P-Image-Try-On

P-Image-Try-On is Pruna’s virtual garment try-on model. Provide a photo of a person and up to eleven garment reference images — one per supported category — and the model realistically dresses the person while preserving pose, face, and background. Garment refs can be flat lays, product shots, or lifestyle photos with multiple items; optionally pass reference_pose to match a target pose before dressing.

Use it for catalog PDP imagery, virtual fitting rooms, B2B line sheets, lookbook campaigns, UGC ad variations, and outfit recommendation carousels. Generate base model and garment reference stills with P-Image, then run try-on. For pose variants on the dressed output, use P-Image-Edit as a separate post-step.

Tip

API reference: p-image-try-on on Pruna API.

At a glance

Metric	P-Image-Try-On
Price	Standard: $0.015 for the first garment, $0.008 for each additional. Turbo: $0.008 per garment.
Speed	Standard: < 2 seconds per garment. Turbo: < 4 seconds total (`turbo: true`, off by default).
Output	Match input dimensions (`preserve_input_size: true` by default)
Max garments per request	11 (one per category)
Character consistency	Very high — face, identity, and likeness preserved across outputs
API model header	`Model: p-image-try-on`

Typical workflow

Step	Model	Role
Base model + garment refs	P-Image	Generate the person photo and isolated garment stills (optional if you already have catalog assets)
Virtual try-on	P-Image-Try-On (this page)	Dress the person in `garment_images` while preserving face and background; pass `reference_pose` to match a different pose before dressing
Pose variants (optional)	P-Image-Edit	Re-pose the dressed output without regenerating garments or the base identity

Turbo mode

Turbo mode applies multiple garment edits in one larger edit instead of processing them separately. This makes it faster than standard mode, especially when dressing more than one garment.

Topic	Details
Default	Disabled — set `turbo: true` to enable
Speed	Usually < 4 seconds total, regardless of garment count
Pricing	$0.008 per garment (flat rate in turbo mode)
Compatibility	Works with all other model features and the same inputs — switch modes without changing your pipeline
Quality trade-off	May produce slightly lower quality than standard mode; some garments may not be added correctly
Multi-garment limit	Performance tends to decrease when editing more than 4–5 garments at once

Use turbo for high-throughput previews, fitting-room flows, and outfit carousels where speed matters most. Use standard mode when quality is the priority. Actual latency and pricing may vary.

Supported garment categories

You can pass up to 11 garment reference images per request — one item per category. The model auto-classifies each image, so you do not need to label categories in the API payload. Up to 6 garments is recommended for consistent quality.

Garment compatibility at a glance

Category	Reliability	Notes
`tops`	High	Core apparel — t-shirts, shirts, blouses, polos, tanks. Best with flat-lays and studio product shots.
`top-layers`	High	Hoodies, sweaters, cardigans, vests, blazers. Layer over `tops`; stacking order is not configurable.
`outerwear`	High	Coats, jackets, parkas, fleeces. Flat-lays and studio product shots work best.
`dresses`	High	Dresses, jumpsuits, rompers, robes. Replaces separate top + bottom — do not combine with `tops` + `bottoms`.
`bottoms`	High	Skirts, shorts, trousers, jeans, chinos, leggings, belts. Pair with `tops` or a `dresses` item.
`underwear`	Moderate	Underwear, bras, swimsuits, bikinis. Supported but less common in catalog workflows.
`feet`	Variable	Sneakers, boots, heels, sandals, socks. Least reliable category — person photo must show feet and ankles.
`headwear`	Moderate	Hats, caps, beanies, face masks. Isolated refs on neutral backgrounds work best.
`neckwear`	Moderate	Scarves, ties, necklaces, chokers. Keep refs clean and centered.
`bags`	Moderate	Handbags, totes, backpacks, crossbody bags. Single bag per request; placement can vary.
`wristwear-single`	Moderate	One watch or bracelet — not stacked jewelry.
Ring / earrings	Moderate	Auto-classified as single-item handwear or earwear; one item per type (not gloves or mittens).

Not supported

Anything not listed above is not supported, including:

Category	Examples
Eyewear	Sunglasses, eyeglasses — use p-try-on-glasses instead
Handheld accessories	Wallets, card holders, keychains, umbrellas, phones, cups, handkerchiefs
Handwear (gloves/mittens)	Full-hand coverage items — not single rings
Multi-item wristwear	Arm warmers, stacked bracelets
General accessories	Pocket squares, suspenders, brooches
Duplicate categories	Two tops, two bags, etc. in the same request
Non-flatlay refs without `prompt`	Lifestyle or multi-garment photos need the experimental `prompt` parameter to select which item to use

Strengths

Capability	Details
Multi-garment outfits	Dress up to 11 categories in a single pass — tops, bottoms, outerwear, footwear, accessories, and more
Character consistency	Very high quality preservation of face, identity, and likeness across outputs
Localized editing	By default, everything outside the garment regions stays intact — background, pose, and skin texture
Garment auto-classification	No manual category labels; the model infers body region from each reference image
Flexible reference formats	Flat lays, studio product shots, worn-on-form photos, and non-flatlay images with multiple garments — use `prompt` to select which garment to use from each image
Reference pose	Pass `reference_pose` to re-pose the person to match a target pose before try-on
Turbo mode	Optional < 4 s total runs for multi-garment outfits at $0.008 per garment — off by default
Cost and throughput	$0.015 for the first garment, $0.008 for each additional garment — built for high-volume catalog and creative workflows

Limits

Limit	Impact
11 items max, one per category	Duplicate categories in `garment_images` are not supported
Footwear variability	`feet` produces the least consistent results — expect occasional shape or placement drift
Accessory placement	`bags`, `headwear`, and `neckwear` can land off-position with busy or worn-on-body refs
Framing	Non-full-body inputs may artifact — best results with full-body or 3/4 shots
Color and texture drift	Complex prints and saturated colors may shift slightly in output
Layering order	Cannot explicitly specify stacking (e.g. blouse under dress vs. over)
Turbo quality	`turbo: true` may miss garments or reduce fidelity — prefer standard mode for hero PDP shots
Turbo at scale	Editing more than 4–5 garments in turbo mode can reduce reliability

Person image guidelines

Try-on preserves the person’s pose, face, and background from the input photo by default. To match a different pose, pass reference_pose with a photo of a person in the target pose — the model re-poses before dressing. For pose variants after try-on, use P-Image-Edit as a separate post-step.

Reference pose

reference_pose accepts an image URL of a person in the pose you want. The model adapts the generated person’s pose to match the reference before applying garments. Use this when your person_image pose differs from the look you need — for example, converting a standing catalog photo into a walking lifestyle shot without regenerating the identity.

Framing and quality

	Do	Avoid
Framing	Full-body or 3/4 length — head to at least the knees; head-to-toe for pants, dresses, and footwear	Chest-up or head-only crops — common source of hemline, layering, and footwear artifacts
Subject	Single person, centered, torso and limbs unobstructed where garments will land	Heavy occlusion (crossed arms over torso, large props, other people in frame)
Identity	Face visible; frontal or three-quarter angles for character consistency	Back-only or extreme side profile when face identity must stay consistent
Baseline outfit	Simple fitted tank/tee + leggings or jeans that can be replaced cleanly	Busy patterns or bulky layers that fight new garments
Footwear try-on	Feet and ankles in frame when passing `feet` category refs	Cropped feet when trying on shoes
Image quality	Clear, in-focus photos with even lighting; phone-camera UGC is fine when sharp	Motion blur, heavy noise, or extreme exposure

Supported poses

The model is not limited to catalog standing poses. These input poses work well:

Pose type	Notes
Standing, facing camera	Arms at sides; best default for B2B and fitting-room flows
Three-quarter turn	Over-shoulder or slight angle; face still visible
Walking mid-stride	Natural arm swing; works for e-commerce and lifestyle reco
Mirror selfie	Phone at chest height; good for VFR and UGC — try-on keeps the selfie pose
Dynamic street motion	Mid-step / stepping off curb; UGC ads
Seated	Works when feet stay visible (e.g. cafe terrace); avoid poses that hide legs under tables
Environmental lean	Rooftop ledge, balustrade; keep the full silhouette in frame

Not every pose is ideal for every garment. Mirror selfies and dynamic motion work well for try-on, but pose edits afterward can artifact on hands (phones, pockets). Prefer distinct pose prompts without re-introducing a phone when generating variants.

Orientations and angles

Aspect ratio	Typical use	Best for
Portrait 3:4	Default catalog framing	E-commerce, B2B, outfit recommendations
Portrait 9:16	UGC, lookbook, mobile fitting-room selfies	Virtual fitting room, lookbook, UGC ads
Landscape 16:9	Wide fitting-room mirror shots	In-store mirror try-on; use `preserve_input_size: true`

Camera angle should keep the garment regions you are dressing visible: frontal and three-quarter angles work best; steep overhead or worm’s-eye angles are untested and may drift.

Garment reference tips

Topic	Recommendation
Count and categories	Up to 11 refs, one per category; auto-classified — no manual labels
Reference formats	Flat-lays, studio product shots, torso-form refs, and lifestyle photos with multiple garments
Multi-garment images	One photo can contain several items — use `prompt` to specify which garment to use (e.g. `the green t-shirt from image 1 and the trousers from image 2`)
`prompt` (experimental)	Required for non-flatlay or multi-item refs; refer to images as `image 1`, `image 2`, etc.
Footwear (`feet`)	Clear shoe product or flat-lay ref; person photo must show feet; least reliable category
Small accessories	Scarves, hats, jewelry — isolated refs on neutral backgrounds
Unsupported items	Gloves, handheld accessories, pocket squares, and other types listed under Not supported above
Coherent outfits	Gender-matched, non-clashing combinations (e.g. avoid dress + separate pants)

Domain Use Cases

Six common production workflows, with examples in each tab below.

Tab	Who it’s for	Typical workflow
E-commerce	Marketplaces, fashion brands, POD sellers	Dress a house model in SKU refs (flat-lay, worn photo, white-bg product shot) for PDP imagery
Virtual fitting room	Retail apps, in-store mirror experiences	Consumer mirror, boutique, or phone-selfie poses; preserve pose and background
Wholesale & B2B	Wholesalers, line-sheet producers	One reusable house model across SKUs; consistent studio lighting and framing
Lookbook & campaigns	Brand marketing, seasonal campaigns	Editorial 3/4 poses; multi-garment outfit swaps on the same base model
UGC ads	Performance marketers, social ad teams	Bedroom, street, and testimonial framing; native backgrounds preserved
Outfit recommendations	Styling apps, recommendation engines	Same person photo, different multi-garment combos — carousel-style looks

feat_ecom_golden_hour_portrait garment 1

API Usage

Upload a person photo and one or more garment reference images, then call the p-image-try-on model.

Upload files

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: YOUR_API_KEY" \
  -F "content=@/path/to/person.jpg"

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: YOUR_API_KEY" \
  -F "content=@/path/to/garment.jpg"

Synchronous try-on

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H 'apikey: YOUR_API_KEY' \
  -H 'Model: p-image-try-on' \
  -H 'Try-Sync: true' \
  -d '{
    "input": {
      "person_image": "https://api.pruna.ai/v1/files/file-abc123",
      "garment_images": [
        "https://api.pruna.ai/v1/files/file-def456",
        "https://api.pruna.ai/v1/files/file-ghi789"
      ]
    }
  }'

Required parameters

Parameter	Type	Description
person_image	string	Image URL of the person to dress
garment_images	array	Image URLs of garment references (up to 11, one per category). Flat-lays, product shots, and multi-garment lifestyle photos are supported — use `prompt` for the latter.

Optional parameters

Parameter	Type	Default	Description
seed	integer	—	Random seed for reproducible outputs
turbo	boolean	false	Apply all garment edits in one pass for faster inference (< 4 s total). Not recommended for more than 4–5 garments.
prompt	string	“”	EXPERIMENTAL. For non-flatlay garment images, indicates which garment to use from each image (e.g. `the green t-shirt from image 1 and the trousers from image 2`).
reference_pose	string	—	Optional image URL of a person in the target pose. The model re-poses the person to match before try-on.
output_format	string	jpg	Output image format
output_quality	integer	95	JPEG output quality (1–100)
preserve_input_size	boolean	true	Keep output dimensions aligned with the person input

Example configuration

{
  "input": {
    "person_image": "https://api.pruna.ai/v1/files/file-abc123",
    "garment_images": [
      "https://api.pruna.ai/v1/files/file-def456",
      "https://api.pruna.ai/v1/files/file-ghi789"
    ],
    "reference_pose": "https://api.pruna.ai/v1/files/file-jkl012",
    "prompt": "the blazer from image 1 and the chinos from image 2",
    "turbo": true,
    "preserve_input_size": true,
    "output_format": "jpg",
    "output_quality": 95,
    "seed": 42
  }
}