Virtual dressing room

The Virtual dressing room mode creates a photoreal virtual try-on from separate subject and garment images.

This is the right workflow when you want to see how a dress, jacket, shirt, or full outfit looks on a real person without doing a simple manual collage. The goal is not to “paste” a garment, but to produce believable fit, fabric behavior, folds, lighting, and contact points.

When the garment or accessory reference contains unwanted elements, you can use the mask to isolate only the part that should actually guide the try-on.

Use Virtual dressing room if:

• you want to apply a garment to a person while keeping the subject recognizable;
• you want to create outfit previews, lookbooks, fashion images, or professional profile photos;
• you want to preserve face, pose, and framing while changing the outfit;
• you want to guide accessories such as bags, glasses, hats, or jewelry with an optional third image.

• If you only need local retouching or a small fix on the same photo: use Single image edit.
• If you want to merge multiple images into a general scene and you are not transferring clothing: use Compose multiple images.
• If you want manual placement of extracted subjects inside a scene: use Seamless Blend (subjects).

• Image #1: subject photo to dress (required).
• Image #2: garment reference photo (required).
• Image #3: accessory reference (optional).
• Supported total inputs: 2 or 3 images.
• The mask is available for garment and accessory references, not for the subject photo.
• No free-form prompt in the current interface.
• Optional structured builder in advanced controls: use it to guide presentation, styling, accessories, preserve rules, and failure modes to avoid.

When processing starts, the app interprets the inputs in a strict order:

• the first image as the subject photo, trying to preserve identity, face, hair, skin tone, body proportions, pose, and framing;
• the second image as the garment reference, trying to preserve garment category, silhouette, colors, pattern, trims, seams, and fabric behavior;
• the third image, if present, as an accessory reference to use only when it integrates naturally.

The model then tries to:

• dress the subject in a realistic way;
• keep the garment recognizable;
• align light, shadows, drape, and contact points;
• avoid duplicated garments/accessories and unrequested face or body changes.

In this mode you typically see:

• Subject photo: required upload.
• Garment photo: required upload.
• Accessory reference: toggle that enables an optional third image.
• Mask item / Re-edit mask: opens the mask editor so you can isolate only the garment or accessory you want to use.
• Use full photo: cancels the extraction and returns to the full reference image.
• Show advanced: opens the extra controls for this mode.
• Virtual try-on builder: structured controls for presentation, styling priorities, accessories, preserve rules, and failure modes to avoid.
• Start try-on: launches generation.
• Generate variants (2-4) and Use as input: compare results and refine.

1. Choose Virtual dressing room in the mode selector.
2. Upload the subject photo.
3. Upload the garment photo.
4. If the garment photo contains extra elements, use the mask to isolate only the relevant garment.
5. If needed, enable and upload an accessory photo.
6. If the accessory photo includes a hand, person, packaging, or other irrelevant context, use the mask there too.
7. If you need constraints or preferences, open Show advanced and use the Virtual try-on builder.
8. Start the try-on.
9. Generate 2-4 variants and pick the most believable one.
10. If small issues remain, use Use as Input and clean them up in Edit.

In the “Mode” selector, choose Virtual dressing room.

Choose the person who should wear the outfit. This usually works better with:

• a clearly visible subject;
• a readable pose;
• a clear face if you want stronger identity preservation.

Upload the dress, jacket, shirt, or full outfit you want to transfer. The clearer the garment is, the easier it is to preserve:

• silhouette;
• texture;
• pattern;
• seams and construction details.

The mask tells the app which part of the reference image should actually matter.

Use it especially when:

• the garment photo contains a second subject or another person;
• the reference person is wearing multiple garments, but you only want one of them;
• you want only the shirt and not the pants, skirt, jacket, or shoes too;
• the accessory photo also includes a hand, box, table, mannequin, bust, or other elements you do not want to carry into the result.

Practical example:

• the reference image shows a person wearing a shirt and pants, but you only want the shirt;
• in that case, open the garment mask and paint only the shirt;
• this keeps the try-on focused on the right item instead of the whole look.

In general, the mask is useful whenever you want to reduce ambiguity. The cleaner the reference, the easier it is to get a coherent result.

Enable the accessory reference only if you want to guide elements such as:

• bag;
• hat;
• glasses;
• jewelry;
• styling extras.

The accessory is not required and may still be ignored if it does not integrate well.

If the accessory photo is already clean, you can leave it as-is.

If it contains too much context, open the mask and isolate only the useful item. This is especially helpful for:

• bags photographed in someone’s hand;
• glasses worn by a person;
• jewelry shown inside a wider scene;
• accessories photographed together with other products.

In the current interface there is no free-form prompt field for Virtual dressing room.

If the three inputs are enough, you can start the try-on directly.

If you need more control, open Show advanced and use the Virtual try-on builder. Its selections are converted internally into structured instructions for rendering.

The builder lets you:

• choose a presentation style (professional portrait, lookbook, outfit preview, etc.);
• define fit, silhouette, and material priorities;
• request compatible accessories;
• say what to preserve (face, hair, background, hands, proportions);
• state what to avoid (changed face, warped fabric, floating accessories, overly synthetic look).

Click Start try-on and evaluate:

• garment fidelity;
• natural fit;
• coherent light and shadows;
• preservation of face and pose.

Use Generate variants (2-4) to compare:

• fit;
• detail rendering;
• accessory integration;
• cleanliness of hands, fingers, and edges.

If one variant is almost right but still has small issues, move to Single image edit with Use as input.

Open the builder when you want:

• a tighter or more relaxed fit;
• stronger fidelity to texture, buttons, seams, and pattern;
• explicit preservation of face, hair, background, or proportions;
• constraints against common problems such as duplicated garments or misaligned accessories.

Practical setup:

• Styling priorities: fit, silhouette, fabric texture;
• Preserve: face, pose, background, or proportions when they matter;
• Avoid: duplicated garments, distorted hands, synthetic look;
• Accessories: only when they materially improve the result.

• Paint only the garment or accessory you really want to use.
• Try to cover the full visible shape of the item, not just a small patch.
• Do not include body parts, face, hands, background, or other garments unless they should influence the result.
• If the reference is already clean, do not mask it just for the sake of masking: the full photo may work perfectly well.
• If the extraction does not look right, use Re-edit mask to refine it or Use full photo to go back to the complete reference.

• Virtual dressing room is not a manual garment warp tool: the final result is a new AI composition.
• If you want to keep the original background too, set it in the builder under preserve rules.
• Accessories are optional, and the model may ignore them when they do not fit the scene or pose naturally.
• If the garment reference is hard to read, partially hidden, or too compressed, fidelity can drop.

• in the builder, set identity, face, and hair as preserve rules;
• use a sharp, readable subject photo.

• in the builder, prioritize texture, pattern, seams, and drape;
• use a clearer garment image.

• enable the third image only when truly needed;
• add accessories only when they materially help, then compare variants.

• in the builder, add duplicated garments to the avoid list;
• generate variants and keep the cleanest integration.

• use the mask on the garment photo and isolate only the correct item;
• if the reference shows a shirt plus pants, paint only the shirt;
• avoid crowded references when you can replace them with a cleaner image.

• use the mask on the accessory reference;
• isolate only the bag, glasses, hat, or jewelry;
• if the original reference was already fairly clean, also test Use full photo and compare variants.