Update app.py

app.py

@@ -1,286 +1,125 @@
 import gradio as gr
 import numpy as np
+import random
 import spaces
 import torch
-import random
-import json
-import os
-from PIL import Image
-from diffusers import FluxKontextPipeline
-from diffusers.utils import load_image
-from huggingface_hub import hf_hub_download, HfFileSystem, ModelCard
-from safetensors.torch import load_file
-import requests
-import re
-
-# Load Kontext model from your local path
-MAX_SEED = np.iinfo(np.int32).max
+from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler, FluxTransformer2DModel
+from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
 
-# Use the local path for the base model as in your test.py
-pipe = FluxKontextPipeline.from_pretrained(
-    "black-forest-labs/FLUX.1-Kontext-dev", 
-    torch_dtype=torch.bfloat16
-).to("cuda")
-
-# Load LoRA data from our custom JSON file
-with open("kontext_loras.json", "r") as file:
-    data = json.load(file)
-    # Add default values for keys that might be missing, to prevent errors
-    flux_loras_raw = [
-        {
-            "image": item["image"],
-            "title": item["title"],
-            "repo": item["repo"],
-            "weights": item.get("weights", "pytorch_lora_weights.safetensors"),
-            "prompt": item.get("prompt", f"Turn this image into {item['title']} style."),
-            # The following keys are kept for compatibility with the original demo structure,
-            # but our simplified logic doesn't heavily rely on them.
-            "lora_type": item.get("lora_type", "flux"),
-            "lora_scale_config": item.get("lora_scale", 1.0), # Default scale set to 1.0
-            "prompt_placeholder": item.get("prompt_placeholder", "You can edit the prompt here..."),
-        }
-        for item in data
-    ]
-print(f"Loaded {len(flux_loras_raw)} LoRAs from kontext_loras.json")
-
-def update_selection(selected_state: gr.SelectData, flux_loras):
-    """Update UI when a LoRA is selected"""
-    if selected_state.index >= len(flux_loras):
-        return "### No LoRA selected", gr.update(), None, gr.update()
-    
-    selected_lora = flux_loras[selected_state.index]
-    lora_repo = selected_lora["repo"]
-    default_prompt = selected_lora.get("prompt")
-    
-    updated_text = f"### Selected: [{lora_repo}](https://huggingface.co/{lora_repo})"
+dtype = torch.bfloat16
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
-    optimal_scale = selected_lora.get("lora_scale_config", 1.0)
-    print("Selected Style: ", selected_lora['title'])
-    print("Optimal Scale: ", optimal_scale)
-    return updated_text, gr.update(value=default_prompt), selected_state.index, optimal_scale
-
-# This wrapper is kept for compatibility with the Gradio event triggers
-def infer_with_lora_wrapper(input_image, prompt, selected_index, lora_state, custom_lora, seed=0, guidance_scale=2.5, num_inference_steps=28, lora_scale=1.0, flux_loras=None, progress=gr.Progress(track_tqdm=True)):
-    """Wrapper function to handle state serialization"""
-    # The 'custom_lora' and 'lora_state' arguments are no longer used but kept in the signature
-    return infer_with_lora(input_image, prompt, selected_index, seed, guidance_scale, num_inference_steps, lora_scale, flux_loras, progress)
-
-@spaces.GPU # This decorator is only for Hugging Face Spaces hardware, not needed for local execution
-def infer_with_lora(input_image, prompt, selected_index, seed=0, guidance_scale=2.5, num_inference_steps=28, lora_scale=1.0, flux_loras=None,  progress=gr.Progress(track_tqdm=True)):
-    """Generate image with selected LoRA"""
-    global pipe
-    
-    # The seed is now always taken directly from the input. Randomization has been removed.
-    
-    # Unload any previous LoRA to ensure a clean state
-    if "selected_lora" in pipe.get_active_adapters():
-        pipe.unload_lora_weights()
-    
-    # Determine which LoRA to use from our gallery
-    lora_to_use = None
-    if selected_index is not None and flux_loras and selected_index < len(flux_loras):
-        lora_to_use = flux_loras[selected_index]
-    
-    if lora_to_use:
-        print(f"Applying LoRA: {lora_to_use['title']}")
-        try:
-            # Load LoRA directly from the Hugging Face Hub
-            pipe.load_lora_weights(
-                lora_to_use["repo"], 
-                weight_name=lora_to_use["weights"], 
-                adapter_name="selected_lora"
-            )
-            pipe.set_adapters(["selected_lora"], adapter_weights=[lora_scale])
-            print(f"Loaded {lora_to_use['repo']} with scale {lora_scale}")
-            
-        except Exception as e:
-            print(f"Error loading LoRA: {e}")
+pipe = DiffusionPipeline.from_pretrained("prithivMLmods/Flux.1-krea-Merge-Transformer", torch_dtype=dtype).to(device)
 
-    # Use the prompt from the textbox directly.
-    final_prompt = prompt
-    print(f"Using prompt: {final_prompt}")
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 2048
 
-    input_image = input_image.convert("RGB")
-    
-    try:
-        image = pipe(
-            image=input_image,
-            width=input_image.size[0],
-            height=input_image.size[1],
-            prompt=final_prompt,
-            guidance_scale=guidance_scale,
-            num_inference_steps=num_inference_steps,
-            generator=torch.Generator().manual_seed(seed)
-        ).images[0]
-        
-        # The seed value is no longer returned, as it's not being changed.
-        return image, lora_scale
-    
-    except Exception as e:
-        print(f"Error during inference: {e}")
-        # Return an error state for all outputs
-        return None, lora_scale
+@spaces.GPU()
+def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=8, progress=gr.Progress(track_tqdm=True)):
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    generator = torch.Generator().manual_seed(seed)
+    image = pipe(
+        prompt = prompt, 
+        width = width,
+        height = height,
+        num_inference_steps = num_inference_steps, 
+        generator = generator,
+        guidance_scale=guidance_scale
+    ).images[0] 
+    return image, seed
+ 
+examples = [
+    "a tiny astronaut hatching from an egg on the moon",
+    "a cat holding a sign that says hello world",
+    "an anime illustration of a wiener schnitzel",
+]
 
-# CSS styling
-css = """
-#gen_btn{height: 100%}
-#gen_column{align-self: stretch}
-#main_app {
-    display: flex;
-    gap: 20px;
-}
-#box_column {
-    min-width: 400px;
-}
-#title{text-align: center}
-#title h1{font-size: 3em; display:inline-flex; align-items:center}
-#title img{width: 100px; margin-right: 0.5em}  
-#selected_lora {
-    color: #2563eb;
-    font-weight: bold;
-}
-#prompt {
-    flex-grow: 1;
-}
-#run_button {
-    background: linear-gradient(45deg, #2563eb, #3b82f6);
-    color: white;
-    border: none;
-    padding: 8px 16px;
-    border-radius: 6px;
-    font-weight: bold;
-}
-.custom_lora_card {
-    background: #f8fafc;
-    border: 1px solid #e2e8f0;
-    border-radius: 8px;
-    padding: 12px;
-    margin: 8px 0;
-}
-#gallery{
-    overflow: scroll !important
-}
-/* Custom CSS to ensure the input image is fully visible */
-#input_image_display div[data-testid="image"] img {
-    object-fit: contain !important;
+css="""
+#col-container {
+    margin: 0 auto;
+    max-width: 520px;
 }
 """
 
-# Create Gradio interface
-with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
-    gr_flux_loras = gr.State(value=flux_loras_raw)
+with gr.Blocks(css=css) as demo:
     
-    title = gr.HTML(
-        """<h1>FLUX Kontext Super LoRAs🖖</h1>""",
-        elem_id="title",
-    )
-    selected_state = gr.State(value=None)
-    # The following states are no longer used by the simplified logic but kept for component structure
-    custom_loaded_lora = gr.State(value=None)
-    lora_state = gr.State(value=1.0)
-    
-    with gr.Row(elem_id="main_app"):
-        with gr.Column(scale=4, elem_id="box_column"):
-            with gr.Group(elem_id="gallery_box"):
-                input_image = gr.Image(
-                    label="Upload a picture of yourself", 
-                    type="pil", 
-                    height=300,
-                    elem_id="input_image_display"
-                )
-                gallery = gr.Gallery(
-                    label="Pick a LoRA",
-                    allow_preview=False,
-                    columns=4,
-                    elem_id="gallery",
-                    show_share_button=False,
-                    height=300,
-                    object_fit="contain"
-                )
-                
-                custom_model = gr.Textbox(
-                    label="Or enter a custom HuggingFace FLUX LoRA", 
-                    placeholder="e.g., username/lora-name",
-                    visible=False
-                )
-                custom_model_card = gr.HTML(visible=False)
-                custom_model_button = gr.Button("Remove custom LoRA", visible=False)
+    with gr.Column(elem_id="col-container"):
         
-        with gr.Column(scale=5):
-            with gr.Row():
-                prompt = gr.Textbox(
-                    label="Editing Prompt",
-                    show_label=False,
-                    lines=1,
-                    max_lines=1,
-                    placeholder="opt - describe the person/subject, e.g. 'a man with glasses and a beard'",
-                    elem_id="prompt"
-                )
-                run_button = gr.Button("Generate", variant="primary", elem_id="gen_btn")
+        with gr.Row():
             
-            result = gr.Image(label="Generated Image", interactive=False, height=512)
+            prompt = gr.Text(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt",
+                container=False,
+            )
+            run_button = gr.Button("Run", scale=0)
+        
+        num_inference_steps = gr.Slider(
+            label="Number of inference steps",
+            minimum=1,
+            maximum=50,
+            step=1,
+            value=8,
+        )
+        
+        result = gr.Image(label="Result", show_label=False)
+        
+        with gr.Accordion("Advanced Settings", open=False):
             
-            with gr.Accordion("Advanced Settings", open=False):
-                lora_scale = gr.Slider(
-                    label="LoRA Scale",
-                    minimum=0,
-                    maximum=2,
-                    step=0.1,
-                    value=1.0,
-                    info="Controls the strength of the LoRA effect"
+            seed = gr.Slider(
+                label="Seed",
+                minimum=0,
+                maximum=MAX_SEED,
+                step=1,
+                value=0,
+            )
+            
+            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+            
+            with gr.Row():
+                
+                width = gr.Slider(
+                    label="Width",
+                    minimum=256,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=32,
+                    value=1024,
                 )
-                seed = gr.Slider(
-                    label="Seed",
-                    minimum=0,
-                    maximum=MAX_SEED,
-                    step=1,
-                    value=0,
+                
+                height = gr.Slider(
+                    label="Height",
+                    minimum=256,
+                    maximum=MAX_IMAGE_SIZE,
+                    step=32,
+                    value=1024,
                 )
+            
+            with gr.Row():
+
                 guidance_scale = gr.Slider(
                     label="Guidance Scale",
                     minimum=1,
-                    maximum=10,
+                    maximum=15,
                     step=0.1,
-                    value=2.5,
+                    value=3.5,
                 )
-                num_inference_steps = gr.Slider(
-                    label="Timesteps",
-                    minimum=1,
-                    maximum=100,
-                    step=1,
-                    value=28,
-                    info="Number of inference steps"
-                )
-            
-            prompt_title = gr.Markdown(
-                value="### Click on a LoRA in the gallery to select it",
-                visible=True,
-                elem_id="selected_lora",
-            )
+        
+        gr.Examples(
+            examples = examples,
+            fn = infer,
+            inputs = [prompt],
+            outputs = [result, seed],
+            cache_examples="lazy"
+        )
 
-    # Event handlers
-    # The custom model inputs are no longer needed as we've hidden them.
-    
-    gallery.select(
-        fn=update_selection,
-        inputs=[gr_flux_loras],
-        outputs=[prompt_title, prompt, selected_state, lora_scale],
-        show_progress=False
-    )
-    
     gr.on(
         triggers=[run_button.click, prompt.submit],
-        fn=infer_with_lora_wrapper,
-        inputs=[input_image, prompt, selected_state, lora_state, custom_loaded_lora, seed, guidance_scale, num_inference_steps, lora_scale, gr_flux_loras],
-        outputs=[result, lora_state]
-    )
-    
-    # Initialize gallery
-    demo.load(
-        fn=lambda loras: ([(item["image"], item["title"]) for item in loras], loras), 
-        inputs=[gr_flux_loras], 
-        outputs=[gallery, gr_flux_loras]
+        fn = infer,
+        inputs = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
+        outputs = [result, seed]
     )
 
-demo.queue(default_concurrency_limit=None)
 demo.launch()