romi2001 commited on
Commit
93d88c1
·
verified ·
1 Parent(s): 37bcea1

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +162 -649
app.py CHANGED
@@ -1,670 +1,183 @@
1
- """
2
- AI Fashion Stylist — Hugging Face Space app
3
- Combines: CLIP recommendation engine (Part 3) + 3 GenAI patterns (Part 4):
4
- 1. Small Language Model -> generates text (Qwen2.5-0.5B-Instruct)
5
- 2. Small Vision-Language Model -> answers questions (BLIP-VQA-base)
6
- 3. Small Vision-Language pipeline -> generates an image (clothing segmentation + SD inpainting)
7
-
8
- Deploy notes:
9
- - Upload this file + requirements.txt + final_image_embeddings.npy + catalog_metadata.parquet
10
- to your HF Space repo root.
11
- - Update HF_DATASET_REPO / HF_WINNING_MODEL below if your repo names differ.
12
- - Quick Starter sample photos are generated automatically from the dataset at startup -
13
- no manual file upload needed.
14
- """
15
-
16
- import base64
17
- import os
18
- import urllib.parse
19
- from io import BytesIO
20
-
21
- import numpy as np
22
- import pandas as pd
23
- import torch
24
  import gradio as gr
25
- import faiss
26
- from datasets import load_dataset
27
  from PIL import Image
28
- from scipy import ndimage
29
- from transformers import (
30
- CLIPModel, CLIPProcessor, pipeline as hf_pipeline,
31
- BlipProcessor, BlipForQuestionAnswering,
32
- SegformerImageProcessor, AutoModelForSemanticSegmentation,
33
- )
34
- from diffusers import StableDiffusionInpaintPipeline
35
-
36
- # ---------------------------------------------------------------------------
37
- # CONFIG — update these to match your own HF repos
38
- # ---------------------------------------------------------------------------
39
- HF_DATASET_REPO = "lihicarmeli/fashion-stylist-multimodal-v2" # your HF dataset repo
40
- HF_WINNING_MODEL = "openai/clip-vit-base-patch32" # winning embedding model (Part 3)
41
- EMBEDDINGS_FILE = "final_image_embeddings.npy" # uploaded next to this app.py
42
- METADATA_FILE = "catalog_metadata.parquet" # uploaded next to this app.py
43
-
44
- DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
45
-
46
- # ---------------------------------------------------------------------------
47
- # LOAD DATA + WINNING EMBEDDING MODEL (runs once, on Space startup)
48
- # ---------------------------------------------------------------------------
49
- print("Loading dataset from HF Hub...")
50
- ds = load_dataset(HF_DATASET_REPO)
51
- df = ds["train"].to_pandas()
52
- images = [ds["train"][i]["image_improved"] for i in range(len(ds["train"]))]
53
-
54
- print("Generating Quick Starter sample photos from the dataset...")
55
- SAMPLES_DIR = "samples_cache"
56
- os.makedirs(SAMPLES_DIR, exist_ok=True)
57
-
58
-
59
- def _pick_index(filter_fn, fallback_idx=0):
60
- matches = df[df.apply(filter_fn, axis=1)]
61
- return int(matches.index[0]) if len(matches) else fallback_idx
62
-
63
-
64
- _sample_specs = [
65
- ("demo_woman.jpg", lambda r: r["gender"] == "woman"),
66
- ("demo_man.jpg", lambda r: r["gender"] == "man"),
67
- ("demo_teen.jpg", lambda r: r["age_group"] == "teen"),
68
- ]
69
-
70
- SAMPLE_PHOTOS = []
71
- for _filename, _filt in _sample_specs:
72
- _idx = _pick_index(_filt, fallback_idx=len(SAMPLE_PHOTOS))
73
- _path = os.path.join(SAMPLES_DIR, _filename)
74
- images[_idx].convert("RGB").save(_path)
75
- SAMPLE_PHOTOS.append(_path)
76
- print(f" -> {len(SAMPLE_PHOTOS)} sample photos ready: {SAMPLE_PHOTOS}")
77
-
78
- print("Loading precomputed embeddings...")
79
- image_embeddings = np.load(EMBEDDINGS_FILE).astype("float32")
80
-
81
- print("Loading winning embedding model (CLIP) from HF Hub...")
82
- win_model = CLIPModel.from_pretrained(HF_WINNING_MODEL).to(DEVICE).eval()
83
- win_processor = CLIPProcessor.from_pretrained(HF_WINNING_MODEL)
84
-
85
- print("Building FAISS index...")
86
- dimension = image_embeddings.shape[1]
87
- faiss_img_index = faiss.IndexFlatL2(dimension)
88
- faiss.normalize_L2(image_embeddings)
89
- faiss_img_index.add(image_embeddings)
90
-
91
- # ---------------------------------------------------------------------------
92
- # GENERATION MODELS — the 3 "Good Examples" patterns
93
- # ---------------------------------------------------------------------------
94
- print("Loading small language model for text generation (Qwen2.5-0.5B-Instruct)...")
95
- caption_gen_pipe = hf_pipeline(
96
- "text-generation",
97
- model="Qwen/Qwen2.5-0.5B-Instruct",
98
- device=0 if DEVICE == "cuda" else -1,
99
- )
100
-
101
- print("Loading small vision-language model for VQA (BLIP-VQA-base)...")
102
- vqa_processor = BlipProcessor.from_pretrained("Salesforce/blip-vqa-base")
103
- vqa_model = BlipForQuestionAnswering.from_pretrained("Salesforce/blip-vqa-base").to(DEVICE)
104
-
105
- print("Loading clothing segmentation model (helper - finds where the clothes are)...")
106
- seg_processor = SegformerImageProcessor.from_pretrained("mattmdjaga/segformer_b2_clothes")
107
- seg_model = AutoModelForSemanticSegmentation.from_pretrained("mattmdjaga/segformer_b2_clothes").to(DEVICE)
108
-
109
- print("Loading image-generation model for the new outfit (Stable Diffusion Inpainting)...")
110
- inpaint_pipe = StableDiffusionInpaintPipeline.from_pretrained(
111
- "runwayml/stable-diffusion-inpainting",
112
- torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
113
- safety_checker=None,
114
- ).to(DEVICE)
115
-
116
- GENDERS = sorted(df["gender"].unique().tolist())
117
- AGE_GROUPS = sorted(df["age_group"].unique().tolist())
118
- SKIN_TONES = sorted(df["skin_tone"].unique().tolist())
119
- UNDERTONES = sorted(df["undertone"].unique().tolist())
120
- STYLES = sorted(df["style_preference"].unique().tolist())
121
-
122
- print("Space ready.")
123
-
124
-
125
- # ---------------------------------------------------------------------------
126
- # EMBEDDING + SEARCH (Part 3 logic, unchanged)
127
- # ---------------------------------------------------------------------------
128
- @torch.no_grad()
129
- def embed_query_image(pil_image):
130
- inputs = win_processor(images=pil_image, return_tensors="pt").to(DEVICE)
131
- feats = win_model.get_image_features(**inputs)
132
- feats = feats.pooler_output if hasattr(feats, "pooler_output") else feats
133
- return feats.cpu().numpy().astype("float32")
134
-
135
-
136
- @torch.no_grad()
137
- def embed_query_text(sentence):
138
- inputs = win_processor(text=[sentence], return_tensors="pt", padding=True, truncation=True).to(DEVICE)
139
- feats = win_model.get_text_features(**inputs)
140
- feats = feats.pooler_output if hasattr(feats, "pooler_output") else feats
141
- return feats.cpu().numpy().astype("float32")
142
-
143
-
144
- def build_feature_sentence(skin_tone, undertone, style_preference, gender=None, age_group=None):
145
- descriptor = " ".join(p for p in [age_group, gender] if p) or "person"
146
- return (
147
- f"a {descriptor} with {skin_tone} skin tone and {undertone} undertone, "
148
- f"wearing a {style_preference} style outfit"
149
- )
150
-
151
-
152
- def faiss_filtered_search(query_emb, top_k=3, exclude_idx=None, gender=None, age_group=None):
153
- faiss.normalize_L2(query_emb)
154
- k = len(df)
155
- distances, indices = faiss_img_index.search(query_emb, k)
156
- distances, indices = distances[0], indices[0]
157
-
158
- def collect(require_gender, require_age):
159
- kept_i, kept_d = [], []
160
- for idx, dist in zip(indices, distances):
161
- if idx == -1 or (exclude_idx is not None and idx == exclude_idx):
162
- continue
163
- row = df.iloc[idx]
164
- if require_gender and gender and str(row["gender"]).lower() != str(gender).lower():
165
- continue
166
- if require_age and age_group and str(row["age_group"]).lower() != str(age_group).lower():
167
- continue
168
- kept_i.append(idx)
169
- kept_d.append(dist)
170
- if len(kept_i) == top_k:
171
- break
172
- return kept_i, kept_d
173
-
174
- kept_i, kept_d = collect(True, True)
175
- if len(kept_i) < top_k:
176
- kept_i, kept_d = collect(True, False)
177
- if len(kept_i) < top_k:
178
- kept_i, kept_d = collect(False, False)
179
-
180
- return np.array(kept_i), df.iloc[kept_i], np.array(kept_d)
181
-
182
-
183
- # ---------------------------------------------------------------------------
184
- # GENERATION — 3 "Good Examples" patterns (Part 4)
185
- # ---------------------------------------------------------------------------
186
-
187
- # --- Pattern 1: small Language Model -> generate text ---
188
- def generate_stylist_caption(row):
189
- user_prompt = (
190
- f"Write one short, warm sentence (max 25 words) from a fashion stylist, recommending this look: "
191
- f"a {row['style_preference']} style outfit in {row['primary_color']} and {row['secondary_color']}, "
192
- f"best colors: {row['recommended_colors']}. Be specific and stylish, no hashtags."
193
- )
194
- messages = [{"role": "user", "content": user_prompt}]
195
- output = caption_gen_pipe(messages, max_new_tokens=40, do_sample=True, temperature=0.7)
196
- return output[0]["generated_text"][-1]["content"].strip()
197
-
198
-
199
- # --- Pattern 2: small Vision-Language Model -> answer questions about an image ---
200
- @torch.no_grad()
201
- def answer_question_about_image(pil_image, question):
202
- inputs = vqa_processor(pil_image.convert("RGB"), question, return_tensors="pt").to(DEVICE)
203
- output_ids = vqa_model.generate(**inputs, max_new_tokens=20)
204
- return vqa_processor.decode(output_ids[0], skip_special_tokens=True)
205
-
206
-
207
- # --- Pattern 3: small Vision-Language Model -> generate the FULL LOOK (full-body image) ---
208
- FACE_LABEL_ID = 11
209
- HAIR_LABEL_ID = 2
210
-
211
-
212
- @torch.no_grad()
213
- def get_face_protect_mask(pil_image, use_geometric_fallback=True):
214
- """Returns a boolean mask (True = protect) covering the face + hair, at pil_image's own
215
- resolution. Combines the segmentation model's prediction with a fixed geometric ellipse
216
- (centered, where a face statistically sits in a headshot crop) as a safety net - so even
217
- if segmentation misclassifies an unusual headwear/turban, the visible face is still
218
- guaranteed to be protected."""
219
- inputs = seg_processor(images=pil_image, return_tensors="pt").to(DEVICE)
220
- logits = seg_model(**inputs).logits
221
- upsampled = torch.nn.functional.interpolate(
222
- logits, size=pil_image.size[::-1], mode="bilinear", align_corners=False
223
- )
224
- pred_seg = upsampled.argmax(dim=1)[0].cpu().numpy()
225
- seg_protect = np.isin(pred_seg, [FACE_LABEL_ID, HAIR_LABEL_ID])
226
-
227
- if not use_geometric_fallback:
228
- return seg_protect
229
-
230
- h, w = seg_protect.shape
231
- yy, xx = np.mgrid[0:h, 0:w]
232
- cy, cx = h * 0.42, w * 0.5 # face center: slightly above vertical middle of a headshot
233
- ry, rx = h * 0.30, w * 0.22 # ellipse radii tuned for a tight headshot/bust crop
234
- geometric_protect = (((xx - cx) / rx) ** 2 + ((yy - cy) / ry) ** 2) <= 1.0
235
-
236
- return seg_protect | geometric_protect
237
-
238
-
239
- def build_full_look_prompt(row):
240
- gender_word = "man" if str(row["gender"]).lower() in ("man", "male") else "woman"
241
- return (
242
- f"full body fashion photo of a {row['age_group']} {gender_word}, standing, "
243
- f"wearing a {row['style_preference']} style outfit: {row['outfit_top']}, "
244
- f"{row['outfit_bottom']}, {row['outfit_shoes']}, {row['outfit_accessory']}, "
245
- f"in {row['primary_color']} and {row['secondary_color']}, "
246
- f"studio lighting, plain background, head to toe, high quality fashion photography"
247
- )
248
-
249
-
250
- def generate_new_outfit_image(pil_image, row, target_gender=None, canvas_size=(512, 1024),
251
- head_width_frac=0.45, steps=40, guidance_scale=8.0):
252
- """Generates the FULL LOOK: keeps the original face pixel-identical (scaled down to a
253
- realistic head-to-body proportion) and generates the rest of a standing figure wearing
254
- the complete recommended outfit (top, bottom, shoes, accessory) around it."""
255
- if target_gender is not None and str(row["gender"]).lower() != str(target_gender).lower():
256
- print(f"⚠️ Warning: recommended row gender ({row['gender']}) != expected gender "
257
- f"({target_gender}) - double-check which matched_rows was passed in.")
258
-
259
- base = pil_image.convert("RGB")
260
- protect = get_face_protect_mask(base) # (H, W) bool, at base's own resolution
261
 
262
- canvas_w, canvas_h = canvas_size
263
- head_w = int(canvas_w * head_width_frac)
264
- scale = head_w / base.width
265
- head_h = int(base.height * scale)
266
-
267
- resized_face_crop = base.resize((head_w, head_h))
268
- resized_protect_img = Image.fromarray(protect.astype(np.uint8) * 255).resize(
269
- (head_w, head_h), resample=Image.NEAREST
270
- )
271
-
272
- canvas = Image.new("RGB", canvas_size, color=(128, 128, 128))
273
- paste_x = (canvas_w - head_w) // 2
274
- paste_y = int(canvas_h * 0.03)
275
- canvas.paste(resized_face_crop, (paste_x, paste_y))
276
-
277
- mask_arr = np.full((canvas_h, canvas_w), 255, dtype=np.uint8) # 255 = let the model generate
278
- protect_resized_arr = np.array(resized_protect_img) > 127
279
- mask_arr[paste_y:paste_y + head_h, paste_x:paste_x + head_w][protect_resized_arr] = 0
280
- mask = Image.fromarray(mask_arr).convert("L")
281
-
282
- prompt = build_full_look_prompt(row)
283
- generated = inpaint_pipe(
284
- prompt=prompt,
285
- image=canvas,
286
- mask_image=mask,
287
- num_inference_steps=steps,
288
- guidance_scale=guidance_scale,
289
- height=canvas_h,
290
- width=canvas_w,
291
- ).images[0]
292
-
293
- # Hard-composite: guarantees zero change on the protected face/hair pixels
294
- final_image = Image.composite(generated, canvas, mask)
295
- return final_image, prompt
296
-
297
-
298
- # ---------------------------------------------------------------------------
299
- # SHOP LINKS (Part 3 logic, unchanged)
300
- # ---------------------------------------------------------------------------
301
- RETAILER_SEARCH_URLS = {
302
- "zara": "https://www.zara.com/us/en/search?searchTerm={query}&section={section}",
303
- "hm": "https://www2.hm.com/en_us/search-results.html?q={query}",
304
- "asos": "https://www.asos.com/us/{dept}/search/?q={query}",
305
- "mango": "https://shop.mango.com/us/en/search?kw={query}",
306
- "shein": "https://us.shein.com/pdsearch/{query}/",
307
- }
308
-
309
-
310
- def normalize_gender(gender):
311
- g = str(gender).strip().lower() if gender is not None else ""
312
- return "men" if g in ("male", "man", "men", "m") else "women"
313
-
314
-
315
- def to_shop_link(retailer, value, gender=None):
316
- dept = normalize_gender(gender)
317
- if retailer == "zara":
318
- section = "MAN" if dept == "men" else "WOMAN"
319
- return RETAILER_SEARCH_URLS["zara"].format(query=urllib.parse.quote(str(value)), section=section)
320
- if retailer == "asos":
321
- return RETAILER_SEARCH_URLS["asos"].format(query=urllib.parse.quote(str(value)), dept=dept)
322
- gender_word = "men's" if dept == "men" else "women's"
323
- return RETAILER_SEARCH_URLS[retailer].format(query=urllib.parse.quote(f"{gender_word} {value}"))
324
-
325
-
326
- def shop_links_markdown(row):
327
- lines = []
328
- for retailer, col in [("zara", "search_query_zara"), ("hm", "search_query_hm"),
329
- ("asos", "search_query_asos"), ("mango", "search_query_mango"),
330
- ("shein", "search_query_shein")]:
331
- link = to_shop_link(retailer, row[col], gender=row["gender"])
332
- lines.append(f"- **{retailer.upper()}**: [{row[col]}]({link})")
333
- return "\n".join(lines)
334
-
335
-
336
- # ---------------------------------------------------------------------------
337
- # UI CARDS — original visual layer, built on top of the same Part 3+4 data
338
- # ---------------------------------------------------------------------------
339
- _SWATCH_CSS = {
340
- "ice white": "#f4f3ee", "bold blue": "#1d4ed8", "royal blue": "#1e3a8a",
341
- "fuchsia": "#c026d3", "cool red": "#dc2626", "deep teal": "#0f766e",
342
- "plum": "#6b21a8", "silver": "#cbd5e1", "bright pink": "#ec4899",
343
- "deep jewel tones": "#581c87", "rust": "#b45309", "camel": "#c19a6b",
344
- "coral": "#fb7185", "ivory": "#fffff0", "olive": "#65730a",
345
- "terracotta": "#c2643a",
346
  }
347
 
 
 
 
 
 
 
 
 
 
348
 
349
- def _swatch_color(name):
350
- key = name.strip().lower()
351
- if key in _SWATCH_CSS:
352
- return _SWATCH_CSS[key]
353
- # generic fallback so unseen color names still render *something* sensible
354
- palette = ["#b45309", "#1d4ed8", "#c026d3", "#0f766e", "#dc2626", "#6b21a8"]
355
- return palette[hash(key) % len(palette)]
356
-
357
-
358
- def _lighten_hex(hex_color, factor=0.82):
359
- hex_color = hex_color.lstrip("#")
360
- r, g, b = int(hex_color[0:2], 16), int(hex_color[2:4], 16), int(hex_color[4:6], 16)
361
- r = int(r + (255 - r) * factor)
362
- g = int(g + (255 - g) * factor)
363
- b = int(b + (255 - b) * factor)
364
- return f"#{r:02x}{g:02x}{b:02x}"
365
-
366
-
367
- _PALETTE_NAMES = {
368
- "fair": {"cool": "Frost Bloom", "warm": "Sunlit Honey", "neutral": "Soft Linen"},
369
- "light": {"cool": "Moonlit Rose", "warm": "Amber Glow", "neutral": "Quiet Dawn"},
370
- "medium": {"cool": "Dusty Orchid", "warm": "Spiced Caramel", "neutral": "Warm Clay"},
371
- "olive": {"cool": "Sage Twilight", "warm": "Bronzed Olive", "neutral": "Earth Tone"},
372
- "tan": {"cool": "Copper Mist", "warm": "Golden Sand", "neutral": "Desert Bloom"},
373
- "deep": {"cool": "Midnight Jewel", "warm": "Mahogany Sun", "neutral": "Onyx Velvet"},
374
- }
375
-
376
-
377
- def get_palette_name(skin_tone, undertone):
378
- return _PALETTE_NAMES.get(str(skin_tone).strip().lower(), {}).get(
379
- str(undertone).strip().lower(), "Signature Palette"
380
- )
381
-
382
-
383
- def pil_to_base64(img, max_size=420):
384
- img = img.convert("RGB").copy()
385
- img.thumbnail((max_size, max_size))
386
- buf = BytesIO()
387
- img.save(buf, format="JPEG", quality=85)
388
- return base64.b64encode(buf.getvalue()).decode("utf-8")
389
-
390
-
391
- def build_feature_pills_html(labels, selected, title):
392
- """Pill-style recap row (her visual pattern) - but built from your real, actually-used
393
- attributes (skin tone / undertone / style), not placeholder categories."""
394
- pills = ""
395
- for label in labels:
396
- active = label == selected
397
- border = "2px solid #C94A72" if active else "1.5px solid #E8E0E8"
398
- bg = "#FFF0F5" if active else "#FFFFFF"
399
- tcol = "#C94A72" if active else "#444"
400
- dot = _swatch_color(label) if title != "Style" else "#C19A6B"
401
- pills += (
402
- f'<div style="display:inline-flex;align-items:center;gap:7px;padding:8px 14px;'
403
- f'border-radius:30px;border:{border};background:{bg};margin:4px;">'
404
- f'<div style="width:14px;height:14px;border-radius:50%;background:{dot};'
405
- f'border:1px solid rgba(0,0,0,.12);"></div>'
406
- f'<span style="font-size:13px;font-weight:500;color:{tcol};">{label.title()}</span></div>'
407
- )
408
- return (
409
- f'<div style="margin-bottom:14px;"><div style="font-size:11px;font-weight:600;color:#aaa;'
410
- f'text-transform:uppercase;letter-spacing:.1em;margin-bottom:8px;">{title}</div>'
411
- f'<div style="display:flex;flex-wrap:wrap;margin:-4px;">{pills}</div></div>'
412
- )
413
-
414
-
415
- def build_features_recap_html(skin_tone, undertone, style):
416
- return (
417
- '<div style="background:#fff;border-radius:16px;padding:18px 20px;'
418
- 'margin-bottom:14px;border:1px solid #F0E8F0;">'
419
- '<div style="font-size:11px;font-weight:600;color:#C94A72;text-transform:uppercase;'
420
- 'letter-spacing:.1em;margin-bottom:14px;">Your Features</div>'
421
- + build_feature_pills_html(SKIN_TONES, skin_tone, "Skin tone")
422
- + build_feature_pills_html(UNDERTONES, undertone, "Undertone")
423
- + build_feature_pills_html(STYLES, style, "Style")
424
- + "</div>"
425
- )
426
- colors = [c.strip() for c in str(row["recommended_colors"]).split(",") if c.strip()][:4]
427
- swatches = "".join(
428
- f'<div style="text-align:center;margin-right:16px;">'
429
- f'<div style="width:38px;height:38px;border-radius:50%;background:{_swatch_color(c)};'
430
- f'border:2px solid #fff;box-shadow:0 0 0 1px #e5e7eb;margin:0 auto 6px;"></div>'
431
- f'<span style="font-size:11px;color:#6b7280;">{c.title()}</span></div>'
432
- for c in colors
433
- )
434
- palette_name = get_palette_name(row["skin_tone"], row["undertone"])
435
- return f"""
436
- <div style="font-family:inherit;">
437
- <div style="position:relative;overflow:hidden;background:#1a1a1a;color:white;
438
- border-radius:18px;padding:24px 28px;margin-bottom:16px;">
439
- <div style="position:absolute;top:-40px;right:-40px;width:150px;height:150px;border-radius:50%;
440
- background:rgba(201,74,114,.25);"></div>
441
- <div style="font-size:11px;letter-spacing:1.5px;color:#C94A72;font-weight:700;">YOUR COLOR PALETTE</div>
442
- <div style="font-size:30px;font-weight:800;margin:6px 0 4px;">{palette_name}</div>
443
- <div style="font-size:13px;color:#bdb8c9;">
444
- {row['skin_tone'].title()} skin · {row['undertone'].title()} undertone · {row['style_preference'].title()} style
445
  </div>
446
- </div>
447
- <div style="display:flex;gap:14px;flex-wrap:wrap;">
448
- <div style="flex:1;min-width:220px;background:#f9fafb;border-radius:14px;padding:18px;">
449
- <div style="font-size:11px;letter-spacing:1.5px;color:#9ca3af;font-weight:700;margin-bottom:10px;">COLORS FOR YOU</div>
450
- <div style="display:flex;flex-wrap:wrap;">{swatches}</div>
 
 
 
451
  </div>
452
- <div style="flex:1;min-width:220px;background:#f9fafb;border-radius:14px;padding:18px;">
453
- <div style="font-size:11px;letter-spacing:1.5px;color:#C94A72;font-weight:700;margin-bottom:8px;">STYLIST TIP</div>
454
- <div style="font-size:14px;color:#1f2937;line-height:1.4;">{caption}</div>
 
 
 
 
 
 
455
  </div>
456
- </div>
457
  </div>
458
  """
459
-
460
-
461
- def build_outfit_component_cards_html(row):
462
- """One card per garment piece of the TOP match (top/bottom/shoes/accessory) -
463
- mirrors a 'your outfit, broken into shoppable pieces' layout."""
464
- colors = [c.strip() for c in str(row["recommended_colors"]).split(",") if c.strip()] or ["neutral"]
465
- components = [
466
- ("TOP", row.get("outfit_top", "Top"), "zara", "search_query_zara"),
467
- ("BOTTOM", row.get("outfit_bottom", "Bottom"), "hm", "search_query_hm"),
468
- ("SHOES", row.get("outfit_shoes", "Shoes"), "asos", "search_query_asos"),
469
- ("ACCESSORY", row.get("outfit_accessory", "Accessory"), "mango", "search_query_mango"),
470
- ]
471
- cards = []
472
- for i, (label, item_name, retailer, col) in enumerate(components):
473
- dot_color = _swatch_color(colors[i % len(colors)])
474
- banner_color = _lighten_hex(dot_color)
475
- link = to_shop_link(retailer, row[col], row["gender"])
476
- cards.append(f"""
477
- <div style="border:1px solid #eee;border-radius:14px;overflow:hidden;width:185px;
478
- box-shadow:0 1px 3px rgba(0,0,0,0.05);">
479
- <div style="background:{banner_color};height:64px;display:flex;align-items:center;justify-content:center;">
480
- <div style="width:30px;height:30px;border-radius:50%;background:{dot_color};border:2px solid white;"></div>
481
- </div>
482
- <div style="padding:12px 14px;">
483
- <div style="font-size:10px;letter-spacing:1px;color:#C94A72;font-weight:700;">{label}</div>
484
- <div style="font-size:13px;font-weight:600;margin:4px 0 2px;color:#111827;">{str(item_name).title()}</div>
485
- <div style="font-size:11px;color:#9ca3af;margin-bottom:10px;">{retailer.upper()}</div>
486
- <a href="{link}" target="_blank" style="display:block;text-align:center;background:#111827;
487
- color:white;border-radius:8px;padding:8px;text-decoration:none;font-size:12px;font-weight:600;">
488
- Shop ↗
489
- </a>
490
- </div>
491
- </div>
492
- """)
493
- return f'<div style="display:flex;gap:14px;flex-wrap:wrap;">{"".join(cards)}</div>'
494
-
495
-
496
- def build_more_matches_html(matched_indices, matched_rows, similarities, skip_first=True):
497
- """Smaller secondary row: the other catalog matches, as full-look thumbnails
498
- (keeps the '3 similar items' requirement visible alongside the per-garment cards above)."""
499
- items = list(zip(matched_indices, similarities))[1:] if skip_first else list(zip(matched_indices, similarities))
500
- if not items:
501
- return ""
502
- cards = []
503
- for idx, sim in items:
504
- row = df.iloc[idx]
505
- img_b64 = pil_to_base64(images[idx])
506
- link = to_shop_link("zara", row["search_query_zara"], row["gender"])
507
- cards.append(f"""
508
- <div style="border:1px solid #eee;border-radius:12px;overflow:hidden;width:140px;">
509
- <img src="data:image/jpeg;base64,{img_b64}" style="width:100%;height:150px;object-fit:cover;display:block;"/>
510
- <div style="padding:8px 10px;">
511
- <div style="font-size:10px;color:#9ca3af;">{sim}</div>
512
- <a href="{link}" target="_blank" style="display:block;text-align:center;background:#111827;
513
- color:white;border-radius:6px;padding:6px;text-decoration:none;font-size:11px;margin-top:6px;">
514
- Shop ↗
515
- </a>
516
- </div>
517
- </div>
518
- """)
519
  return (
520
- '<div style="font-size:11px;letter-spacing:1.5px;color:#9ca3af;font-weight:700;'
521
- 'margin:18px 0 8px;">MORE MATCHES LIKE THIS</div>'
522
- f'<div style="display:flex;gap:12px;flex-wrap:wrap;">{"".join(cards)}</div>'
523
- )
524
-
525
-
526
- # ---------------------------------------------------------------------------
527
- # MAIN PIPELINE — shared by both input paths, runs all 3 GenAI patterns
528
- # ---------------------------------------------------------------------------
529
- def run_pipeline(matched_indices, matched_rows, matched_scores, base_image=None, question=None,
530
- expected_gender=None):
531
- if len(matched_indices) == 0:
532
- return "<i>No matches found — try different filters.</i>", None, "", "<div></div>"
533
-
534
- similarity = [round(1.0 - (d / 2.0), 3) for d in matched_scores]
535
-
536
- top_row = matched_rows.iloc[0]
537
- # Pattern 3 needs a base photo: the user's own photo (Path A) or the top catalog match (Path B)
538
- edit_base_image = base_image if base_image is not None else images[matched_indices[0]]
539
- # If no explicit gender filter was given, fall back to the matched row's own gender, so
540
- # the generated outfit's wording always matches the person actually shown in the photo.
541
- gender_for_generation = expected_gender or top_row["gender"]
542
-
543
- caption = generate_stylist_caption(top_row) # Pattern 1
544
- new_image, _prompt = generate_new_outfit_image(
545
- edit_base_image, top_row, target_gender=gender_for_generation
546
- ) # Pattern 3
547
-
548
- answer = ""
549
- if question:
550
- answer = answer_question_about_image(edit_base_image, question) # Pattern 2
551
-
552
- style_card_html = build_style_card_html(top_row, caption)
553
- outfit_cards_html = (
554
- build_outfit_component_cards_html(top_row)
555
- + build_more_matches_html(matched_indices, matched_rows, similarity)
556
  )
557
- return style_card_html, new_image, answer, outfit_cards_html
558
-
559
-
560
- def recommend_from_photo(photo, gender, age_group, question):
561
- if photo is None:
562
- return "<i>Please upload a photo or pick a Quick Starter.</i>", None, "", "<div></div>"
563
- query_emb = embed_query_image(photo)
564
- idx, rows, scores = faiss_filtered_search(query_emb, gender=gender or None, age_group=age_group or None)
565
- return run_pipeline(idx, rows, scores, base_image=photo, question=question, expected_gender=gender)
566
 
567
-
568
- def recommend_from_features(skin_tone, undertone, style, gender, age_group, question):
569
- sentence = build_feature_sentence(skin_tone, undertone, style, gender, age_group)
570
- query_emb = embed_query_text(sentence)
571
- idx, rows, scores = faiss_filtered_search(query_emb, gender=gender or None, age_group=age_group or None)
572
- return run_pipeline(idx, rows, scores, base_image=None, question=question, expected_gender=gender)
573
-
574
-
575
- # ---------------------------------------------------------------------------
576
- # GRADIO UI
577
- # ---------------------------------------------------------------------------
578
- CUSTOM_CSS = """
579
- @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800&display=swap');
580
- .gradio-container {font-family: 'Inter', 'Helvetica Neue', Arial, sans-serif !important;
581
- max-width: 820px !important; margin: auto !important; background: #FAF5F8 !important;}
582
- .tab-nav button {font-size: 15px !important; font-weight: 600 !important; padding: 12px 24px !important;}
583
- .tab-nav button.selected {color: #C94A72 !important; border-bottom: 2px solid #C94A72 !important;}
584
- #find-btn, .big-btn {background: #1a1a1a !important; color: #fff !important; border: none !important;
585
- border-radius: 12px !important; font-weight: 700 !important; font-size: 15px !important;
586
- padding: 13px !important; transition: background .2s !important;}
587
- #find-btn:hover, .big-btn:hover {background: #C94A72 !important;}
588
- .dark-panel {background: #1a1a1a !important; border-radius: 16px !important; padding: 18px !important;}
589
- .dark-panel label, .dark-panel span {color: #f0e6ea !important;}
590
- footer {display: none !important;}
 
 
 
 
 
 
 
591
  """
592
 
593
- with gr.Blocks(title="AI Fashion Stylist", css=CUSTOM_CSS, theme=gr.themes.Soft(primary_hue="purple")) as demo:
594
- gr.Markdown(
595
- "# 👗 AI Fashion Stylist\n"
596
- "Upload a photo **or** describe your style — get real catalog matches, shop links, "
597
- "a brand-new AI-generated full-look photo, and a personal stylist note."
 
 
 
 
 
 
 
 
 
 
598
  )
599
 
600
- with gr.Tab("📸 Upload a Photo"):
601
- with gr.Row():
602
- photo_in = gr.Image(type="pil", label="Your photo")
603
- with gr.Column():
604
- gender_a = gr.Dropdown(GENDERS, label="Gender (optional)")
605
- age_a = gr.Dropdown(AGE_GROUPS, label="Age group (optional)")
606
- question_a = gr.Textbox(label="Ask the stylist a question about your photo (optional)",
607
- placeholder="e.g. What style would suit me best?")
608
- btn_a = gr.Button("Get My Look ✨", elem_id="find-btn", variant="primary")
609
- style_card_a = gr.HTML(label="Style Profile")
610
- new_img_a = gr.Image(label="✨ Your New AI-Generated Look")
611
- answer_a = gr.Textbox(label="Answer to your question")
612
- outfit_cards_a = gr.HTML(label="Outfit Picks")
613
- btn_a.click(
614
- recommend_from_photo,
615
- [photo_in, gender_a, age_a, question_a],
616
- [style_card_a, new_img_a, answer_a, outfit_cards_a],
617
- )
618
-
619
- gr.Examples(
620
- examples=[[p, None, None, "What style would suit me best?"] for p in SAMPLE_PHOTOS],
621
- inputs=[photo_in, gender_a, age_a, question_a],
622
- outputs=[style_card_a, new_img_a, answer_a, outfit_cards_a],
623
- fn=recommend_from_photo,
624
- cache_examples=True, # precomputed at Space startup -> truly "1-click and see a result"
625
- label="Quick Starters",
626
- )
627
-
628
- with gr.Tab("🎨 Choose Manually"):
629
- gr.Markdown("Select your skin tone, undertone and style below.")
630
- with gr.Group(elem_classes="dark-panel"):
631
- with gr.Row():
632
- skin_b = gr.Dropdown(SKIN_TONES, label="Skin tone", value=SKIN_TONES[0])
633
- undertone_b = gr.Dropdown(UNDERTONES, label="Undertone", value=UNDERTONES[0])
634
- style_b = gr.Dropdown(STYLES, label="Style preference", value=STYLES[0])
635
- with gr.Row():
636
- gender_b = gr.Dropdown(GENDERS, label="Gender", value=GENDERS[0])
637
- age_b = gr.Dropdown(AGE_GROUPS, label="Age group", value=AGE_GROUPS[0])
638
- question_b = gr.Textbox(label="Ask the stylist a question about the top match (optional)",
639
- placeholder="e.g. Is this outfit formal or casual?")
640
- btn_b = gr.Button("Get My Look ✨", elem_id="find-btn", elem_classes="big-btn", variant="primary")
641
- features_recap_b = gr.HTML(build_features_recap_html(SKIN_TONES[0], UNDERTONES[0], STYLES[0]))
642
- for _dropdown in (skin_b, undertone_b, style_b):
643
- _dropdown.change(
644
- build_features_recap_html, [skin_b, undertone_b, style_b], features_recap_b
645
- )
646
- style_card_b = gr.HTML(label="Style Profile")
647
- new_img_b = gr.Image(label="✨ Your New AI-Generated Look")
648
- answer_b = gr.Textbox(label="Answer to your question")
649
- outfit_cards_b = gr.HTML(label="Outfit Picks")
650
- btn_b.click(
651
- recommend_from_features,
652
- [skin_b, undertone_b, style_b, gender_b, age_b, question_b],
653
- [style_card_b, new_img_b, answer_b, outfit_cards_b],
654
- )
655
-
656
- gr.Examples(
657
- examples=[
658
- ["deep", "cool", "boho", "woman", "young adult", "Is this outfit formal or casual?"],
659
- ["tan", "warm", "elegant", "man", "adult", "What occasion suits this outfit?"],
660
- ["fair", "neutral", "minimalist", "woman", "teen", "What season is this outfit best for?"],
661
- ],
662
- inputs=[skin_b, undertone_b, style_b, gender_b, age_b, question_b],
663
- outputs=[style_card_b, new_img_b, answer_b, outfit_cards_b],
664
- fn=recommend_from_features,
665
- cache_examples=True, # precomputed at Space startup -> truly "1-click and see a result"
666
- label="Quick Starters",
667
- )
668
 
669
  if __name__ == "__main__":
670
  demo.launch()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  import gradio as gr
2
+ import numpy as np
 
3
  from PIL import Image
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4
 
5
+ # ==========================================
6
+ # 1. CORE DATA & ANALYSIS PIPELINE
7
+ # ==========================================
8
+
9
+ COLOR_PALETTES = {
10
+ "Fair-Blonde-Blue": {
11
+ "name": "Icy Cool Blonde",
12
+ "description": "Fair skin · Blonde hair · Blue eyes",
13
+ "swatches": [
14
+ {"name": "Light Blue", "hex": "#A0C4DF"},
15
+ {"name": "Soft Lavender", "hex": "#C39BD3"},
16
+ {"name": "Ivory", "hex": "#F9F9F6"},
17
+ {"name": "Dusty Rose", "hex": "#E8A7A1"}
18
+ ],
19
+ "tip": "Avoid harsh black — opt for navy or charcoal instead."
20
+ }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
  }
22
 
23
+ # The specific items recommended directly based on user traits
24
+ RECOMMENDED_LOOK = [
25
+ {"category": "TOP", "name": "Powder Blue Linen Blouse", "brand": "Zara", "hex": "#A0C4DF", "url": "https://shopping.google.com"},
26
+ {"category": "TROUSERS", "name": "Soft White Wide-Leg Trousers", "brand": "ASOS", "hex": "#F9F9F6", "url": "https://shopping.google.com"},
27
+ {"category": "SHOES", "name": "Nude Strappy Sandals", "brand": "SHEIN", "hex": "#E5C2B3", "url": "https://shopping.google.com"},
28
+ {"category": "JACKET", "name": "Lilac Knit Cardigan", "brand": "H&M", "hex": "#C39BD3", "url": "https://shopping.google.com"},
29
+ {"category": "BAG", "name": "Dusty Rose Mini Bag", "brand": "SHEIN", "hex": "#E8A7A1", "url": "https://shopping.google.com"},
30
+ {"category": "DRESS", "name": "Lavender Midi Wrap Dress", "brand": "ASOS", "hex": "#D2B4DE", "url": "https://shopping.google.com"}
31
+ ]
32
 
33
+ def analyze_and_recommend(image, skin, hair, eye, choice_method):
34
+ # 1. Detect features based on input method
35
+ if choice_method == "Upload a photo" and image is not None:
36
+ # Connect your SigLIP / feature classification logic here
37
+ skin_tone, hair_color, eye_color = "Fair", "Blonde", "Blue"
38
+ else:
39
+ skin_tone, hair_color, eye_color = skin, hair, eye
40
+
41
+ # 2. Get corresponding color palette archetype
42
+ palette_key = f"{skin_tone}-{hair_color}-{eye_color}"
43
+ palette = COLOR_PALETTES.get(palette_key, COLOR_PALETTES["Fair-Blonde-Blue"])
44
+
45
+ # 3. Build HTML for Color Swatches
46
+ swatches_html = "".join([
47
+ f'<div class="swatch-card"><div class="color-bubble" style="background-color: {s["hex"]};"></div><p>{s["name"]}</p></div>'
48
+ for s in palette["swatches"]
49
+ ])
50
+
51
+ # 4. Build HTML for Outfit Picks (mapping your FAISS database results here)
52
+ outfit_html = "".join([
53
+ f'''
54
+ <div class="product-card">
55
+ <div class="prod-bubble" style="background-color: {item["hex"]};"></div>
56
+ <div class="prod-meta">
57
+ <span class="prod-cat">{item["category"]}</span>
58
+ <p class="prod-title">{item["name"]}</p>
59
+ <span class="prod-brand">{item["brand"]}</span>
60
+ <a href="{item["url"]}" target="_blank" class="shop-btn">Shop ↗</a>
61
+ </div>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  </div>
63
+ ''' for item in RECOMMENDED_LOOK
64
+ ])
65
+
66
+ palette_results = f"""
67
+ <div class="results-box">
68
+ <div class="palette-banner">
69
+ <h2>{palette["name"]}</h2>
70
+ <p>{palette["description"]}</p>
71
  </div>
72
+ <div class="section-split">
73
+ <div>
74
+ <h3>COLORS FOR YOU</h3>
75
+ <div class="swatch-grid">{swatches_html}</div>
76
+ </div>
77
+ <div class="tip-box">
78
+ <h3>STYLIST TIP</h3>
79
+ <p>{palette["tip"]}</p>
80
+ </div>
81
  </div>
 
82
  </div>
83
  """
84
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
85
  return (
86
+ gr.update(value=skin_tone),
87
+ gr.update(value=hair_color),
88
+ gr.update(value=eye_color),
89
+ palette_results,
90
+ outfit_html,
91
+ gr.update(visible=True)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
92
  )
 
 
 
 
 
 
 
 
 
93
 
94
+ # ==========================================
95
+ # 2. MATCHING MINIMALIST UI CSS (No Orange)
96
+ # ==========================================
97
+ custom_css = """
98
+ body { background-color: #F9F8F6 !important; font-family: -apple-system, BlinkMacSystemFont, sans-serif; }
99
+ .gradio-container { max-width: 800px !important; margin: 0 auto !important; padding-top: 40px !important; }
100
+ .center-text { text-align: center; }
101
+ .title-main { font-size: 40px; font-weight: 800; color: #111111; margin-bottom: 4px; letter-spacing: -1px; }
102
+ .subtitle-main { font-size: 16px; color: #777777; margin-bottom: 32px; }
103
+ .pill-container { display: flex; gap: 8px; flex-wrap: wrap; margin-top: 10px; }
104
+ .results-box { background: white; border-radius: 16px; padding: 24px; border: 1px solid #EAE9E6; margin-top: 24px; }
105
+ .palette-banner { background: #1A1A1A; color: white; padding: 24px; border-radius: 12px; margin-bottom: 24px; }
106
+ .palette-banner h2 { font-size: 26px; font-weight: 700; margin: 0 0 4px 0; }
107
+ .palette-banner p { color: #A0A0A0; margin: 0; font-size: 14px; }
108
+ .section-split { display: grid; grid-template-columns: 1.2fr 0.8fr; gap: 24px; margin-bottom: 24px; }
109
+ @media (max-width: 768px) { .section-split { grid-template-columns: 1fr; } }
110
+ .swatch-grid { display: flex; gap: 16px; margin-top: 12px; }
111
+ .swatch-card { text-align: center; font-size: 12px; font-weight: 500; color: #555; }
112
+ .color-bubble { width: 52px; height: 52px; border-radius: 50%; margin: 0 auto 8px auto; border: 1px solid rgba(0,0,0,0.05); }
113
+ .tip-box { background: #F4F3F0; border-radius: 12px; padding: 20px; border: 1px solid #EAE9E6; }
114
+ .tip-box h3, .section-split h3 { font-size: 11px; font-weight: 700; color: #A24A70; letter-spacing: 1px; margin-top: 0; }
115
+ .tip-box p { font-size: 14px; color: #333; line-height: 1.5; margin: 8px 0 0 0; }
116
+ .outfit-grid { display: grid; grid-template-columns: repeat(3, 1fr); gap: 16px; margin-top: 16px; }
117
+ @media (max-width: 640px) { .outfit-grid { grid-template-columns: repeat(2, 1fr); } }
118
+ .product-card { background: white; border: 1px solid #EAE9E6; border-radius: 12px; padding: 16px; display: flex; flex-direction: column; align-items: center; text-align: center; }
119
+ .prod-bubble { width: 36px; height: 36px; border-radius: 50%; margin-bottom: 12px; border: 1px solid rgba(0,0,0,0.04); }
120
+ .prod-cat { font-size: 10px; font-weight: 700; color: #A24A70; letter-spacing: 0.5px; }
121
+ .prod-title { font-size: 14px; font-weight: 600; color: #111; margin: 4px 0; min-height: 40px; display: flex; align-items: center; }
122
+ .prod-brand { font-size: 12px; color: #888; margin-bottom: 12px; display: block; }
123
+ .shop-btn { display: block; width: 100%; background: #111111; color: white !important; text-align: center; padding: 8px 0; border-radius: 6px; font-size: 13px; font-weight: 500; text-decoration: none !important; }
124
+ .primary-action-btn { background: #111111 !important; color: white !important; border: none !important; border-radius: 8px !important; font-weight: 600 !important; font-size: 16px !important; padding: 12px 0 !important; }
125
  """
126
 
127
+ # ==========================================
128
+ # 3. GRADIO INTERFACE LAYOUT
129
+ # ==========================================
130
+ with gr.Blocks(css=custom_css) as demo:
131
+
132
+ with gr.Group(elem_classes="center-text"):
133
+ gr.Markdown("PERSONAL COLOR STYLING", elem_classes="prod-cat")
134
+ gr.Markdown("# LookMatch", elem_classes="title-main")
135
+ gr.Markdown("Discover your perfect color palette and outfit", elem_classes="subtitle-main")
136
+
137
+ choice_method = gr.Radio(
138
+ ["Upload a photo", "Choose manually"],
139
+ value="Upload a photo",
140
+ show_label=False,
141
+ interactive=True
142
  )
143
 
144
+ with gr.Row():
145
+ with gr.Column() as upload_col:
146
+ image_input = gr.Image(label="Upload your photo — Gemini AI will detect your features", type="pil")
147
+
148
+ with gr.Column(visible=False) as manual_col:
149
+ skin_dropdown = gr.Dropdown(["Fair", "Light", "Medium", "Olive", "Tan", "Deep"], value="Fair", label="Skin tone")
150
+ hair_dropdown = gr.Dropdown(["Blonde", "Auburn/Red", "Brunette", "Black", "Grey/White"], value="Blonde", label="Hair color")
151
+ eye_dropdown = gr.Dropdown(["Blue", "Green", "Hazel", "Brown", "Dark brown/Black"], value="Blue", label="Eye color")
152
+
153
+ submit_btn = gr.Button("Get my look ✨", elem_classes="primary-action-btn")
154
+
155
+ # Lower Dashboard Area (Revealed on click)
156
+ with gr.Column(visible=False) as output_dashboard:
157
+ gr.Markdown("### YOUR FEATURES")
158
+ with gr.Row(elem_classes="pill-container"):
159
+ skin_tag = gr.Button(value="Fair", size="sm")
160
+ hair_tag = gr.Button(value="Blonde", size="sm")
161
+ eye_tag = gr.Button(value="Blue", size="sm")
162
+
163
+ palette_output = gr.HTML()
164
+
165
+ gr.Markdown("### OUTFIT PICKS")
166
+ outfit_output = gr.HTML(elem_classes="outfit-grid")
167
+
168
+ # Navigation controller
169
+ def toggle_input_ui(choice):
170
+ if choice == "Choose manually":
171
+ return gr.update(visible=False), gr.update(visible=True)
172
+ return gr.update(visible=True), gr.update(visible=False)
173
+
174
+ choice_method.change(toggle_input_ui, inputs=choice_method, outputs=[upload_col, manual_col])
175
+
176
+ submit_btn.click(
177
+ fn=analyze_and_recommend,
178
+ inputs=[image_input, skin_dropdown, hair_dropdown, eye_dropdown, choice_method],
179
+ outputs=[skin_tag, hair_tag, eye_tag, palette_output, outfit_output, output_dashboard]
180
+ )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
181
 
182
  if __name__ == "__main__":
183
  demo.launch()