Update app.py

app.py

@@ -1,12 +1,12 @@
-# app.py — InstantID SDXL + (optionnel) IP-Adapter Style (2D total)
-# Hugging Face Space – prêt à déployer
+# app.py — InstantID SDXL (officiel) + IP-Adapter Style (optionnel, rendu 2D)
+# Hugging Face Space ready
 
-# 0) Environnement AVANT TOUT IMPORT
+# 0) Environnement AVANT imports
 import os, sys
-os.environ["OMP_NUM_THREADS"] = "4"  # safe pour libgomp
+os.environ["OMP_NUM_THREADS"] = "4"
 os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
 
-# rendre importable ./instantid (où se trouve pipeline_stable_diffusion_xl_instantid_full.py)
+# rendre importable ./instantid (pipeline officielle à placer ici)
 sys.path.insert(0, os.path.abspath("./instantid"))
 
 # 1) Imports
@@ -16,22 +16,27 @@ from PIL import Image, ImageOps, ImageDraw
 from huggingface_hub import hf_hub_download
 from diffusers.models import ControlNetModel
 
+# InsightFace
+from insightface.app import FaceAnalysis
+
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 DTYPE  = torch.float16 if DEVICE == "cuda" else torch.float32
 
-# 2) Chemins & Hub
-ASSETS_REPO = "InstantX/InstantID"  # poids InstantID: IdentityNet + ip-adapter instantid
+# 2) Chemins & Hub (poids InstantID officiels + IP-Adapter Style SDXL)
+ASSETS_REPO = "InstantX/InstantID"
 CHECKPOINTS_DIR  = "./checkpoints"
 CN_LOCAL_DIR     = os.path.join(CHECKPOINTS_DIR, "ControlNetModel")
 IP_ADAPTER_LOCAL = os.path.join(CHECKPOINTS_DIR, "ip-adapter.bin")
 
-# IP-Adapter Style (SDXL) — pour forcer le rendu 2D
 IP_STYLE_REPO      = "h94/IP-Adapter"
 IP_STYLE_SUBFOLDER = "sdxl_models"
 IP_STYLE_WEIGHT    = "ip-adapter_sdxl.bin"
-IP_STYLE_LOCAL     = os.path.join(CHECKPOINTS_DIR, "ip-adapter_sdxl.bin")
 
-# 3) Téléchargements sûrs (détecte fichiers vides)
+# Modèle de base : SDXL (tu peux remplacer par un checkpoint stylé anime/one-piece-like)
+BASE_MODEL = "stabilityai/stable-diffusion-xl-base-1.0"
+# Exemple alternatif (anime) : "wangqixun/YamerMIX_v8"
+
+# 3) Téléchargements sûrs
 def safe_download(repo, filename, local_dir, min_bytes, label, subfolder=None):
     os.makedirs(local_dir, exist_ok=True)
     local_path = os.path.join(local_dir, os.path.basename(filename))
@@ -60,16 +65,11 @@ def ensure_assets_or_download():
     # IdentityNet (ControlNet) + ip-adapter (InstantID)
     safe_download(ASSETS_REPO, "ControlNetModel/config.json", CHECKPOINTS_DIR, 1_000, "IdentityNet config")
     safe_download(ASSETS_REPO, "ControlNetModel/diffusion_pytorch_model.safetensors", CHECKPOINTS_DIR, 100_000_000, "IdentityNet weights")
-    safe_download(ASSETS_REPO, "ip-adapter.bin", CHECKPOINTS_DIR, 100_000_000, "IP-Adapter InstantID")
-    # IP-Adapter Style (SDXL)
-    p = safe_download(IP_STYLE_REPO, IP_STYLE_WEIGHT, CHECKPOINTS_DIR, 20_000_000, "IP-Adapter Style (SDXL)", subfolder=IP_STYLE_SUBFOLDER)
-    try:
-        if not os.path.exists(IP_STYLE_LOCAL):
-            import shutil; shutil.copy2(p, IP_STYLE_LOCAL)
-    except Exception as e:
-        print(f"ℹ️ Copie locale IP-Adapter Style ignorée: {e}")
+    safe_download(ASSETS_REPO, "ip-adapter.bin", CHECKPOINTS_DIR, 100_000_000, "IP-Adapter (InstantID)")
+    # IP-Adapter Style (SDXL) — optionnel
+    safe_download(IP_STYLE_REPO, IP_STYLE_WEIGHT, CHECKPOINTS_DIR, 20_000_000, "IP-Adapter Style (SDXL)", subfolder=IP_STYLE_SUBFOLDER)
 
-# 4) Import dynamique de la pipeline InstantID SDXL (officielle)
+# 4) Import dynamique de la pipeline SDXL officielle
 def import_pipeline_or_fail():
     candidates = [
         "./instantid/pipeline_stable_diffusion_xl_instantid_full.py",
@@ -77,14 +77,13 @@ def import_pipeline_or_fail():
     ]
     pipeline_file = next((p for p in candidates if os.path.exists(p)), None)
     if pipeline_file is None:
-        raise RuntimeError("❌ Fichier pipeline manquant.\nPlace `pipeline_stable_diffusion_xl_instantid_full.py` dans ./instantid/")
+        raise RuntimeError("❌ Pipeline manquante. Place `pipeline_stable_diffusion_xl_instantid_full.py` dans ./instantid/")
     if os.path.getsize(pipeline_file) < 1024:
-        raise RuntimeError("❌ Fichier pipeline trop petit (vide ?). Colle la version SDXL officielle.")
+        raise RuntimeError("❌ Pipeline trop petite (vide ?). Utilise la version SDXL officielle.")
 
     spec = importlib.util.spec_from_file_location("instantid_pipeline", pipeline_file)
     mod  = importlib.util.module_from_spec(spec)
     spec.loader.exec_module(mod)
-    # Chercher la classe SDXL officielle
     for name, obj in vars(mod).items():
         if isinstance(obj, type) and "InstantID" in name and hasattr(obj, "from_pretrained"):
             print(f"✅ Pipeline trouvée : {name}")
@@ -92,7 +91,7 @@ def import_pipeline_or_fail():
     avail = [n for n, o in vars(mod).items() if isinstance(o, type)]
     raise RuntimeError("❌ Aucune classe pipeline InstantID trouvée. Classes dispo: " + ", ".join(avail))
 
-# 5) draw_kps local (remplace la dépendance draw_kps du repo)
+# 5) util — dessin landmarks (kps)
 def draw_kps_local(img_pil, kps):
     w, h = img_pil.size
     out = Image.new("RGB", (w, h), "white")
@@ -102,30 +101,30 @@ def draw_kps_local(img_pil, kps):
         d.ellipse((x - r, y - r, x + r, y + r), fill="black")
     return out
 
-# 6) Chargement pipeline (ControlNet = objet unique)
+# 6) Chargement pipeline
 load_logs = []
 HAS_STYLE_ADAPTER = False
+
 try:
+    # a) pipeline
     SDXLInstantID = import_pipeline_or_fail()
     ensure_assets_or_download()
 
-    BASE_MODEL = "stabilityai/stable-diffusion-xl-base-1.0"
-    load_logs.append(f"Chargement base: {BASE_MODEL}")
-
+    # b) IdentityNet (ControlNet)
     controlnet_identitynet = ControlNetModel.from_pretrained(CN_LOCAL_DIR, torch_dtype=DTYPE)
 
     pipe = SDXLInstantID.from_pretrained(
         BASE_MODEL,
-        controlnet=controlnet_identitynet,   # objet unique
+        controlnet=controlnet_identitynet,              # objet unique
         torch_dtype=DTYPE,
         safety_checker=None,
         feature_extractor=None,
     ).to(DEVICE)
 
-    # Charger l’IP-Adapter InstantID (identité)
+    # c) IP-Adapter InstantID (identité)
     pipe.load_ip_adapter_instantid(IP_ADAPTER_LOCAL)
 
-    # Charger (optionnel) un IP-Adapter Style SDXL, nommé "style"
+    # d) IP-Adapter Style SDXL (optionnel), nommé "style"
     try:
         pipe.load_ip_adapter(
             IP_STYLE_REPO,
@@ -137,8 +136,8 @@ try:
         HAS_STYLE_ADAPTER = True
     except Exception as e:
         load_logs.append(f"ℹ️ IP-Adapter Style non chargé: {e}")
-        HAS_STYLE_ADAPTER = False
 
+    # e) devices
     if DEVICE == "cuda":
         if hasattr(pipe, "image_proj_model"): pipe.image_proj_model.to("cuda")
         if hasattr(pipe, "unet"):             pipe.unet.to("cuda")
@@ -151,8 +150,7 @@ except Exception:
 if pipe is None:
     raise RuntimeError("Échec de chargement du pipeline.\n" + "\n".join(load_logs))
 
-# 7) InsightFace (robuste : antelopev2 → buffalo_l)
-from insightface.app import FaceAnalysis
+# 7) InsightFace (comme le Space officiel : CPUExecutionProvider)
 def load_face_analyser():
     errors = []
     for name in ("antelopev2", "buffalo_l"):
@@ -168,7 +166,6 @@ def load_face_analyser():
 
 fa = load_face_analyser()
 
-# — util pour extraire embedding visage + landmarks (kps) depuis la photo
 def extract_face_embed_and_kps(pil_img):
     import numpy as np, cv2
     img_cv2 = cv2.cvtColor(np.array(pil_img.convert("RGB")), cv2.COLOR_RGB2BGR)
@@ -176,15 +173,11 @@ def extract_face_embed_and_kps(pil_img):
     if not faces:
         raise ValueError("Aucun visage détecté dans la photo.")
     face = faces[-1]
-    face_emb = face["embedding"]              # <— Embedding InsightFace attendu par la pipeline SDXL officielle
+    face_emb = face["embedding"]              # ← attendu par la pipeline officielle
     kps_img  = draw_kps_local(pil_img, face["kps"])
-    # Convertir en torch tensor si besoin (la pipeline accepte souvent ndarray directement)
-    if isinstance(face_emb, (list, tuple)):
-        import numpy as np
-        face_emb = np.array(face_emb)
     return face_emb, kps_img
 
-# 8) Génération (Option A : on passe l’embedding InsightFace -> image_embeds)
+# 8) Génération
 def generate(face_image, style_image, prompt, negative_prompt,
              identity_strength, adapter_strength, style_strength,
              steps, cfg, width, height, seed):
@@ -194,15 +187,15 @@ def generate(face_image, style_image, prompt, negative_prompt,
 
         gen = None if seed is None or int(seed) < 0 else torch.Generator(device=DEVICE).manual_seed(int(seed))
 
-        # Préparer visage carré (512) pour détection consistante
+        # visage → carré 512 pour détection stable
         face = ImageOps.exif_transpose(face_image).convert("RGB")
         ms = min(face.size); x = (face.width - ms) // 2; y = (face.height - ms) // 2
         face_sq = face.crop((x, y, x + ms, y + ms)).resize((512, 512), Image.Resampling.LANCZOS)
 
-        # Embedding InsightFace + landmarks (kps)
+        # InsightFace : embedding + landmarks
         face_emb, kps_img = extract_face_embed_and_kps(face_sq)
 
-        # Régler l’échelle des IP-Adapters (identité & style)
+        # IP-Adapter scales
         try:
             if HAS_STYLE_ADAPTER and style_image is not None:
                 pipe.set_ip_adapter_scale({"instantid": float(adapter_strength), "style": float(style_strength)})
@@ -211,7 +204,7 @@ def generate(face_image, style_image, prompt, negative_prompt,
         except Exception as e:
             print(f"ℹ️ set_ip_adapter_scale ignoré: {e}")
 
-        # Compat multi-ControlNet
+        # compat multi-ControlNet (même si on en a qu’un)
         cn = getattr(pipe, "controlnet", None)
         if isinstance(cn, (list, tuple)):
             n_cn = len(cn)
@@ -223,12 +216,12 @@ def generate(face_image, style_image, prompt, negative_prompt,
         scale_val = float(identity_strength)
         scale_arg = [scale_val] * n_cn if n_cn > 1 else ([scale_val] if isinstance(cn, (list, tuple)) else scale_val)
 
-        # Préparer kwargs (NOTE: image_embeds = face_emb)
+        # kwargs d’inférence
         gen_kwargs = dict(
-            prompt=prompt.strip(),
+            prompt=(prompt or "").strip(),
             negative_prompt=(negative_prompt or "").strip(),
-            image=image_arg,                             # IdentityNet (landmarks)
-            image_embeds=face_emb,                       # <— embedding InsightFace
+            image=image_arg,                         # IdentityNet (landmarks)
+            image_embeds=face_emb,                   # ← embedding InsightFace
             controlnet_conditioning_scale=scale_arg,
             num_inference_steps=int(steps),
             guidance_scale=float(cfg),
@@ -237,7 +230,7 @@ def generate(face_image, style_image, prompt, negative_prompt,
             generator=gen,
         )
 
-        # Fournir l’image de style à l’IP-Adapter Style si dispo
+        # passer l’image de style à l’IP-Adapter Style (si dispo + fournie)
         if HAS_STYLE_ADAPTER and style_image is not None:
             try:
                 gen_kwargs["ip_adapter_image"] = ImageOps.exif_transpose(style_image).convert("RGB")
@@ -252,31 +245,31 @@ def generate(face_image, style_image, prompt, negative_prompt,
     except Exception:
         return None, "Erreur:\n" + traceback.format_exc(), "\n".join(load_logs)
 
-# 9) UI
+# 9) UI (One Piece-friendly par défaut, mais neutre)
 EX_PROMPT = (
-    "one piece style, Eiichiro Oda style, anime portrait, upper body, pirate outfit, straw hat, "
-    "clean lineart, cel shading, vibrant colors, expressive eyes, symmetrical face, looking at camera, "
-    "dynamic composition, simple background"
+    "one piece style, Eiichiro Oda style, anime portrait, upper body, pirate outfit, "
+    "clean lineart, cel shading, vibrant colors, expressive eyes, dynamic composition, simple background"
 )
 EX_NEG = (
     "realistic, photo, photorealistic, skin pores, complex lighting, "
     "low quality, worst quality, lowres, blurry, noisy, watermark, text, logo, jpeg artifacts, "
-    "bad anatomy, distorted eyes, deformed, multiple faces, nsfw"
+    "bad anatomy, deformed, multiple faces, nsfw"
 )
 
 with gr.Blocks(css="footer{display:none !important}") as demo:
-    gr.Markdown("# 🏴‍☠️ One Piece – InstantID SDXL + IP-Adapter Style (2D total) — Option A officielle")
+    gr.Markdown("# 🏴‍☠️ InstantID SDXL + IP-Adapter Style (2D) — visage → perso One Piece")
 
     with gr.Row():
         with gr.Column():
             face_image  = gr.Image(type="pil", label="Photo visage (obligatoire)", height=260)
-            style_image = gr.Image(type="pil", label="Image de style (IP-Adapter) — optionnel", height=260)
-            prompt  = gr.Textbox(label="Prompt", value=EX_PROMPT, lines=3)
-            negative= gr.Textbox(label="Negative Prompt", value=EX_NEG, lines=3)
+            style_image = gr.Image(type="pil", label="Image de style (optionnel)", height=260,
+                                   info="Poster/planche One Piece, ou visuel manga servant de style global")
+            prompt   = gr.Textbox(label="Prompt", value=EX_PROMPT, lines=3)
+            negative = gr.Textbox(label="Negative Prompt", value=EX_NEG, lines=3)
 
             with gr.Row():
                 identity_strength = gr.Slider(0.2, 1.5, 0.95, 0.05, label="Fidélité visage (IdentityNet)")
-                adapter_strength  = gr.Slider(0.1, 1.5, 0.85, 0.05, label="Détails anime (InstantID IP-Adapter)")
+                adapter_strength  = gr.Slider(0.1, 1.5, 0.85, 0.05, label="Détails anime (InstantID)")
 
             style_strength = gr.Slider(0.1, 1.5, 0.95, 0.05, label="Force style (IP-Adapter Style)")
 
@@ -304,6 +297,6 @@ with gr.Blocks(css="footer{display:none !important}") as demo:
         outputs=[out_image, err_box, log_box],
     )
 
-demo.queue()
+demo.queue(api_open=False)
 if __name__ == "__main__":
     demo.launch(server_name="0.0.0.0", server_port=7860, ssr_mode=False)