Update app.py

app.py

@@ -1,6 +1,9 @@
-# app.py — InstantID (SDXL) résilient: essaie plusieurs chemins de pipeline dans InstantX/InstantID
+# app.py — InstantID SDXL (Option 1: téléchargements auto des poids depuis un repo Model)
+# - Tu uploades localement SEULEMENT la pipeline .py (texte) dans ./instantid/
+# - Les poids (safetensors/bin) sont téléchargés au runtime depuis InstantX/InstantID (repo Model)
 import os, traceback, importlib.util
 
+# Eviter l'erreur libgomp
 os.environ.setdefault("OMP_NUM_THREADS", "4")
 os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
 
@@ -9,95 +12,86 @@ from PIL import Image, ImageOps
 from huggingface_hub import hf_hub_download, HfHubHTTPError
 from diffusers.models import ControlNetModel
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-dtype  = torch.float16 if device == "cuda" else torch.float32
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+DTYPE  = torch.float16 if DEVICE == "cuda" else torch.float32
 
-REPO = "InstantX/InstantID"
+# -------- Références Hub (repo MODEL public qui contient les poids) --------
+ASSETS_REPO = "InstantX/InstantID"   # tu peux le remplacer par ton propre repo MODEL si besoin
 
-# --- candidats de fichiers possibles dans le repo (les noms varient selon les versions)
+# -------- Chemins locaux attendus dans le Space --------
 PIPE_CANDIDATES = [
-    "pipeline_stable_diffusion_xl_instantid_full.py",
-    "pipeline_stable_diffusion_xl_instantid.py",
-    "pipelines/pipeline_stable_diffusion_xl_instantid_full.py",
-    "pipelines/pipeline_stable_diffusion_xl_instantid.py",
+    "./instantid/pipeline_stable_diffusion_xl_instantid.py",
+    "./instantid/pipeline_stable_diffusion_xl_instantid_full.py",
 ]
+CHECKPOINTS_DIR   = "./checkpoints"
+CN_LOCAL_DIR      = os.path.join(CHECKPOINTS_DIR, "ControlNetModel")
+IP_ADAPTER_LOCAL  = os.path.join(CHECKPOINTS_DIR, "ip-adapter.bin")
+
+# -------- Utilitaires --------
+def import_pipeline_or_fail():
+    pipeline_file = next((p for p in PIPE_CANDIDATES if os.path.exists(p)), None)
+    if pipeline_file is None:
+        raise RuntimeError(
+            "Pipeline InstantID introuvable.\n"
+            "➡️ Uploade l’un de ces fichiers (texte) dans ton Space :\n"
+            "   - instantid/pipeline_stable_diffusion_xl_instantid.py\n"
+            "   - instantid/pipeline_stable_diffusion_xl_instantid_full.py\n"
+            "Les poids seront téléchargés automatiquement."
+        )
+    spec = importlib.util.spec_from_file_location("instantid_pipeline", pipeline_file)
+    mod  = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    SDXLInstantID = getattr(mod, "StableDiffusionXLInstantIDPipeline", None)
+    draw_kps = getattr(mod, "draw_kps", None)
+    if SDXLInstantID is None or draw_kps is None:
+        raise RuntimeError("Le fichier pipeline ne contient pas StableDiffusionXLInstantIDPipeline/draw_kps.")
+    return SDXLInstantID, draw_kps
+
+def ensure_assets_or_download():
+    os.makedirs(CHECKPOINTS_DIR, exist_ok=True)
+    os.makedirs(CN_LOCAL_DIR, exist_ok=True)
+    # Télécharge/valide ControlNet IdentityNet
+    try:
+        if not os.path.isfile(os.path.join(CN_LOCAL_DIR, "config.json")):
+            hf_hub_download(ASSETS_REPO, "ControlNetModel/config.json", local_dir=CHECKPOINTS_DIR)
+        if not os.path.isfile(os.path.join(CN_LOCAL_DIR, "diffusion_pytorch_model.safetensors")):
+            hf_hub_download(ASSETS_REPO, "ControlNetModel/diffusion_pytorch_model.safetensors", local_dir=CHECKPOINTS_DIR)
+    except HfHubHTTPError as e:
+        raise RuntimeError(f"Echec téléchargement IdentityNet depuis {ASSETS_REPO} : {e}")
+
+    # Télécharge/valide ip-adapter.bin
+    try:
+        if not os.path.isfile(IP_ADAPTER_LOCAL):
+            hf_hub_download(ASSETS_REPO, "ip-adapter.bin", local_dir=CHECKPOINTS_DIR)
+    except HfHubHTTPError as e:
+        raise RuntimeError(f"Echec téléchargement ip-adapter.bin depuis {ASSETS_REPO} : {e}")
 
+# -------- Chargement pipeline --------
 load_logs = []
+try:
+    SDXLInstantID, draw_kps = import_pipeline_or_fail()
+    ensure_assets_or_download()
 
-def _download_first_existing(repo_id: str, candidates: list[str], local_dir: str) -> str | None:
-    for fname in candidates:
-        try:
-            p = hf_hub_download(repo_id=repo_id, filename=fname, local_dir=local_dir)
-            load_logs.append(f"✅ Pipeline trouvée: {fname}")
-            return p
-        except HfHubHTTPError as e:
-            load_logs.append(f"… {fname} introuvable ({e.__class__.__name__})")
-        except Exception as e:
-            load_logs.append(f"… {fname} erreur: {e}")
-    return None
-
-# 1) Télécharger la pipeline InstantID (un des fichiers ci-dessus)
-PIPE_LOCAL = _download_first_existing(REPO, PIPE_CANDIDATES, "./instantid")
-if PIPE_LOCAL is None:
-    # Abort propre avec aide utilisateur
-    msg = (
-        "Aucun fichier pipeline *.py trouvé dans InstantX/InstantID.\n"
-        "Solutions:\n"
-        " - Uploade manuellement dans /instantid/ un fichier pipeline nommé, par ex:\n"
-        "   pipeline_stable_diffusion_xl_instantid_full.py\n"
-        "   (copie-le depuis le repo InstantX/InstantID)\n"
-        " - Ou change la variable PIPE_CANDIDATES pour matcher le nom exact dans le repo.\n"
-    )
-    raise RuntimeError(msg + "\n" + "\n".join(load_logs))
+    BASE_MODEL = "stabilityai/stable-diffusion-xl-base-1.0"  # remplaçable par un SDXL style anime si tu en as
 
-# 2) Télécharger les poids IdentityNet (ControlNet) + ip-adapter.bin
-try:
-    cn_cfg = hf_hub_download(repo_id=REPO, filename="ControlNetModel/config.json", local_dir="./checkpoints")
-    cn_dir = os.path.dirname(cn_cfg)  # ./checkpoints/ControlNetModel
-    hf_hub_download(repo_id=REPO, filename="ControlNetModel/diffusion_pytorch_model.safetensors", local_dir="./checkpoints")
-    ip_adapter_path = hf_hub_download(repo_id=REPO, filename="ip-adapter.bin", local_dir="./checkpoints")
-    load_logs.append("✅ IdentityNet + ip-adapter téléchargés.")
-except Exception as e:
-    raise RuntimeError(f"Echec téléchargement IdentityNet/ip-adapter: {e}")
-
-# 3) Import dynamique de la pipeline
-import importlib.util
-spec = importlib.util.spec_from_file_location("instantid_pipeline", PIPE_LOCAL)
-mod  = importlib.util.module_from_spec(spec)
-spec.loader.exec_module(mod)
-
-# Ces symboles existent dans les implémentations InstantID SDXL
-StableDiffusionXLInstantIDPipeline = getattr(mod, "StableDiffusionXLInstantIDPipeline", None)
-draw_kps = getattr(mod, "draw_kps", None)
-if StableDiffusionXLInstantIDPipeline is None or draw_kps is None:
-    raise RuntimeError("La pipeline importée ne contient pas StableDiffusionXLInstantIDPipeline/draw_kps.")
-
-# 4) Modèle de base SDXL — prends un SDXL stylé anime ou le base officiel
-BASE_MODEL = "stabilityai/stable-diffusion-xl-base-1.0"
-# Astuce: si tu as un SDXL anime (ex: YamerMIX), mets-le ici pour rendu plus manga:
-# BASE_MODEL = "wangqixun/YamerMIX_v8"
-
-# 5) Charger IdentityNet + pipeline
-try:
     load_logs.append("Chargement ControlNet IdentityNet…")
-    controlnet_identitynet = ControlNetModel.from_pretrained(cn_dir, torch_dtype=dtype)
+    controlnet_identitynet = ControlNetModel.from_pretrained(CN_LOCAL_DIR, torch_dtype=DTYPE)
 
     load_logs.append(f"Chargement pipeline InstantID (base={BASE_MODEL})…")
-    pipe = StableDiffusionXLInstantIDPipeline.from_pretrained(
+    pipe = SDXLInstantID.from_pretrained(
         BASE_MODEL,
         controlnet=[controlnet_identitynet],
-        torch_dtype=dtype,
+        torch_dtype=DTYPE,
         safety_checker=None,
         feature_extractor=None,
-    ).to(device)
+    ).to(DEVICE)
 
-    # ip-adapter d’InstantID
     if hasattr(pipe, "load_ip_adapter_instantid"):
-        pipe.load_ip_adapter_instantid(ip_adapter_path)
+        pipe.load_ip_adapter_instantid(IP_ADAPTER_LOCAL)
     else:
-        raise RuntimeError("La méthode pipe.load_ip_adapter_instantid est absente dans cette pipeline.")
+        raise RuntimeError("La méthode load_ip_adapter_instantid est absente de cette pipeline.")
 
-    if device == "cuda":
+    if DEVICE == "cuda":
         if hasattr(pipe, "image_proj_model"): pipe.image_proj_model.to("cuda")
         if hasattr(pipe, "unet"):             pipe.unet.to("cuda")
 
@@ -107,36 +101,35 @@ except Exception:
     pipe = None
 
 if pipe is None:
-    raise RuntimeError("Échec de chargement de la pipeline InstantID.\n" + "\n".join(load_logs))
+    raise RuntimeError("Échec chargement pipeline InstantID.\n" + "\n".join(load_logs))
 
-# 6) InsightFace pour landmarks
+# -------- InsightFace pour landmarks --------
 from insightface.app import FaceAnalysis
 fa = FaceAnalysis(name="antelopev2", root="./", providers=["CPUExecutionProvider"])
 fa.prepare(ctx_id=0, det_size=(640, 640))
 
-def extract_face_info(pil_img: Image.Image):
+def extract_kps_image(pil_img: Image.Image):
     import numpy as np, cv2
     img_cv2 = cv2.cvtColor(np.array(pil_img.convert("RGB")), cv2.COLOR_RGB2BGR)
     faces = fa.get(img_cv2)
     if not faces:
-        raise ValueError("Aucun visage détecté. Utilise un portrait net (buste, visage bien centré).")
-    face = faces[-1]
-    kps_image = draw_kps(pil_img, face["kps"])
-    return face, kps_image
+        raise ValueError("Aucun visage détecté. Utilise un portrait net (visage centré).")
+    face = faces[-1]  # visage principal
+    return draw_kps(pil_img, face["kps"])
 
-# 7) Inference
+# -------- Inference --------
 def generate(face_image, prompt, negative_prompt, identity_strength, adapter_strength, steps, cfg, width, height, seed):
     try:
         if face_image is None:
             return None, "Merci d'ajouter une photo visage.", "\n".join(load_logs)
 
-        gen = None if seed is None or int(seed) < 0 else torch.Generator(device=device).manual_seed(int(seed))
+        gen = None if seed is None or int(seed) < 0 else torch.Generator(device=DEVICE).manual_seed(int(seed))
 
         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)
 
-        face_info, face_kps = extract_face_info(face_sq)
+        kps_img = extract_kps_image(face_sq)
 
         if hasattr(pipe, "set_ip_adapter_scale"):
             pipe.set_ip_adapter_scale(float(adapter_strength))
@@ -144,7 +137,7 @@ def generate(face_image, prompt, negative_prompt, identity_strength, adapter_str
         images = pipe(
             prompt=prompt.strip(),
             negative_prompt=(negative_prompt or "").strip(),
-            image=face_kps,
+            image=kps_img,  # landmarks pour IdentityNet
             controlnet_conditioning_scale=float(identity_strength),
             num_inference_steps=int(steps),
             guidance_scale=float(cfg),
@@ -154,14 +147,13 @@ def generate(face_image, prompt, negative_prompt, identity_strength, adapter_str
         ).images
 
         return images[0], "", "\n".join(load_logs)
-
     except torch.cuda.OutOfMemoryError as oom:
-        msg = "CUDA OOM: baisse résolution (ex: 640×768 → 576×704), steps 24–28, CFG 5–7."
+        msg = "CUDA OOM: baisse la résolution (ex: 640×768 → 576×704), steps 24–28, CFG 5–6."
         return None, f"{msg}\n{oom}", "\n".join(load_logs)
     except Exception:
         return None, "Erreur:\n"+traceback.format_exc(), "\n".join(load_logs)
 
-# 8) UI
+# -------- UI --------
 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, "
@@ -173,7 +165,7 @@ EX_NEG = (
 )
 
 with gr.Blocks(css="footer{display:none !important}") as demo:
-    gr.Markdown("# 🏴‍☠️ One Piece — InstantID (SDXL)")
+    gr.Markdown("# 🏴‍☠️ One Piece — InstantID (SDXL) — Poids auto depuis HF")
 
     with gr.Row():
         with gr.Column():
@@ -181,7 +173,7 @@ with gr.Blocks(css="footer{display:none !important}") as demo:
             prompt = gr.Textbox(label="Prompt", value=EX_PROMPT, lines=3)
             negative = gr.Textbox(label="Negative Prompt", value=EX_NEG, lines=3)
 
-            identity_strength = gr.Slider(0.2, 1.5, value=0.90, step=0.05, label="IdentityNet strength (fidélité)")
+            identity_strength = gr.Slider(0.2, 1.5, value=0.95, step=0.05, label="IdentityNet strength (fidélité)")
             adapter_strength  = gr.Slider(0.2, 1.5, value=0.85, step=0.05, label="Adapter strength (détails)")
             steps  = gr.Slider(10, 60, value=30, step=1, label="Steps")
             cfg    = gr.Slider(0.1, 12.0, value=5.5, step=0.1, label="CFG")
@@ -208,4 +200,3 @@ with gr.Blocks(css="footer{display:none !important}") as demo:
 demo.queue()
 if __name__ == "__main__":
     demo.launch(ssr_mode=False, server_name="0.0.0.0", server_port=7860)
-