try-On. When brands invest in influencer posts, they care about conversion and are willing to set up a proper automated funnel.

About Nano Banana (v1 and v2)

At the moment, Apiway uses Nano Banana 1 — an image generation model from Google, available via Gemini, which we use for virtual try-on. In simple terms, this AI model transfers clothes onto a person, preserving the face, pose, and body proportions, and renders the final image to look like a real photograph.

Nano Banana 1 is the first version of this model. It works fairly fast and is relatively inexpensive, but sometimes it can produce strange results. In some cases, details of the clothing, the logic of layers, or the shape of the silhouette may be distorted. This is not an Apiway bug — it is a limitation of the current generation of AI models.

Google has already released Nano Banana 2, the next and more advanced version of the same model. It understands garment construction better, works more accurately with fabrics, keeps body proportions more consistently, and overall produces more stable and photorealistic results. This version is noticeably more expensive to run, but the quality is significantly higher. In the near future, we plan to switch to it as the main engine for virtual try-on.

It is important to understand that there is no such thing as a perfect AI try-on yet. Occasionally, artifacts and strange results can happen — this is a normal stage in the development of the technology.

Below, we show a real example using the same model and the same outfit. In the result generated by Nano Banana 1, the model “decided” to put shorts on top of the pants. It looks strange, but this kind of glitch clearly shows the limitations of the first version. In the same scenario, Nano Banana 2 produces a much more correct and realistic result.

This example clearly shows why we are moving to a newer model and why the quality of AI try-on keeps improving over time.

Garment Photo Recommendations
for AI Processing

The garment image you send to AI is the most important part of the try-on flow that you can control.
(The second most important part is the user’s photo — but you can’t fully control that).

If you provide a bad or unclear garment image, the result will also be bad.

This step is critical.

From our observations, the images that brands already use for Instagram posts and ads often work well.

AI can only work with what it clearly sees.

When AI does not see enough information, it starts to guess — and that’s when problems appear.

Example: unclear garment length.

In the example below, it is clear that this is a dress.

However, it is not obvious how long the dress really is.

As a result, in our case, the AI generated a shorter version of the dress than in the original product.

And in general:

For one user, AI may generate a long dress
For another user, AI may generate a shorter version

This happens because the AI does not have enough visual information and starts guessing and completing missing details.

Key rule:
The clearer and more understandable the garment photo, the better the try-on result.

Advanced poses are bad for AI
(even if they look great on Instagram)

Instagram shops often post creative, dynamic poses — and they look amazing for marketing.

But for AI processing, these images are not ideal.

For example:

Is the model wearing shorts or a mini skirt?
Does the jacket have front pockets?
What does the front view of the garment look like?

If these details are hidden, AI will guess.

Side-view garments vs front-view selfies

In some photos, AI can clearly identify only the side view of the garment.

But users usually send front-facing mirror selfies.

In this situation, AI has no clear front reference and will invent missing details, creating a garment that may not match the real product.

Best practices for garment photos

For best results, use garment images that:

Show a clear front view
Clearly show length and shape
Have good lighting
Do not hide important details with poses or accessories
Are easy to visually understand in one glance

Ghost Mannequin (Best Practice)

The best results are usually achieved with ghost mannequin garment images, like in the example below.

A ghost mannequin does not guarantee perfect results, but it significantly increases the chances of getting a high-quality AI try-on image.

Why ghost mannequins work best

Ghost mannequin images:

Show the exact shape of the garment
Clearly display front structure and length
Remove confusion caused by poses, body angles, or accessories
Give AI a clean and complete view of the product

Because of this, AI has less room to guess and produces more stable results.

How to create a ghost mannequin AI image

You can create a ghost mannequin image using Gemini Thinking Model (Nano Banana 2).

Below is a prompt example you can use:

Convert the scene into a clean, bright white photography studio. The background must be pure white (RGB 255, 255, 255), seamless, and minimal.
Keep the human in the exact same position, pose, camera angle, framing, and proportions in the frame.
Replace the human with a ghost mannequin — an invisible body form with no head, neck, arms, or legs visible. Only the natural garment shape should remain, as if worn by an invisible figure.
Important:
Use all garments from the original photo. Every clothing item must be preserved exactly as it appears.
Preserve all colors, textures, materials, stitching, folds, patterns, and garment design details exactly as in the original. No changes.
Copy the original lighting, color tones, reflections, and shadows from the source image. Keep the same light direction and intensity.
Maintain the same silhouette and clothing fit exactly as it appears on the human.
Do not stylize, beautify, enhance, or alter the clothing in any way.
The final output must look like a professional ghost-mannequin product photo on a pure white (RGB 255, 255, 255) background

Common AI Problems You Will Face — and Why They’re Normal

AI Virtual Try-On is powerful, but it is not perfect yet.
You will face issues — and that is expected.

This already happened in history:

In the 1920s, starting a car was difficult
Early household appliances were unreliable
In the early 2000s, you couldn’t watch videos online because internet was too slow

Today, we can’t live without these technologies.

The first innovators who used the internet in the 1990s built massive businesses because they saw opportunity while others said:
“It’s just a toy for kids.”

With AI, we are at the same stage now.

AI models are improving fast, but today they are still unstable and can hallucinate.

You will face this 100%.

AI Hallucinations (AI glitches)

AI models hallucinate.
You will face this 100%.

Sometimes:

You generate an image and it looks wrong
You resend the same image
And the next result looks perfect

That’s not a bug.
That’s how AI works today.

How often does AI hallucinate?

From our experience:

Nano Banana 1.0 → around 1 time out of 10
Nano Banana 2.0 → around 1 time out of 100

With every model update, this happens less and less.

I believe that somewhere around 2027–2029 this problem will almost disappear.
But in 2026, you will still face it.

AI Can Slightly Change the Garment

In one photo, your garment can look perfect.

In another photo, the same garment can look slightly different.

Important notes:

This does not always depend on garment complexity
Sometimes a complex dress works perfectly
Sometimes a simple white T-shirt fails

AI behavior is not linear.

The good news:
With every major model update, garment accuracy becomes better and more stable.

Not a Perfect Fit (AI Is Not 3D)

AI does not understand real sizing (S, M, L, measurements).

What AI does:

It visually overlays the garment
It edits the image like Photoshop, not like 3D modeling

AI May Reject User Photos

Sometimes, AI rejects user photos in Instagram DMs and asks for another one.

Important:

The same photo may be rejected once and accepted the next time
A photo that worked yesterday may fail today

Common reasons:

AI thinks the image looks like a celebrity photo
Fashion-style photos look “too professional”
Too much skin or transparent underwear
Low-quality lighting or blur

Example:
A normal photo may be rejected because AI thinks the person is a public figure (for example, a tennis player or influencer).
This is frustrating, but normal.

AI May Reject Your Garment Image

Sometimes AI simply does not accept your garment image.

This happens mostly with:

transparent fabrics
lace
mesh
very thin materials

It happens because AI models have nudity protection.

AI didn’t accept this dress:

AI Can Put Garments Over Pants
(Layering Issue)

From my experience, Nano Banana 1.0 can put shirts over pants.

For example, you are selling shirts and a user uploads a photo wearing pants.
AI should remove the pants, put the shirt on the body, and show bare legs.
But sometimes AI wears the shirt over the pants.

At the same time, when you are selling jackets, AI should put the jacket over a T-shirt.
AI can’t clearly understand in which case the previous garment should be removed, and in which case it should be added on top.

With the same garment, but a different user photo, it can work perfectly.

When we were testing Nano Banana 2.0, this problem was almost gone.
That is why Nano Banana 2.0 is more expensive than Nano Banana 1.0.

AI Limitations with Tight Long Dresses

For example, you are selling tight long dresses, but the user sends a photo of herself wearing jeans with her legs spread wide.
In this case, AI has to bring the legs together.

From our experience:

Nano Banana 1.0 makes mistakes
Nano Banana 2.0 works better

Sometimes it works.
Sometimes it doesn’t. 😂🤷‍♂️

An iconic Silicon Valley scene that explains AI perfectly:

AI is unpredictable.
You need to run some tests to start “feeling” the AI and understand how it works.

There is no template or guide that works 100% of the time.
People who say that AI works 100% are lying.

What Works Well vs What Doesn’t

From our observation, the best categories for AI Virtual Try-On are:

Works well:

Top body clothing
Outerwear
Suits
Shirts and T-shirts
Hats and glasses
Bags and other accessories
Swimsuits

Works badly:

Lower body clothing
Slim (tight) dresses
Underwear
Shorts

Users Will Send You Ugly Photos

Users will send you ugly photos.
It is a big problem, and it will take time before people start to understand what works for AI and what doesn’t.

For our luck, big brands like Zara have already launched AI Virtual Try-On in their apps.
This will push demand and help educate the audience on how AI try-on works.

Good photos from users give around 60% success rate.
If a user sends a good photo and your garment is not complicated, AI can generate a good result in 80–90% of cases.

If the photo is not good, AI will fail.

Why did we recommend using
mirror selfies in the DM flow?

Actually, it is not strictly necessary.
AI can generate images even if the person is sitting on a chair.

But what we see in real life:

In 90% of cases, users send mirror selfies
Even if you ask them to send studio photos or other types of photos

Mirror selfies are the type of photo almost every woman has.
That is why we built our strategy around mirror selfies.

How People React to AI Try-Ons

I have seen hundreds of reactions when people try AI Virtual Try-On.

60% of people have a WOW effect.

They send AI images to their friends and start talking about it.

20% of people have no strong reaction.

Maybe a little interest, but nothing more.

15% of people think it does not really work.

They think logically and say:
“The garment on me will look different.”
“How will AI figure out my size?”
They think from a rational point of view.

5% of people have negative reactions.

Some say:
“It’s not me.”
“AI made me fatter or thinner.”

Virtual Try-On simulates the process of looking at yourself in the mirror.
And let’s be honest — many people are not confident and don’t fully accept themselves when they look in the mirror.

If you are an ads expert or SMM manager and you set up Apiway for your clients, you will face the same reactions.

Some brand owners will be surprised and say:
“Let’s do it.”

Some will be skeptical.

Some will say:
“Let’s try it, but if AI generates three arms, I can’t show this to my followers. The technology is not ready.”

From my observation, in most cases people see AI Try-On as something fun.
But 100% you will face people who think it is evil or believe it will never work properly.

You just need to be ready for that.

How to Increase the Number of Try-Ons

Look how our clients make the call to action at the end:

When you use the Comment-to-Post strategy, users receive a DM from you.
The problem is that Meta sends these DMs to Hidden Requests by default.

Apiway can’t control this.
The page owner has to teach users to check Hidden Requests.

To increase the chance that users actually start the try-on process, we recommend adding a clear call to action as the last image in your carousel, like this:

Users scroll through 3–4 images in your carousel, and at the end they see a direct instruction to leave a comment and start the try-on.

Important note:
If you only mention the try-on opportunity in the post description, it usually works very poorly. Most people do not read captions.
If you want to increase the number of try-ons, you need to push this clearly to your audience. This is a new use case, and people don’t understand what’s happening by default.

Reels work the same way.

That’s why we recommend:

talking about the try-on feature directly in Reels
adding clear visual notes or text explaining what to do
Make direct “call to action”

Prompt Playing

There are two prompt strategies that you can use in Apiway:

Overlay on the original user selfie
Generating a new image based on the user selfie

Image 3 was generated from scratch using the following prompt:

Use image #1 as the base for the face, hairstyle, and body shape. High-fidelity reproduction of the face from image #1. Preserve all facial features and subtle expression precisely. Full-body mirror selfie of a person taking a photo with their phone. Keep the person’s face, head, hairstyle, body shape, and proportions from image #1.
**Crucially, retain the exact skin tone, natural pores, and subtle skin imperfections from image #1. Do not enhance, smooth, or apply any beauty filters to the face; maintain its authentic texture and appearance.** Replace the clothing with the outfit from image #2, keeping the exact fabric, texture, patterns, color, and cut.
One arm holds the phone.
The pose is confident and stylish: relaxed shoulders, weight shifted casually to one leg, hips slightly angled.
Crucially, retain the exact facial angle and head position as seen in image #1.
The person should fill about 90% of the frame. Scene: A luxury modern BoConcept-style living room with wooden plank flooring, a designer sofa, wall art, refined shelving, and subtle upscale textures.
Cool daylight from one side, warm lamp light from the other. A large thin black-framed mirror reflects the interior.
Apply a **high-fashion, polished Instagram look, resembling a Vogue editorial selfie** — **crisp details, vibrant yet balanced colors, sophisticated lighting, and a refined aesthetic.** Use a 9:16 vertical cinematic crop. The photo should look like a polished Instagram mirror selfie. Add a small watermark in the bottom-right corner: Text: "Created by Apiway.ai" Proportions: The watermark must be small — no larger than 5–7% of the image’s width. Keep it clean, minimal, and unobtrusive.

By default, all automations include a prebuilt prompt that works with the Overlay strategy.

Below, I explain the difference and the pros and cons.

Overlay Strategy

The overlay strategy means we take the user’s photo and tell AI to wear the garment on this exact image.

AI works only with the garment area and does not change:

background
user’s face

With this strategy, the user always sees himself, with only very small changes in rare cases.

However, in most cases user photos are not perfect, and the garment can look different.
In bad photographs, things don’t fit well. In real life it’s the same — fashion photoshoots always look better than mirror selfies.

Pros:

User always sees himself

Cons:

Garment fit may not look perfect
Small visual changes can appear

New Image Generation Strategy

This strategy works differently.

The user uploads a photo, and we tell AI:
“Based on the user’s face and photo, dress the user in this garment and place them in a new environment (for example, a Star Wars ship).”

AI takes the face and creates a new person using it.

As a result, you get a new character that may look similar to the original person.

Pros:

Garment looks perfect

Cons:

User’s face and body structure can look completely different

Which AI generative strategy to use

We use the first strategy by default, but you can change the prompt and customize it however you want.

From our experience:

The second strategy works much better for men
If you sell men’s clothing, you can use the second strategy
If you sell women’s clothing, it is better to use the first strategy

In the future, we believe AI will be able to create avatars that look 100% like real people.
Right now, the result is not at that level.

My suggestion: experiment with different prompts.
It’s hard to explain — this needs to be felt.

If you don’t want to spend time on this, just use the default prompt.
It works.

Quick Start and Test AI Virtual Try-On

In most cases, our customers set up AI Try-On for old posts in their feed first.
They test it themselves and ask friends and employees to try it.

After that, they simply repost this content in Stories and get their first try-ons.
When everything works well, they start to customize try-ons for each new post.

When you test Try-On on your own posts, you should comment from a different account.
For example, use your personal account.

If you comment on your own post from the brand account, automation and AI Virtual Try-On will not work.