Update app.py

app.py

@@ -1,4 +1,3 @@
-# app.py - KORRIGIERTE VERSION
 import gradio as gr
 from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline
 import torch
@@ -9,7 +8,6 @@ import time
 device = "cuda" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if device == "cuda" else torch.float32
 IMG_SIZE = 512
-STEPS = 35
 
 print(f"Running on: {device}")
 
@@ -28,6 +26,11 @@ def load_txt2img():
             safety_checker=None,
             requires_safety_checker=False
         ).to(device)
+        
+        # DPMSolver für Text-to-Image
+        from diffusers import DPMSolverMultistepScheduler
+        pipe_txt2img.scheduler = DPMSolverMultistepScheduler.from_config(pipe_txt2img.scheduler.config)
+        
         pipe_txt2img.enable_attention_slicing()
     return pipe_txt2img
 
@@ -42,16 +45,22 @@ def load_img2img():
             safety_checker=None,
             requires_safety_checker=False
         ).to(device)
+        
+        # DPMSolver für Image-to-Image
+        from diffusers import DPMSolverMultistepScheduler
+        pipe_img2img.scheduler = DPMSolverMultistepScheduler.from_config(pipe_img2img.scheduler.config)
+        
         pipe_img2img.enable_attention_slicing()
     return pipe_img2img
 
 # === FUNKTIONEN ===
-def text_to_image(prompt):
+def text_to_image(prompt, steps, guidance_scale):
     try:
         if not prompt or not prompt.strip():
             return None
             
         print(f"Starting generation for: {prompt}")
+        print(f"Parameters - Steps: {steps}, Guidance Scale: {guidance_scale}")
         start_time = time.time()
         
         pipe = load_txt2img()
@@ -60,8 +69,8 @@ def text_to_image(prompt):
             prompt=prompt,
             height=IMG_SIZE,
             width=IMG_SIZE,
-            num_inference_steps=STEPS,
-            guidance_scale=3.0,
+            num_inference_steps=steps,
+            guidance_scale=guidance_scale,
         ).images[0]
         
         end_time = time.time()
@@ -75,23 +84,23 @@ def text_to_image(prompt):
         traceback.print_exc()
         return None
 
-def img_to_image(image, prompt="", strength=0.6):
+def img_to_image(image, prompt, strength, steps, guidance_scale):
     try:
         if image is None:
             return None
             
-        print(f"Image-to-Image mit Strength: {strength}")
+        print(f"Image-to-Image mit Strength: {strength}, Steps: {steps}, Guidance: {guidance_scale}")
         start_time = time.time()
         
         pipe = load_img2img()
-        img_resized = image.convert("RGB").resize((512, 512))
+        img_resized = image.convert("RGB").resize((IMG_SIZE, IMG_SIZE))
         
         result = pipe(
             prompt=prompt,
             image=img_resized,
             strength=strength,
-            num_inference_steps=35,
-            guidance_scale=7.5
+            num_inference_steps=steps,
+            guidance_scale=guidance_scale
         )
         
         end_time = time.time()
@@ -111,11 +120,33 @@ with gr.Blocks() as app:
     
     with gr.Tab("Text zu Bild"):
         gr.Markdown("**Beschreibe dein gewünschtes Bild:**")
-        txt_input = gr.Textbox(
-            placeholder="z.B. a red apple on a wooden table, photorealistic",
-            lines=2,
-            label="Prompt (Englisch)"
-        )
+        
+        with gr.Row():
+            txt_input = gr.Textbox(
+                placeholder="z.B. a red apple on a wooden table, photorealistic",
+                lines=2,
+                label="Prompt (Englisch)"
+            )
+        
+        with gr.Row():
+            with gr.Column():
+                txt_steps = gr.Slider(
+                    minimum=10, maximum=100, value=35, step=1,
+                    label="Steps (Qualität vs. Geschwindigkeit)"
+                )
+            with gr.Column():
+                txt_guidance = gr.Slider(
+                    minimum=1.0, maximum=20.0, value=7.5, step=0.5,
+                    label="Guidance Scale (Prompt-Treue)"
+                )
+        
+        with gr.Row():
+            gr.Markdown(
+                "**Parameter-Erklärung:**  "
+                "• **Steps:** Mehr = bessere Qualität, aber langsamer  "
+                "• **Guidance:** Höher = stärkere Prompt-Treue, kann verzerren  "
+            )
+        
         generate_btn = gr.Button("🎨 Bild generieren", variant="primary")
         txt_output = gr.Image(
             label="Generiertes Bild", 
@@ -124,7 +155,7 @@ with gr.Blocks() as app:
         
         generate_btn.click(
             fn=text_to_image,
-            inputs=txt_input,
+            inputs=[txt_input, txt_steps, txt_guidance],
             outputs=txt_output,
             concurrency_limit=1
         )
@@ -147,17 +178,28 @@ with gr.Blocks() as app:
             )
         
         with gr.Row():
-            strength_slider = gr.Slider(
-                0.1, 0.8, 0.4,  # Jetzt ab 0.3 möglich!
-                label="Stärke der Veränderung"
-            )
+            with gr.Column():
+                strength_slider = gr.Slider(
+                    minimum=0.1, maximum=0.9, value=0.4, step=0.05,
+                    label="Strength (Veränderungs-Stärke)"
+                )
+            with gr.Column():
+                img_steps = gr.Slider(
+                    minimum=10, maximum=100, value=35, step=1,
+                    label="Steps (Qualität vs. Geschwindigkeit)"
+                )
+            with gr.Column():
+                img_guidance = gr.Slider(
+                    minimum=1.0, maximum=20.0, value=7.5, step=0.5,
+                    label="Guidance Scale (Prompt-Treue)"
+                )
         
         with gr.Row():
             gr.Markdown(
-                "**Stärke-Einstellung:**  "
-                "• **0.5-0.6:** Behält fast alles Original bei  "
-                "• **0.65-0.7:** Gute Balance  "  
-                "• **0.75-0.8:** Stärkere Veränderungen"
+                "**Parameter-Erklärung:**  "
+                "• **Strength:** Niedrig = behält Original, Hoch = starke Veränderung  "
+                "• **Steps:** Mehr = bessere Qualität, aber langsamer  "
+                "• **Guidance:** Höher = stärkere Prompt-Treue  "
             )
         
         transform_btn = gr.Button("🔄 Bild transformieren", variant="primary")
@@ -170,7 +212,7 @@ with gr.Blocks() as app:
         
         transform_btn.click(
             fn=img_to_image,
-            inputs=[img_input, img_prompt, strength_slider],
+            inputs=[img_input, img_prompt, strength_slider, img_steps, img_guidance],
             outputs=img_output,
             concurrency_limit=1
         )