Skip to content
GenMedia Creative Studio

Virtual Try-On (VTO)

The Virtual Try-On (VTO) workflow (/vto) provides a dedicated studio for Vertex AI’s specialized Virtual Try-On model (virtual-try-on-001 via client.models.recontext_image), paired with a synthetic Virtual Model Generator (VirtualModelGenerator) and one-click Veo video animation.

Virtual Try-On Workflow Architecture

Source Architecture

Section titled “Source Architecture”
Layer File Path Responsibility
Mesop Page & State pages/vto.py (PageState) Person & garment uploaders, synthetic model generator trigger, sample count & seed controls, result gallery, and inline Veo animation.
VTO Model Adapter models/vto.py (generate_vto_image, call_virtual_try_on) Wraps client.models.recontext_image() (virtual-try-on-001) and the GAPIC PredictionServiceClient helper for fitting product images onto a person image.
Synthetic Model Generator models/virtual_model_generator.py (VirtualModelGenerator, DEFAULT_PROMPT) Generates diverse, studio-lit full-body fashion model portraits (imagen-3.0-generate-002 / Nano Banana) using randomized gender, silhouette, and pose attributes from config/virtual_model_options.json.

Step-by-Step Pipeline

Section titled “Step-by-Step Pipeline”

Step 1: Person Image Selection or Synthetic Model Generation

Section titled “Step 1: Person Image Selection or Synthetic Model Generation”

Users can provide the target person portrait (person_image_gcs) in three ways:

  1. Upload / Library Selection: Upload a full-body or half-body portrait or pick an existing model from the Studio Library.
  2. Generate Virtual Model (on_generate_person_image): Click Generate Virtual Model to invoke VirtualModelGenerator, which samples a randomized combination of gender presentation, body silhouette, and camera framing from config/virtual_model_options.json and synthesizes a clean studio portrait ready for apparel fitting.

Step 2: Product Garment Selection

Section titled “Step 2: Product Garment Selection”
  • Upload or select a product garment image (product_image_gcs)—such as a top, dress, jacket, or bottom—on a flat lay, mannequin, or model.

Step 3: Vertex AI VTO Synthesis (models.recontext_image)

Section titled “Step 3: Vertex AI VTO Synthesis (models.recontext_image)”
  • generate_vto_image() calls client.models.recontext_image() with model=cfg.VTO_MODEL_ID (virtual-try-on-001), passing:
    • source=types.RecontextImageSource(person_image=..., product_images=[types.ProductImage(product_image=...)])
    • config=types.RecontextImageConfig(number_of_images=sample_count, base_steps=base_steps, safety_filter_level=..., person_generation=...)
  • Each generated try-on image is uploaded to gs://<GENMEDIA_BUCKET>/vto_result_images/ and logged to Firestore (MediaItem with media_type="vto").

Step 4: Optional Inline Veo Video Animation (on_animate_click)

Section titled “Step 4: Optional Inline Veo Video Animation (on_animate_click)”
  • Every synthesized VTO result card includes an Animate with Veo button. Clicking it submits the try-on result directly to models/veo.py (veo-3.1-fast-generate-001, 9:16 portrait aspect ratio) to render a natural fashion movement clip (result_videos on the same card).