vhup 1.0__tar.gz → 1.2__tar.gz

{vhup-1.0/vhup.egg-info → vhup-1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vhup
-Version: 1.0
+Version: 1.2
 Summary: Vho: İşlevsel ve kolaylaştırılmış görsel oluşturma kütüphanesi
 Author: Bedirhan
 Author-email: bedirhan.oytpass@gmail.com

{vhup-1.0 → vhup-1.2}/setup.py

@@ -5,7 +5,7 @@ with open("README.md", "r", encoding="utf-8") as f:
 
 setup(
     name="vhup",
-    version="1.0",
+    version="1.2",
     packages=find_packages(),
     install_requires=[
         "torch>=2.0.0",

vhup-1.2/vhup/__init__.py

@@ -0,0 +1,2 @@
+from .vhup import create_text
+from .horba import create_loop

vhup-1.2/vhup/horba.py

@@ -0,0 +1,61 @@
+import torch
+import cv2
+import numpy as np
+import os
+from PIL import Image
+from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, DPMSolverMultistepScheduler
+from deep_translator import GoogleTranslator
+
+# Donanım ayarları
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.float16 if device == "cuda" else torch.float32
+
+# 1. ControlNet ve Model Yükleme
+controlnet = ControlNetModel.from_pretrained(
+    "lllyasviel/sd-controlnet-canny", 
+    torch_dtype=dtype
+)
+
+_pipe = StableDiffusionControlNetPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    controlnet=controlnet,
+    torch_dtype=dtype,
+    safety_checker=None
+).to(device)
+
+_pipe.scheduler = DPMSolverMultistepScheduler.from_config(_pipe.scheduler.config)
+translator = GoogleTranslator(source="tr", target="en")
+
+def create_loop(prompt: str, foto: str) -> str:
+    # A. Referans Görsel İşleme (Canny Edge)
+    input_image = Image.open(foto).convert("RGB")
+    image_np = np.array(input_image)
+    
+    # Kenar algılama ile iskelet çıkarma
+    image_canny = cv2.Canny(image_np, 100, 200)
+    image_canny = image_canny[:, :, None]
+    image_canny = np.concatenate([image_canny, image_canny, image_canny], axis=2)
+    canny_image = Image.fromarray(image_canny)
+
+    # B. Prompt Hazırlığı
+    # Türkçe giriş gelirse İngilizce'ye çevirir
+    translated_prompt = translator.translate(prompt)
+    
+    # Kalite takıları (Quality Tags)
+    final_prompt = f"{translated_prompt}, highly detailed, masterpiece, 8k wallpaper"
+    negative_prompt = "lowres, bad anatomy, worst quality, blurry, deformed face"
+
+    # C. Üretim Süreci
+    output = _pipe(
+        prompt=final_prompt,
+        image=canny_image, 
+        negative_prompt=negative_prompt,
+        num_inference_steps=30,
+        controlnet_conditioning_scale=0.7 # 0.7 referansa iyi sadık kalır ama yaratıcılığa da alan bırakır
+    ).images[0]
+
+    # D. Kayıt
+    output_path = "vho_result.png"
+    output.save(output_path)
+    return os.path.abspath(output_path)
+

{vhup-1.0 → vhup-1.2/vhup.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vhup
-Version: 1.0
+Version: 1.2
 Summary: Vho: İşlevsel ve kolaylaştırılmış görsel oluşturma kütüphanesi
 Author: Bedirhan
 Author-email: bedirhan.oytpass@gmail.com

vhup-1.0/vhup/__init__.py

@@ -1,2 +0,0 @@
-from .vhup import create_text
-from .horba import create_hore, create_hore_Bef

vhup-1.0/vhup/horba.py

@@ -1,120 +0,0 @@
-import torch
-import random
-import os
-from PIL import Image
-from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, DPMSolverMultistepScheduler
-from transformers import CLIPVisionModelWithProjection, CLIPImageProcessor
-from deep_translator import GoogleTranslator
-
-# Cihaz ayarları
-device = "cuda" if torch.cuda.is_available() else "cpu"
-dtype = torch.float16 if device == "cuda" else torch.float32
-
-FIXED_OUTPUT_FILE = "vho1.png"
-translator = GoogleTranslator(source="tr", target="en")
-
-TR_STYLE_MAP = {
-    "gerçekçi": "realistic", "sinematik": "cinematic lighting",
-    "anime": "anime style", "çizgi film": "cartoon style",
-    "karanlık": "dark dramatic mood", "neon": "neon cyberpunk lighting",
-    "portre": "portrait photography", "detaylı": "highly detailed",
-    "fantastik": "fantasy art", "bilim kurgu": "science fiction"
-}
-
-def load_pipelines():
-    model_id = "runwayml/stable-diffusion-v1-5"
-    # Karakter koruma için kritik olan görsel işlemci
-    clip_id = "openai/clip-vit-large-patch14"
-    
-    # 1. İşlemcileri yükle
-    img_proc = CLIPImageProcessor.from_pretrained(clip_id)
-    img_enc = CLIPVisionModelWithProjection.from_pretrained(
-        clip_id, 
-        torch_dtype=dtype
-    ).to(device)
-    
-    # 2. Ana Pipeline'ı yükle
-    pipe = StableDiffusionPipeline.from_pretrained(
-        model_id, 
-        image_encoder=img_enc,
-        torch_dtype=dtype, 
-        safety_checker=None
-    ).to(device)
-    
-    # 3. IP-Adapter ağırlıklarını yükle
-    pipe.load_ip_adapter(
-        "h94/IP-Adapter", 
-        subfolder="models", 
-        weight_name="ip-adapter_sd15.bin"
-    )
-    
-    pipe.set_ip_adapter_scale(0.7)
-    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
-    pipe.enable_attention_slicing()
-    
-    # Kesin olarak 2 değer döndür
-    return pipe, img_proc
-
-# Pipeline'ları başlat
-_pipe, _image_processor = load_pipelines()
-
-def _prepare_prompt(prompt_tr: str) -> str:
-    clean_prompt = prompt_tr.replace("#", "the person")
-    translated = translator.translate(clean_prompt)
-    styles = [en for tr, en in TR_STYLE_MAP.items() if tr in prompt_tr.lower()]
-    return f"{translated}, {', '.join(styles)}, masterpiece, high quality, sharp focus"
-
-def _get_image_embeds(hero_path):
-    # Karakter görselini tensor formatına çevirir (Tuple hatasını önler)
-    face_img = Image.open(hero_path).convert("RGB")
-    clip_input = _image_processor(images=face_img, return_tensors="pt").to(device, dtype=dtype)
-    outputs = _pipe.image_encoder(**clip_input)
-    return outputs.image_embeds.unsqueeze(1)
-
-def create_hore(prompt: str, hero: str, kalite: float = 7.5) -> str:
-    final_p = _prepare_prompt(prompt)
-    embeds = _get_image_embeds(hero)
-    
-    seed = random.randint(0, 999999999)
-    gen = torch.Generator(device=device).manual_seed(seed)
-
-    image = _pipe(
-        prompt=final_p,
-        ip_adapter_image_embeds=[embeds],
-        negative_prompt="worst quality, low quality, bad anatomy, deformed",
-        num_inference_steps=30,
-        guidance_scale=kalite,
-        generator=gen
-    ).images[0]
-
-    image.save(FIXED_OUTPUT_FILE)
-    return os.path.abspath(FIXED_OUTPUT_FILE)
-
-def create_hore_Bef(prompt: str, hero: str, strength: float = 0.6) -> str:
-    # Önceki görsel yoksa create_hore'a pasla
-    if not os.path.exists(FIXED_OUTPUT_FILE):
-        return create_hore(prompt, hero)
-    
-    current_bg = Image.open(FIXED_OUTPUT_FILE).convert("RGB")
-    embeds = _get_image_embeds(hero)
-    final_p = _prepare_prompt(prompt)
-
-    # Img2Img için geçici pipeline oluştur
-    i2i_pipe = StableDiffusionImg2ImgPipeline(**_pipe.components)
-    
-    seed = random.randint(0, 999999999)
-    gen = torch.Generator(device=device).manual_seed(seed)
-
-    image = i2i_pipe(
-        prompt=final_p,
-        image=current_bg,
-        ip_adapter_image_embeds=[embeds],
-        strength=strength,
-        num_inference_steps=30,
-        negative_prompt="worst quality, low quality",
-        generator=gen
-    ).images[0]
-
-    image.save(FIXED_OUTPUT_FILE)
-    return os.path.abspath(FIXED_OUTPUT_FILE)
-