Upload 6 files

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+esempi/8bit.png filter=lfs diff=lfs merge=lfs -text
+esempi/architetto.png filter=lfs diff=lfs merge=lfs -text
+esempi/popart.png filter=lfs diff=lfs merge=lfs -text
+esempi/rinascimentale.png filter=lfs diff=lfs merge=lfs -text

app.py

@@ -1,189 +1,177 @@
-import os
-import cv2
-import math
+import gradio as gr
 import torch
+import cv2
 import numpy as np
-import gradio as gr
+import os
+from PIL import Image, ImageDraw, ImageFilter
+from diffusers import StableDiffusionXLInpaintPipeline
 import spaces
-from PIL import Image
-
-from huggingface_hub import hf_hub_download, snapshot_download
 from insightface.app import FaceAnalysis
-from diffusers import ControlNetModel
-
-from pipeline_stable_diffusion_xl_instantid import StableDiffusionXLInstantIDPipeline
+import insightface
+from huggingface_hub import hf_hub_download
 
 # ==============================================================================
-# 1. SETUP E INIZIALIZZAZIONE GLOBALE
+# 1. SETUP DELL'AMBIENTE E DOWNLOAD MODELLI
 # ==============================================================================
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-dtype = torch.float16 if torch.cuda.is_available() else torch.float32
-
-def download_antelopev2():
-    """Scarica automaticamente i modelli di InsightFace se non sono presenti."""
-    model_dir = "./models/antelopev2"
-    if not os.path.exists(model_dir) or len(os.listdir(model_dir)) < 5:
-        print("Scaricamento dei modelli antelopev2 in corso...")
-        os.makedirs(model_dir, exist_ok=True)
-        snapshot_download(repo_id="DIAMONIK7777/antelopev2", local_dir=model_dir)
-        print("Scaricamento completato.")
-    return "./"
-
-model_root = download_antelopev2()
-
-face_app = FaceAnalysis(name='antelopev2', root=model_root, providers=['CPUExecutionProvider'])
+# Inizializza l'analizzatore di volti (InsightFace)
+print("Inizializzazione FaceAnalysis...")
+face_app = FaceAnalysis(name='antelopev2', root='./', providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
 face_app.prepare(ctx_id=0, det_size=(640, 640))
 
-# Scarica il file ip-adapter.bin di InstantID
-face_adapter = hf_hub_download(repo_id="InstantX/InstantID", filename="ip-adapter.bin")
-
-# Carica il ControlNet (IdentityNet) di InstantID
-print("Caricamento ControlNet InstantID...")
-controlnet = ControlNetModel.from_pretrained(
-    "InstantX/InstantID",
-    subfolder="ControlNetModel",
-    torch_dtype=dtype,
-    use_safetensors=True
-)
+# Download e caricamento del modello Inswapper
+print("Download e caricamento Inswapper...")
+try:
+    inswapper_path = hf_hub_download(repo_id="ezioruan/inswapper_128.onnx", filename="inswapper_128.onnx")
+    face_swapper = insightface.model_zoo.get_model(inswapper_path, providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
+except Exception as e:
+    print(f"Errore caricamento Inswapper: {e}")
+    face_swapper = None
+
+# Caricamento Pipeline Inpainting SDXL
+print("Caricamento pipeline SDXL Inpaint...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.float16 if torch.cuda.is_available() else "float32"
 
-# Carica la pipeline dedicata InstantID (NON quella generica ControlNet)
-print("Caricamento Pipeline SDXL InstantID...")
-pipe = StableDiffusionXLInstantIDPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-base-1.0",
-    controlnet=controlnet,
+pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
+    "diffusers/stable-diffusion-xl-1.0-inpainting-0.1",
     torch_dtype=dtype,
+    variant="fp16" if torch.cuda.is_available() else None,
     use_safetensors=True
 )
 
-# Carica il Face Adapter con il metodo specifico di InstantID
-print("Caricamento Face Adapter InstantID...")
-pipe.load_ip_adapter_instantid(face_adapter)
-pipe.to(device)
-
-print("Pipeline pronta!")
-
+# Rimuovi il limitatore di sicurezza se crea problemi (opzionale)
+pipe.watermark = None 
 
 # ==============================================================================
-# 2. CONFIGURAZIONE STILI
+# 2. CONFIGURAZIONE STILI E TEMPLATE
 # ==============================================================================
-
+# I template sono le immagini pre-generate fornite dall'utente.
 STYLES = {
     "Stile 1: Bit-Builder (8-Bit/Pixel Art)": {
+        "template": "esempi/8bit.png",
         "repo_id": "nerijs/pixel-art-xl",
         "weight_name": "pixel-art-xl.safetensors",
-        "prompt": "pixelart style, Pixel art portrait of a tech builder, indie hacker typing on a retro mechanical keyboard. He is wearing a jacket with a small green, white, and red square patch on the chest. Cyberpunk desk setup, glowing monitors displaying code. High quality, 8-bit game style.",
-        "negative_prompt": "ugly, deformed, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, watermark, text"
+        "prompt": "pixelart style, 8-bit game style face, highly detailed, perfect pixel art shading, 8-bit retro aesthetic, matching lighting",
+        "negative_prompt": "ugly, deformed, lowres, realistic, photographic, 3d render, photo, photorealistic"
     },
     "Stile 2: Il Costruttore Rinascimentale": {
+        "template": "esempi/rinascimentale.png",
         "repo_id": "davidmoref/sdxl-lora-adapter-renaissance",
         "weight_name": "pytorch_lora_weights.safetensors",
-        "prompt": "renaissance oil painting, A masterpiece renaissance oil painting of an Italian genius inventor in the style of Leonardo da Vinci. The man is holding a wooden prototype. He is wearing elegant renaissance dark clothes with a subtle green, white and red enameled pin on the lapel. Architectural blueprints on a rustic wooden table. Chiaroscuro lighting.",
-        "negative_prompt": "ugly, deformed, modern, photography, bad anatomy, worst quality, low quality, watermark, text"
+        "prompt": "renaissance oil painting, oil painting face, elegant brush strokes, chiaroscuro lighting, classic masterpiece, perfectly integrated face",
+        "negative_prompt": "ugly, deformed, modern, photography, bad anatomy, worst quality, low quality, pixel art, cartoon"
     },
-    "Stile 3: L'Architetto del Futuro (Design Geometrico)": {
-        "repo_id": "goofyai/flat_illustration_sdxl",
-        "weight_name": "flat_illustration_sdxl.safetensors",
-        "prompt": "flat vector art, Modern flat vector illustration of a tech architect holding a smartphone. He wears a minimalist grey suit jacket featuring a small geometric green, white, and red enamel badge. Clean geometric lines, industrial design background, minimalistic.",
-        "negative_prompt": "ugly, deformed, photorealistic, 3d render, bad anatomy, worst quality, low quality, watermark, text"
-    },
-    "Stile 4: Genio Creativo (Pop-Art)": {
+    "Stile 3: Genio Creativo (Pop-Art)": {
+        "template": "esempi/popart.png",
         "repo_id": None,
         "weight_name": None,
-        "prompt": "vintage pop art, Comic book style pop-art portrait of a creative maker. Vibrant colors. He wears a trendy jacket adorned with an 'Italian Builders' custom patch in green, white and red. Halftone dots, vintage poster aesthetic.",
-        "negative_prompt": "ugly, deformed, photorealistic, bad anatomy, worst quality, low quality, watermark, extra text"
+        "prompt": "vintage pop art, Comic book style pop-art face, vibrant colors, halftone dots, graphic illustration, bold lines",
+        "negative_prompt": "ugly, deformed, photorealistic, realistic photography, oil painting, 3d render"
     }
 }
 
 # ==============================================================================
-# 3. PIPELINE DI GENERAZIONE
+# 3. FUNZIONI CORE DELL'APPLICAZIONE
 # ==============================================================================
 
-def draw_kps(image_pil, kps, color_list=[(255,0,0), (0,255,0), (0,0,255), (255,255,0), (255,0,255)]):
-    """Genera l'immagine dei keypoints per la ControlNet di InstantID."""
-    stickwidth = 4
-    limbSeq = np.array([[0, 2], [1, 2], [3, 2], [4, 2]])
-    kps = np.array(kps)
-    w, h = image_pil.size
-    out_img = np.zeros([h, w, 3])
-
-    for i in range(len(limbSeq)):
-        index = limbSeq[i]
-        color = color_list[index[0]]
-        x = kps[index][:, 0]
-        y = kps[index][:, 1]
-        length = ((x[0] - x[1]) ** 2 + (y[0] - y[1]) ** 2) ** 0.5
-        angle = math.degrees(math.atan2(y[0] - y[1], x[0] - x[1]))
-        polygon = cv2.ellipse2Poly((int(np.mean(x)), int(np.mean(y))), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
-        cv2.fillConvexPoly(out_img, polygon, color)
-    out_img = (out_img * 0.6).astype(np.uint8)
-
-    for idx, kp in enumerate(kps):
-        color = color_list[idx]
-        x, y = kp
-        cv2.circle(out_img, (int(x), int(y)), 10, color, -1)
-
-    return Image.fromarray(out_img.astype(np.uint8))
-
-@spaces.GPU(duration=120)
+def create_face_mask(image, bbox, expansion=0.2, blur_radius=15):
+    """
+    Crea una maschera sfumata attorno al volto per l'inpainting.
+    """
+    mask = Image.new("L", image.size, 0)
+    draw = ImageDraw.Draw(mask)
+    
+    x1, y1, x2, y2 = bbox
+    w = x2 - x1
+    h = y2 - y1
+    
+    # Espandi leggermente la bounding box
+    nx1 = max(0, x1 - w * expansion)
+    ny1 = max(0, y1 - h * expansion)
+    nx2 = min(image.width, x2 + w * expansion)
+    ny2 = min(image.height, y2 + h * (expansion * 1.5)) # Espandi un po' di più verso il basso per il mento
+    
+    # Disegna un ovale
+    draw.ellipse([nx1, ny1, nx2, ny2], fill=255)
+    
+    # Sfuma i bordi per fondere meglio
+    mask = mask.filter(ImageFilter.GaussianBlur(blur_radius))
+    return mask
+
+@spaces.GPU(duration=60)
 def generate_avatar(user_image, selected_style):
     if user_image is None:
         raise gr.Error("Per favore, carica un'immagine del volto.")
-
-    # 1. Analisi del volto
-    cv_image = cv2.cvtColor(np.array(user_image), cv2.COLOR_RGB2BGR)
-    faces = face_app.get(cv_image)
-    if len(faces) == 0:
-        raise gr.Error("Nessun volto rilevato. Riprova con un'altra foto in cui il volto sia ben visibile.")
-
-    face = sorted(faces, key=lambda x: (x.bbox[2]-x.bbox[0])*(x.bbox[3]-x.bbox[1]))[-1]
-    face_emb = face.normed_embedding
-    face_kps = face.kps
-
-    # Ridimensioniamo l'immagine utente a 1024x1024 (standard SDXL)
-    user_image_resized = user_image.resize((1024, 1024))
-    faces_resized = face_app.get(cv2.cvtColor(np.array(user_image_resized), cv2.COLOR_RGB2BGR))
-    if len(faces_resized) > 0:
-        face_kps = sorted(faces_resized, key=lambda x: (x.bbox[2]-x.bbox[0])*(x.bbox[3]-x.bbox[1]))[-1].kps
-
-    kps_image = draw_kps(user_image_resized, face_kps)
-
-    # 2. Gestione LoRA
+        
+    if face_swapper is None:
+        raise gr.Error("Modello Inswapper non caricato correttamente.")
+
+    # 1. Analisi del volto utente
+    cv_user_image = cv2.cvtColor(np.array(user_image), cv2.COLOR_RGB2BGR)
+    user_faces = face_app.get(cv_user_image)
+    if len(user_faces) == 0:
+        raise gr.Error("Nessun volto rilevato nella tua foto. Riprova con un selfie più chiaro.")
+    
+    # Prendi il volto più grande
+    user_face = sorted(user_faces, key=lambda x: (x.bbox[2]-x.bbox[0])*(x.bbox[3]-x.bbox[1]))[-1]
+
+    # 2. Caricamento e analisi del Template
     style_config = STYLES[selected_style]
+    template_path = style_config["template"]
+    
+    if not os.path.exists(template_path):
+        raise gr.Error(f"Errore: il file di template {template_path} non esiste.")
+        
+    template_pil = Image.open(template_path).convert("RGB")
+    cv_template_image = cv2.cvtColor(np.array(template_pil), cv2.COLOR_RGB2BGR)
+    
+    template_faces = face_app.get(cv_template_image)
+    if len(template_faces) == 0:
+        raise gr.Error("Impossibile trovare un volto nel template di base.")
+    template_face = sorted(template_faces, key=lambda x: (x.bbox[2]-x.bbox[0])*(x.bbox[3]-x.bbox[1]))[-1]
+
+    # 3. Swap del Volto (Fotorealistico)
+    print("Eseguendo Face Swap...")
+    swapped_cv = face_swapper.get(cv_template_image, template_face, user_face, paste_back=True)
+    swapped_pil = Image.fromarray(cv2.cvtColor(swapped_cv, cv2.COLOR_BGR2RGB))
+    
+    # 4. Creazione Maschera per l'Inpainting
+    mask_pil = create_face_mask(template_pil, template_face.bbox)
+    
+    # 5. Stylization tramite Inpainting
     repo_id = style_config["repo_id"]
-
+    
+    # Carica il LoRA dello stile se necessario
     if repo_id:
         try:
             print(f"Scaricamento/Caricamento LoRA {repo_id}...")
             pipe.load_lora_weights(repo_id, weight_name=style_config["weight_name"])
         except Exception as e:
             print(f"Errore caricamento LoRA da HF: {e}")
-
-    # 3. Generazione immagine con InstantID
-    prompt = style_config["prompt"]
-    negative_prompt = style_config["negative_prompt"]
-
-    print("Inizio generazione InstantID...")
-    pipe.set_ip_adapter_scale(0.8)
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        image_embeds=face_emb,
-        image=kps_image,
-        controlnet_conditioning_scale=0.8,
+            
+    # Sposta la pipeline su GPU per l'inferenza
+    pipe.to(device)
+
+    print("Inizio Inpainting per fondere lo stile...")
+    stylized_image = pipe(
+        prompt=style_config["prompt"],
+        negative_prompt=style_config["negative_prompt"],
+        image=swapped_pil,
+        mask_image=mask_pil,
+        strength=0.35, # Bassa strength: mantiene forte l'identità, applica solo lo stile superficiale
+        guidance_scale=7.5,
         num_inference_steps=30,
-        guidance_scale=5.0,
     ).images[0]
 
+    # Scarica il LoRA per evitare conflitti al prossimo giro
     if repo_id:
         try:
             pipe.unload_lora_weights()
         except:
             pass
 
-
-
-    return image
+    return stylized_image
 
 # ==============================================================================
 # 4. INTERFACCIA GRADIO (FRONTEND)
@@ -205,11 +193,12 @@ theme = gr.themes.Soft(primary_hue="blue", secondary_hue="emerald").set(
 with gr.Blocks(css=custom_css, theme=theme) as demo:
     with gr.Column(elem_classes="container"):
         gr.HTML("<h1 class='header-title'>🇮🇹 Italian Builders Avatar Generator 🇮🇹</h1>")
-        gr.HTML("<p class='header-subtitle'>Trasforma il tuo selfie. Il tuo brand, la tua identità.</p>")
+        gr.HTML("<p class='header-subtitle'>Trasforma il tuo selfie e unisciti alla community.</p>")
 
         with gr.Row():
             with gr.Column(scale=1):
                 input_image = gr.Image(type="pil", label="1. Carica il tuo Selfie")
+
             with gr.Column(scale=1):
                 style_selector = gr.Radio(
                     choices=list(STYLES.keys()),
@@ -220,7 +209,7 @@ with gr.Blocks(css=custom_css, theme=theme) as demo:
                 generate_button = gr.Button("Genera il tuo Avatar 🚀", elem_classes="generate-btn", variant="primary")
 
             with gr.Column(scale=1):
-                output_image = gr.Image(label="Il tuo Avatar Builder", interactive=False)
+                output_image = gr.Image(label="Il tuo Avatar", interactive=False)
 
         generate_button.click(
             fn=generate_avatar,