Create app.py

app.py

@@ -1,154 +1,132 @@
 import gradio as gr
-import numpy as np
-import random
-
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
 import torch
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-
-    generator = torch.Generator().manual_seed(seed)
-
+from diffusers import StableDiffusionXLPipeline, EulerDiscreteScheduler
+from huggingface_hub import hf_hub_download
+
+# --- Configuration ---
+# The base model your LoRA was trained on.
+base_model_id = "stabilityai/stable-diffusion-xl-base-1.0"
+
+# The path to your LoRA file on the Hugging Face Hub.
+lora_repo_id = "TuringsSolutions/EmilyH"
+lora_filename = "emilyh.safetensors"
+
+# --- Load the Pipeline ---
+# Use a recommended VAE for SDXL
+vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    base_model_id,
+    vae=vae,
+    torch_dtype=torch.float16,
+    variant="fp16",
+    use_safetensors=True
+)
+
+# --- Load and Fuse the LoRA ---
+# Download the LoRA file and load the state dict.
+lora_file_path = hf_hub_download(repo_id=lora_repo_id, filename=lora_filename)
+pipe.load_lora_weights(lora_file_path)
+
+# It's recommended to fuse the LoRA weights for better performance,
+# but this is optional. You can also use pipe.set_adapters(["default"], adapter_weights=[0.9])
+# during inference if you prefer more dynamic control.
+# pipe.fuse_lora(lora_scale=0.9) # Fusing is more efficient
+
+# Move the pipeline to the GPU
+pipe.to("cuda")
+
+# --- Default Settings from your Recommendations ---
+# These are pulled directly from your "Recomendations.txt". 
+default_positive_prompt = "masterpiece, best quality, ultra-detailed, realistic skin, intricate details, highres" # 
+default_negative_prompt = "low quality, worst quality, blurry, (deformed:1.3), extra fingers, cartoon, 3d, anime, bad anatomy" # 
+default_sampler = "DPM++ 2M Karras" # 
+default_cfg = 6.0 # 
+default_steps = 30 # 
+trigger_word = "emilyh" # 
+lora_tag_main = "<lora:emilyh:0.9>" # 
+
+
+# --- Define the Inference Function ---
+def generate_image(prompt, negative_prompt, sampler, steps, cfg, width, height, seed):
+    """
+    Function to generate an image based on user inputs.
+    """
+    # Combine the user prompt with the trigger word and LoRA tag 
+    full_prompt = f"{lora_tag_main}, {trigger_word}, {prompt}"
+
+    # Set the scheduler (sampler)
+    if sampler == "DPM++ 2M Karras":
+        pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
+    elif sampler == "DPM++ SDE Karras":
+        pipe.scheduler = DPMSolverSinglestepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True, algorithm_type="sde-dpmsolver++")
+    else: # Default to DPM++ 2M Karras
+        pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
+
+    # Set seed for reproducibility
+    generator = torch.Generator("cuda").manual_seed(seed) if seed != -1 else None
+
+    # Generate the image
     image = pipe(
-        prompt=prompt,
+        prompt=full_prompt,
         negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
         width=width,
         height=height,
+        guidance_scale=cfg,
+        num_inference_steps=steps,
         generator=generator,
+        cross_attention_kwargs={"scale": 0.9} # This is an alternative way to apply LoRA scale if not fused
     ).images[0]
 
-    return image, seed
-
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
-
-        with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
+    return image
 
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+# --- Create the Gradio Interface ---
+with gr.Blocks(css="style.css") as demo:
+    gr.Markdown("# `emilyh` LoRA Image Generator")
+    gr.Markdown(
+        "A Gradio interface for the `emilyh` LoRA. "
+        "Based on the recommendations provided. "
+    )
 
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Textbox(label="Positive Prompt", value=default_positive_prompt, lines=3)
+            negative_prompt = gr.Textbox(label="Negative Prompt", value=default_negative_prompt, lines=3)
+            
             with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
+                sampler = gr.Radio(
+                    label="Sampler",
+                    choices=["DPM++ 2M Karras", "DPM++ SDE Karras"],
+                    value=default_sampler,
+                ) # 
+                steps = gr.Slider(label="Steps", minimum=15, maximum=50, value=default_steps, step=1) # 
+            
+            cfg = gr.Slider(label="CFG Scale", minimum=1.0, maximum=10.0, value=default_cfg, step=0.5) # 
 
             with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
+                width = gr.Slider(label="Width", minimum=512, maximum=1024, value=1024, step=64)
+                height = gr.Slider(label="Height", minimum=512, maximum=1024, value=1024, step=64)
+
+            seed = gr.Slider(label="Seed", minimum=-1, maximum=999999999, step=1, value=-1, info="Use -1 for a random seed.")
+            
+            generate_button = gr.Button("Generate Image", variant="primary")
+
+        with gr.Column():
+            output_image = gr.Image(label="Generated Image", type="pil")
+            gr.Markdown(
+                """
+                ### 🔧 Usage Guide 
+                * The trigger word `emilyh` and the LoRA tag `<lora:emilyh:0.9>` are automatically added to your prompt. 
+                * For best results, generate images in batches and choose the most consistent ones. 
+                * The LoRA captures the subject's appearance well across various poses and outfits. 
+                * A weight of 0.9 provides a good balance of likeness and flexibility.  Using a weight closer to 1.0 can increase consistency but may cause stiffness. 
+                * This interface does not include ADetailer, which is recommended for final face refinement. 
+                """
+            )
+
+    generate_button.click(
+        fn=generate_image,
+        inputs=[prompt, negative_prompt, sampler, steps, cfg, width, height, seed],
+        outputs=output_image
     )
 
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()