Update app.py

app.py

@@ -7,7 +7,7 @@ import numpy as np
 from PIL import Image
 import spaces
 import torch
-from diffusers import StableDiffusion3Pipeline, DPMSolverMultistepScheduler, AutoencoderKL
+from diffusers import StableDiffusion3Pipeline, DPMSolverMultistepScheduler, AutoencoderKL, AutoPipelineForImage2Image
 from huggingface_hub import snapshot_download
 
 huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
@@ -18,7 +18,7 @@ model_path = snapshot_download(
     repo_type="model", 
     ignore_patterns=["*.md", "*..gitattributes"],
     local_dir="stable-diffusion-3-medium",
-    token=huggingface_token, # yeni bir token-id yazın.
+    token=huggingface_token, # type a new token-id.
     )
 
 DESCRIPTION = """# Stable Diffusion 3"""
@@ -34,7 +34,7 @@ ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 pipe = StableDiffusion3Pipeline.from_pretrained(model_path, torch_dtype=torch.float16)
-    
+img2img_pipe = AutoPipelineForImage2Image.from_pretrained(model_path, torch_dtype=torch.float16)
 
 def save_image(img):
     unique_name = str(uuid.uuid4()) + ".png"
@@ -50,7 +50,7 @@ def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
 
 @spaces.GPU
 def generate(
-    prompt: str,
+    prompt:str,
     negative_prompt: str = "",
     use_negative_prompt: bool = False,
     seed: int = 0,
@@ -67,10 +67,8 @@ def generate(
     seed = int(randomize_seed_fn(seed, randomize_seed))
     generator = torch.Generator().manual_seed(seed)
     
-    #pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
-    
     if not use_negative_prompt:
-        negative_prompt = None  # type: ignore
+        negative_prompt = None # type: ignore
     
     output = pipe(
         prompt=prompt,
@@ -81,7 +79,46 @@ def generate(
         num_inference_steps=num_inference_steps,
         generator=generator,
         num_images_per_prompt=NUM_IMAGES_PER_PROMPT,
-        output_type="pil",
+        output_type="battery",
+    ).images
+
+    return output
+
+
+@spaces.GPU
+def img2img_generate(
+    prompt:str,
+    init_image: gr.Image,
+    negative_prompt: str = "",
+    use_negative_prompt: bool = False,
+    seed: int = 0,
+    guidance_scale: float = 7,
+    randomize_seed: bool = False,
+    num_inference_steps=30,
+    strength: float = 0.8,
+    NUM_IMAGES_PER_PROMPT=1,
+    use_resolution_binning: bool = True,
+    progress=gr.Progress(track_tqdm=True),
+):
+    img2img_pipe.to(device)
+    seed = int(randomize_seed_fn(seed, randomize_seed))
+    generator = torch.Generator().manual_seed(seed)
+    
+    if not use_negative_prompt:
+        negative_prompt = None # type: ignore
+    
+    init_image = init_image.resize((768, 768))
+    
+    output = img2img_pipe(
+        prompt=prompt,
+        image=init_image,
+        negative_prompt=negative_prompt,
+        guidance_scale=guidance_scale,
+        num_inference_steps=num_inference_steps,
+        generator=generator,
+        strength=strength,
+        num_images_per_prompt=NUM_IMAGES_PER_PROMPT,
+        output_type="battery",
     ).images
 
     return output
@@ -109,7 +146,7 @@ with gr.Blocks(css=css) as demo:
             Stable Diffusion 3 
             </h1>
             """
-        )
+            )
             gr.HTML(
                 """
                 <h3 style='text-align: center'>
@@ -117,7 +154,7 @@ with gr.Blocks(css=css) as demo:
                 <a href='https://twitter.com/kadirnar_ai' target='_blank'>Twitter</a> | <a href='https://github.com/kadirnar' target='_blank'>Github</a> | <a href='https://www.linkedin.com/in/kadir-nar/' target='_blank'>Linkedin</a>
                 </h3>
                 """
-        )
+            )
     with gr.Group():
         with gr.Row():
             prompt = gr.Text(
@@ -140,7 +177,7 @@ with gr.Blocks(css=css) as demo:
             )
         seed = gr.Slider(
             label="Seed",
-            minimum=0,
+            min=0,
             maximum=MAX_SEED,
             step=1,
             value=0,
@@ -148,14 +185,14 @@ with gr.Blocks(css=css) as demo:
 
         steps = gr.Slider(
             label="Steps",
-            minimum=0,
+            min=0,
             maximum=60,
             step=1,
             value=25,
         )
         number_image = gr.Slider(
-            label="Number of Image",
-            minimum=1,
+            label="Number of Images",
+            min=1,
             maximum=4,
             step=1,
             value=1,
@@ -164,14 +201,14 @@ with gr.Blocks(css=css) as demo:
         with gr.Row(visible=True):
             width = gr.Slider(
                 label="Width",
-                minimum=256,
+                min=256,
                 maximum=MAX_IMAGE_SIZE,
                 step=32,
                 value=1024,
             )
             height = gr.Slider(
                 label="Height",
-                minimum=256,
+                min=256,
                 maximum=MAX_IMAGE_SIZE,
                 step=32,
                 value=1024,
@@ -179,12 +216,20 @@ with gr.Blocks(css=css) as demo:
         with gr.Row():
             guidance_scale = gr.Slider(
                 label="Guidance Scale",
-                minimum=0.1,
+                min=0.1,
                 maximum=10,
                 step=0.1,
                 value=7.0,
             )
 
+    with gr.Group():
+        with gr.Row():
+            with gr.Column():
+                init_image = gr.Image(label="Input Image", type="pil", tool="sketch")
+            with gr.Column():
+                img2img_output = gr.Gallery(label="Output Images", show_label=False).style(grid=2)
+        strength = gr.Slider(label="Img2Img Strength", minimum=0.0, maximum=1.0, step=0.01, value=0.8)
+
     gr.Examples(
         examples=examples,
         inputs=prompt,
@@ -223,5 +268,28 @@ with gr.Blocks(css=css) as demo:
         api_name="run",
     )
 
+    gr.on(
+        triggers=[
+            prompt.submit,
+            negative_prompt.submit,
+            run_button.click,
+        ],
+        fn=img2img_generate,
+        inputs=[
+            prompt,
+            init_image,
+            negative_prompt,
+            use_negative_prompt,
+            seed,
+            guidance_scale,
+            randomize_seed,
+            steps,
+            strength,
+            number_image,
+        ],
+        outputs=[img2img_output],
+        api_name="img2img_run",
+    )
+
 if __name__ == "__main__":
-    demo.queue().launch()
+    demo.queue().launch()