p-image-try-on

Use when the user wants virtual try-on, dress a person in clothing from reference photos, garment fitting on a model still, or ecommerce fashion compositing.

Pruna AI@pruna-ai

Install

openclaw skills install @pruna-ai/p-image-try-on

p-image-try-on (Pruna)

Virtually fit one or more garments onto a person's photo. Rate limit: 500 requests/minute · Category: Image Editing.

The model's strength is garment-only editing — identity, pose, hair, background, and scene props stay intact. That supports editorial fashion, complex prints / patchwork, and multi-garment stacks, not just simple flat-lay tee swaps.

Canonical API reference: p-image-try-on model docs · operational guide (Runware host): virtual try-on

Showcase quality bar: realistic-persona-showcase.md · try-on checklist: p-image-try-on-quality-checklist.md · examples: example-prompt.md

Shared HTTP patterns: pruna-api.md (upload, poll, download)

Pricing

Per generation (same for normal and turbo mode):

$0.015 for the first garment
$0.008 for each additional garment

Example: 3 garments → $0.015 + 2 × $0.008 = $0.031.

Request shape

One person_image, one garment_images[] entry per piece (up to 11), optional reference_pose. The model auto-classifies each garment — array order does not matter and you do not tag hat vs shoe. No composite garment image required; mixed categories (headwear + top + shoes + bag) belong in one call.

prompt — only when a reference shows multiple garments or is worn on-model; clean flat-lays need no prompt.
preserve_input_size: true (default) — output dimensions follow the person image.

Runware field map (same model, different host): person → person_image, garment → garment_images[], pose → reference_pose, positivePrompt → prompt, settings.turbo → turbo.

HTTP (curl)

Follow the official quickstart: upload files, then call POST /v1/predictions with Model: p-image-try-on.

Start with uploading your images

bash

# Upload person photo
curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: ${PRUNA_API_KEY}" \
  -F "content=@/path/to/person.jpg"

# Upload garment image (repeat for each garment file)
curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: ${PRUNA_API_KEY}" \
  -F "content=@/path/to/garment.png"

Use -F (form) with @ to upload from disk. Use each response urls.get in input.person_image and input.garment_images[].

Optional uploads for extended fields: reference_pose (pose reference person image).

Try On (Synchronous)

bash

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

Try On (Asynchronous)

Omit Try-Sync for production reliability; poll until succeeded:

bash

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-try-on' \
  -d '{
    "input": {
      "person_image": "https://api.pruna.ai/v1/files/PERSON_FILE_ID",
      "garment_images": ["https://api.pruna.ai/v1/files/GARMENT_FILE_ID"]
    }
  }'

Poll and download: pruna-api.md.

Parameters

Tables follow the official model page. Extended fields (turbo, reference_pose, prompt) are listed below.

Required

Parameter	Type	Description
`person_image`	string	Image URL of the person to edit
`garment_images`	array of string	Up to 11 garment refs; ≤6 for reliable finals, 7–8 often lands all pieces, 9–11 may drop the last item(s); extra URLs beyond 11 are ignored

Optional

Parameter	Type	Default	Description
`seed`	integer	—	Random seed. Leave blank for a random seed.
`output_format`	string	`jpg`	Format of the saved output image (`webp`, `jpg`, `png`).
`output_quality`	integer	`95`	Quality for jpg/webp outputs from 0 to 100.
`preserve_input_size`	boolean	`true`	Output matches person_image dimensions (model resizes internally).

Extended optional fields

Parameter	Type	Default	Description
`turbo`	boolean	`false`	Faster multi-garment pass (~2.5–3.5 s); see Turbo mode
`reference_pose`	string	—	Optional person image URL; output pose matches this reference
`prompt`	string	—	EXPERIMENTAL. For non-flatlay garment images; names which garment to use from which image, e.g. `"the green t-shirt from image 1 and the trousers from image 2"`

Before generating

Generation diversity — ritual seed + axis rotation before each try-on job (reuse hero project_seed when dressing an approved plate).
Confirm with the user:

person_image — person photo with clear visibility of the body region to dress
garment_images — up to 11 refs (≤6 finals; 7–8 reliable; one item per body spot — see Multi-garment limits)
turbo, reference_pose, prompt when relevant (see sections below)
seed, output_format, output_quality, preserve_input_size when delivery format matters

Run p-image-try-on-quality-checklist.md on outputs before downstream steps.

Production quality (not basic demos)

Upstream person plate: try-on fidelity is capped by person_image. Follow realistic-persona-showcase.md for photoreal p-image plates — not generic catalog mannequins. Garment tiers: see Showcase tiers below and p-image-try-on-quality-checklist.md.

Showcase tiers to plan for:

Tier	Example capability
Editorial still	Artistic prints, color-block sleeves, seated lifestyle poses
Complex garments	Collaged / patchwork suits, fine pleats, multi-panel streetwear
In-scene accessories	Hats, logo tees, glasses — mirror/street compositions preserved
Multi-garment stack	Jacket + tee + pants + hat + shoes in one pass (normal mode; plan ≤6 for finals)

Anti-slop: avoid white-background-only demos, mushy AI person plates, turbo-only finals on complex stacks, and repeating one default face across examples. Rotate cast and settings per visual-variety-bible.md.

Replicate playground: diversify pinned examples on p-image-try-on per realistic-persona-showcase.md.

Garment inputs

The model accepts a broader range of garment images than flat-lay packshots alone:

Input type	Notes
Flat-lay / packshot	Best default; no `prompt` needed
On-model / lifestyle	Supported; use `prompt` to identify the garment
Multi-garment in one image	Supported; use `prompt` to pick which items to apply

When a garment image shows multiple items or the garment is worn by someone else, set prompt to disambiguate (EXPERIMENTAL).

Multi-garment limits

Rule	Guidance
One item per body spot	Socks + shoes on the feet → expect one winner (usually shoes). Send socks alone if you need them.
Reliable count	≤6 for delivery assets; 7–8 usually lands; 9–11 treat extras as bonus — last pieces not guaranteed.
Restyling	Fix person + base garments; swap one `garment_images[]` URL and rerun for variant tops (catalog A/B).

Person image: full-body or three-quarter works best — the model needs the body region to dress. Tight crops can artifact.

Garment categories

The model auto-classifies each garment image. Misclassified or unsupported refs may be skipped (run can still succeed) — omit known-bad types from garment_images[] rather than hoping they drop.

Works well: tops and shirts; sweaters, hoodies, and blazers; pants, jeans, shorts, and skirts; dresses, jumpsuits, and rompers; jackets and coats; underwear and swimwear; footwear (shoes, boots, sandals, socks — person photo must show feet); headwear (hats, caps, beanies, sunglasses, eyeglasses); neckwear (scarves, ties, necklaces); bags (handbags, totes, backpacks); select jewelry (watches, bracelets, rings, earrings).

Omit from request: gloves and mittens; arm warmers; handheld props (phones, wallets, umbrellas, cups, keychains); pocket squares, suspenders, and brooches.

Turbo mode

Turbo applies multiple garment edits in one larger edit instead of processing garments separately.

	Normal (default)	Turbo (`turbo: true`)
Speed	Scales with garment count	~2.5–3.5 s regardless of garment count
Quality	Highest fidelity	May be slightly lower; some garments may not apply correctly
Pricing	Per-garment table above	Same pricing
Best for	Final assets, 5+ garments	Previews, batch catalogs, speed-critical flows (~4 pieces sweet spot)

Guidance:

Turbo is disabled by default — enable explicitly when the user prioritizes latency.
Same price as normal; works with reference_pose, prompt, etc.
Pruna docs: not recommended above ~4 garments in turbo — use normal mode for larger stacks and final delivery.

Example: extended input (turbo + pose + prompt)

bash

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-try-on' \
  -d '{
    "input": {
      "person_image": "https://api.pruna.ai/v1/files/PERSON_FILE_ID",
      "garment_images": [
        "https://api.pruna.ai/v1/files/MULTI_GARMENT_SHOT_ID",
        "https://api.pruna.ai/v1/files/BOTTOM_ID"
      ],
      "reference_pose": "https://api.pruna.ai/v1/files/POSE_REF_ID",
      "prompt": "the green t-shirt from image 1 and the trousers from image 2",
      "turbo": true,
      "output_format": "jpg",
      "output_quality": 95,
      "preserve_input_size": true
    }
  }'

Typical next steps

Restyle one piece: keep person_image + unchanged garments; swap a single garment_images[] URL per variant.
Upscale for delivery: p-image-upscale
Animate try-on still: p-video (I2V) or p-video-avatar
Ecommerce pack workflows: pruna-generative-pipeline recipe K