feat(ui): add capture mode selector (single/multi/video WIP)

Add a Radio selector in the Capture tab with three modes:
- Single garment: skips YOLO, sends photo directly to VLM with
single-garment prompt (ideal for close-up photos)
- Multiple garments: full pipeline (YOLO + crops + VLM per crop,
fallback to multi-garment VLM analysis)
- Video stream: placeholder showing "coming soon" message

New vision.extract_single_from_path() for direct single-garment
extraction without YOLO detection step.

Co-authored-by: Cursor <cursoragent@cursor.com>

app.py

@@ -11,7 +11,7 @@ import os
 
 import gradio as gr
 
-from src.vision import extract_garments
+from src.vision import extract_garments, extract_single_from_path
 from src.catalog import (
     add_garments,
     load_catalog,
@@ -47,16 +47,23 @@ mediante lenguaje natural. Sube fotos → detecta prendas → pregunta lo que qu
 """
 
 
-def analyze_images(images):
-    """Process uploaded images and extract garments."""
+def analyze_images(images, mode):
+    """Process uploaded images and extract garments based on capture mode."""
     if not images:
         return "Sube al menos una imagen.", []
 
+    if mode == "video":
+        return "Video stream estará disponible próximamente.", []
+
     all_added = []
     for img_data in images:
         image_path = img_data if isinstance(img_data, str) else img_data.name
         try:
-            garments_with_crops = extract_garments(image_path)
+            if mode == "single":
+                garments_with_crops = extract_single_from_path(image_path)
+            else:
+                garments_with_crops = extract_garments(image_path)
+
             if garments_with_crops:
                 added = add_garments(garments_with_crops)
                 all_added.extend(added)
@@ -295,6 +302,15 @@ with gr.Blocks(title="Wardrobe AI") as demo:
 
         with gr.Row():
             with gr.Column(scale=1):
+                capture_mode = gr.Radio(
+                    choices=[
+                        ("Prenda individual", "single"),
+                        ("Múltiples prendas", "multi"),
+                        ("Video stream (próximamente)", "video"),
+                    ],
+                    value="multi",
+                    label="Modo de captura",
+                )
                 image_input = gr.File(
                     label="Fotos del armario",
                     file_count="multiple",
@@ -313,7 +329,7 @@ with gr.Blocks(title="Wardrobe AI") as demo:
 
         analyze_btn.click(
             analyze_images,
-            inputs=[image_input],
+            inputs=[image_input, capture_mode],
             outputs=[status_output, detected_table],
         )
 

src/vision.py

@@ -220,3 +220,18 @@ def extract_garments(image_path: str) -> list[tuple[dict, bytes]]:
     logger.info("Falling back to full-image analysis")
     garments, full_bytes = _extract_from_full_image(image_path)
     return [(g, full_bytes) for g in garments]
+
+
+def extract_single_from_path(image_path: str) -> list[tuple[dict, bytes]]:
+    """Extract a single garment from a photo (no YOLO, direct VLM).
+
+    Use when the user photographs one garment at a time.
+    Returns a list with 0 or 1 (garment_dict, image_bytes) tuples.
+    """
+    logger.info("Single-garment mode: %s", image_path)
+    _, image_bytes = _prepare_image(image_path)
+    garment = _extract_single_garment(image_bytes)
+
+    if garment:
+        return [(garment, image_bytes)]
+    return []