Rollback Working except Avatar

app.py

@@ -3,95 +3,16 @@
 # Author: Vijay S. Chaudhari | 2025
 # ==========================================
 
-import importlib.util
 import gradio as gr
 import spaces
 import torch
 import cv2
 import numpy as np
-from pathlib import Path
-
 from PIL import Image, ImageEnhance, ImageOps
 from rembg import remove
 from diffusers import StableDiffusionImg2ImgPipeline
-from diffusers import StableDiffusionXLPipeline
 import io
-import os, sys, subprocess, warnings, logging
-
-warnings.filterwarnings("ignore", category=UserWarning)
-logging.getLogger("onnxruntime").setLevel(logging.ERROR)
-os.environ["CUDA_VISIBLE_DEVICES"] = ""
-
-# --- Ensure InstantID is available ---
-if not Path("instantid").exists():
-    print("🔄 Cloning InstantID repository...")
-    subprocess.run(["git", "clone", "--depth", "1", "https://github.com/InstantID/InstantID.git", "instantid"],check=True)
-
-repo_root = Path("instantid").resolve()
-
-# 🧭 Search for a pipeline file that matches *instantid*.py under the repo
-candidates = list(repo_root.rglob("pipeline*instantid*.py"))
-if not candidates:
-    # Fallback common names across commits
-    fallback_names = [
-        "pipelines/pipeline_instantid.py",
-        "pipelines/pipeline_stable_diffusion_instantid.py",
-        "pipelines/pipeline_stable_diffusion_xl_instantid.py",
-    ]
-    for name in fallback_names:
-        p = repo_root / name
-        if p.exists():
-            candidates = [p]
-            break
-
-if not candidates:
-    raise FileNotFoundError(
-        "Could not locate an InstantID pipeline file under ./instantid. "
-        "Repo layout may have changed. Please check the repo structure."
-    )
-
-pipeline_file = candidates[0]
-print(f"✅ Using InstantID pipeline file: {pipeline_file.relative_to(repo_root)}")
-
-# 🪄 Import the pipeline module by file path (no package needed)
-spec = importlib.util.spec_from_file_location("instantid_pipeline", str(pipeline_file))
-instantid_mod = importlib.util.module_from_spec(spec)
-spec.loader.exec_module(instantid_mod)  # type: ignore
-
-# 🔎 Pick a pipeline class that looks like an InstantID Pipeline
-InstantIDPipeline = None
-for attr in dir(instantid_mod):
-    if "InstantID" in attr and "Pipeline" in attr:
-        InstantIDPipeline = getattr(instantid_mod, attr)
-        break
-
-if InstantIDPipeline is None:
-    # Helpful diagnostics
-    print("Available names in module:", [a for a in dir(instantid_mod) if "Pipeline" in a])
-    raise ImportError(
-        "Could not find an InstantID pipeline class. "
-        "Looked for a class name containing both 'InstantID' and 'Pipeline'."
-    )
-
-print(f"✅ Imported pipeline class: {InstantIDPipeline.__name__}")
-
-'''
-if os.path.exists("InstantID") and not os.path.exists("instantid"):
-    os.rename("InstantID", "instantid")    
 
-instantid_path = os.path.abspath("instantid")
-sys.path.append(instantid_path)
-sys.path.append(os.path.join(instantid_path, "pipelines"))
-
-#sys.path.append(os.path.abspath("instantid"))
-#sys.path.insert(0, os.path.join(os.getcwd(), 'InstantID'))
-try:
-    from pipelines.pipeline_instantid import InstantIDPipeline
-    print("✅ InstantIDPipeline imported successfully.")
-except Exception as e:
-    print("⚠️ Failed to import InstantIDPipeline:", e)
-    InstantIDPipeline = None  # graceful fallback
-'''
 
 import torchvision
 print("Printing Torch and TorchVision versions:")
@@ -217,10 +138,12 @@ def create_passport(img: Image.Image) -> Image.Image:
 
 @spaces.GPU
 def create_avatar(img: Image.Image, prompt: str, strength: float, guidance_scale: float) -> Image.Image:
-    """
-    Create a stylized AI avatar while preserving facial identity using InstantID.
-    Retains core facial features, skin tone, and expressions of the input photo.
-    """
+    """Stylized AI avatar using Stable Diffusion Img2Img with user inputs"""
+    # Enhance face
+    img_enhanced = enhance_face(img)
+    
+    # Resize for SD (512x512)
+    img_resized = img_enhanced.convert("RGB").resize((512, 512))
 
     # Stylize with SD prompt. We are selecting these from UI now.
     #prompt = "highly detailed, digital portrait, professional lighting, cinematic style, artistic AI avatar"
@@ -228,84 +151,13 @@ def create_avatar(img: Image.Image, prompt: str, strength: float, guidance_scale
     #prompt = "studio portrait, even lighting, neutral background, realistic skin, confident pose"
     #prompt = "realistic professional headshot, soft studio lighting, neutral background, crisp details, natural skin tone"    
 
-    # --- Convert input ---
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    img = img.convert("RGB").resize((512, 512), Image.Resampling.LANCZOS)
-
-    # --- Step 1: Load InstantID + SDXL pipeline ---
-    pipe = StableDiffusionXLPipeline.from_pretrained(
-        "stabilityai/stable-diffusion-xl-base-1.0",
-        torch_dtype=torch.float16
-    ).to(device)
-
-    instantid = InstantIDPipeline.from_pretrained("InstantID/InstantID", torch_dtype=torch.float16,)
-    pipe.to("cuda" if torch.cuda.is_available() else "cpu")
-    #pipe.load_ip_adapter(instantid)
-
-    # --- Step 2: Optimize for ZeroGPU memory ---
-    pipe.enable_attention_slicing()
-    pipe.enable_model_cpu_offload()
-
-    # --- Step 3: Prepare conditioning (face embedding) ---
-    np_img = np.array(img)
-    bgr_img = cv2.cvtColor(np_img, cv2.COLOR_RGB2BGR)
-    face_emb = instantid.extract_face_embedding(bgr_img)  # key step: ID embedding guidance
-
-    # --- Step 4: Stylized generation ---
-    gen = pipe.generate_with_identity(
-        image=img,
-        face_embedding=face_emb,
-        prompt=(
-            prompt
-            + ", portrait of the same person, consistent identity, detailed lighting, "
-              "highly realistic skin texture, cinematic color tones"
-        ),
-        strength=float(strength),
-        guidance_scale=float(guidance_scale),
-        num_inference_steps=30
-    )
-
-    avatar = gen.images[0]
-
-    # --- Step 5 (Optional): Post-process with GFPGAN for crispness ---
-    try:
-        from gfpgan import GFPGANer
-        from realesrgan import RealESRGANer
-        from basicsr.archs.rrdbnet_arch import RRDBNet
-
-        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64,
-                        num_block=23, num_grow_ch=32, scale=2)
-        upsampler = RealESRGANer(
-            scale=2,
-            model_path='https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth',
-            model=model,
-            tile=400,
-            tile_pad=10,
-            pre_pad=0,
-            half=True,
-            device=device
-        )
-        face_enhancer = GFPGANer(
-            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
-            upscale=1,
-            arch='clean',
-            channel_multiplier=2,
-            bg_upsampler=upsampler,
-            device=device
-        )
-
-        img_cv = cv2.cvtColor(np.array(avatar), cv2.COLOR_RGB2BGR)
-        _, _, restored_img = face_enhancer.enhance(
-            img_cv, has_aligned=False, only_center_face=False,
-            paste_back=True, weight=0.4
-        )
-        avatar = Image.fromarray(cv2.cvtColor(restored_img, cv2.COLOR_BGR2RGB))
-    except Exception as e:
-        print(f"[WARN] GFPGAN post-process skipped: {e}")
+    with torch.autocast("cuda"):
+        result = sd_pipe(prompt=prompt, image=img_resized, strength=strength, guidance_scale=guidance_scale)
 
+    avatar = enhance_face(result.images[0])
+    
     return avatar
 
-
 @spaces.GPU
 def process_all(img: Image.Image):
     """Process all three types at once"""

app copy.py → app.py.wip_avatar

@@ -3,18 +3,95 @@
 # Author: Vijay S. Chaudhari | 2025
 # ==========================================
 
+import importlib.util
 import gradio as gr
 import spaces
 import torch
 import cv2
 import numpy as np
+from pathlib import Path
+
 from PIL import Image, ImageEnhance, ImageOps
 from rembg import remove
 from diffusers import StableDiffusionImg2ImgPipeline
 from diffusers import StableDiffusionXLPipeline
-from instantid import InstantID
 import io
+import os, sys, subprocess, warnings, logging
+
+warnings.filterwarnings("ignore", category=UserWarning)
+logging.getLogger("onnxruntime").setLevel(logging.ERROR)
+os.environ["CUDA_VISIBLE_DEVICES"] = ""
+
+# --- Ensure InstantID is available ---
+if not Path("instantid").exists():
+    print("🔄 Cloning InstantID repository...")
+    subprocess.run(["git", "clone", "--depth", "1", "https://github.com/InstantID/InstantID.git", "instantid"],check=True)
+
+repo_root = Path("instantid").resolve()
+
+# 🧭 Search for a pipeline file that matches *instantid*.py under the repo
+candidates = list(repo_root.rglob("pipeline*instantid*.py"))
+if not candidates:
+    # Fallback common names across commits
+    fallback_names = [
+        "pipelines/pipeline_instantid.py",
+        "pipelines/pipeline_stable_diffusion_instantid.py",
+        "pipelines/pipeline_stable_diffusion_xl_instantid.py",
+    ]
+    for name in fallback_names:
+        p = repo_root / name
+        if p.exists():
+            candidates = [p]
+            break
+
+if not candidates:
+    raise FileNotFoundError(
+        "Could not locate an InstantID pipeline file under ./instantid. "
+        "Repo layout may have changed. Please check the repo structure."
+    )
+
+pipeline_file = candidates[0]
+print(f"✅ Using InstantID pipeline file: {pipeline_file.relative_to(repo_root)}")
+
+# 🪄 Import the pipeline module by file path (no package needed)
+spec = importlib.util.spec_from_file_location("instantid_pipeline", str(pipeline_file))
+instantid_mod = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(instantid_mod)  # type: ignore
+
+# 🔎 Pick a pipeline class that looks like an InstantID Pipeline
+InstantIDPipeline = None
+for attr in dir(instantid_mod):
+    if "InstantID" in attr and "Pipeline" in attr:
+        InstantIDPipeline = getattr(instantid_mod, attr)
+        break
+
+if InstantIDPipeline is None:
+    # Helpful diagnostics
+    print("Available names in module:", [a for a in dir(instantid_mod) if "Pipeline" in a])
+    raise ImportError(
+        "Could not find an InstantID pipeline class. "
+        "Looked for a class name containing both 'InstantID' and 'Pipeline'."
+    )
+
+print(f"✅ Imported pipeline class: {InstantIDPipeline.__name__}")
+
+'''
+if os.path.exists("InstantID") and not os.path.exists("instantid"):
+    os.rename("InstantID", "instantid")    
 
+instantid_path = os.path.abspath("instantid")
+sys.path.append(instantid_path)
+sys.path.append(os.path.join(instantid_path, "pipelines"))
+
+#sys.path.append(os.path.abspath("instantid"))
+#sys.path.insert(0, os.path.join(os.getcwd(), 'InstantID'))
+try:
+    from pipelines.pipeline_instantid import InstantIDPipeline
+    print("✅ InstantIDPipeline imported successfully.")
+except Exception as e:
+    print("⚠️ Failed to import InstantIDPipeline:", e)
+    InstantIDPipeline = None  # graceful fallback
+'''
 
 import torchvision
 print("Printing Torch and TorchVision versions:")
@@ -140,12 +217,10 @@ def create_passport(img: Image.Image) -> Image.Image:
 
 @spaces.GPU
 def create_avatar(img: Image.Image, prompt: str, strength: float, guidance_scale: float) -> Image.Image:
-    """Stylized AI avatar using Stable Diffusion Img2Img with user inputs"""
-    # Enhance face
-    img_enhanced = enhance_face(img)
-    
-    # Resize for SD (512x512)
-    img_resized = img_enhanced.convert("RGB").resize((512, 512))
+    """
+    Create a stylized AI avatar while preserving facial identity using InstantID.
+    Retains core facial features, skin tone, and expressions of the input photo.
+    """
 
     # Stylize with SD prompt. We are selecting these from UI now.
     #prompt = "highly detailed, digital portrait, professional lighting, cinematic style, artistic AI avatar"
@@ -153,13 +228,84 @@ def create_avatar(img: Image.Image, prompt: str, strength: float, guidance_scale
     #prompt = "studio portrait, even lighting, neutral background, realistic skin, confident pose"
     #prompt = "realistic professional headshot, soft studio lighting, neutral background, crisp details, natural skin tone"    
 
-    with torch.autocast("cuda"):
-        result = sd_pipe(prompt=prompt, image=img_resized, strength=strength, guidance_scale=guidance_scale)
+    # --- Convert input ---
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    img = img.convert("RGB").resize((512, 512), Image.Resampling.LANCZOS)
+
+    # --- Step 1: Load InstantID + SDXL pipeline ---
+    pipe = StableDiffusionXLPipeline.from_pretrained(
+        "stabilityai/stable-diffusion-xl-base-1.0",
+        torch_dtype=torch.float16
+    ).to(device)
+
+    instantid = InstantIDPipeline.from_pretrained("InstantID/InstantID", torch_dtype=torch.float16,)
+    pipe.to("cuda" if torch.cuda.is_available() else "cpu")
+    #pipe.load_ip_adapter(instantid)
+
+    # --- Step 2: Optimize for ZeroGPU memory ---
+    pipe.enable_attention_slicing()
+    pipe.enable_model_cpu_offload()
+
+    # --- Step 3: Prepare conditioning (face embedding) ---
+    np_img = np.array(img)
+    bgr_img = cv2.cvtColor(np_img, cv2.COLOR_RGB2BGR)
+    face_emb = instantid.extract_face_embedding(bgr_img)  # key step: ID embedding guidance
+
+    # --- Step 4: Stylized generation ---
+    gen = pipe.generate_with_identity(
+        image=img,
+        face_embedding=face_emb,
+        prompt=(
+            prompt
+            + ", portrait of the same person, consistent identity, detailed lighting, "
+              "highly realistic skin texture, cinematic color tones"
+        ),
+        strength=float(strength),
+        guidance_scale=float(guidance_scale),
+        num_inference_steps=30
+    )
+
+    avatar = gen.images[0]
+
+    # --- Step 5 (Optional): Post-process with GFPGAN for crispness ---
+    try:
+        from gfpgan import GFPGANer
+        from realesrgan import RealESRGANer
+        from basicsr.archs.rrdbnet_arch import RRDBNet
+
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64,
+                        num_block=23, num_grow_ch=32, scale=2)
+        upsampler = RealESRGANer(
+            scale=2,
+            model_path='https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth',
+            model=model,
+            tile=400,
+            tile_pad=10,
+            pre_pad=0,
+            half=True,
+            device=device
+        )
+        face_enhancer = GFPGANer(
+            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
+            upscale=1,
+            arch='clean',
+            channel_multiplier=2,
+            bg_upsampler=upsampler,
+            device=device
+        )
+
+        img_cv = cv2.cvtColor(np.array(avatar), cv2.COLOR_RGB2BGR)
+        _, _, restored_img = face_enhancer.enhance(
+            img_cv, has_aligned=False, only_center_face=False,
+            paste_back=True, weight=0.4
+        )
+        avatar = Image.fromarray(cv2.cvtColor(restored_img, cv2.COLOR_BGR2RGB))
+    except Exception as e:
+        print(f"[WARN] GFPGAN post-process skipped: {e}")
 
-    avatar = enhance_face(result.images[0])
-    
     return avatar
 
+
 @spaces.GPU
 def process_all(img: Image.Image):
     """Process all three types at once"""