Update app.py

app.py

@@ -1,15 +1,8 @@
 import spaces
 import gradio as gr
 import torch
-from PIL import Image
-from transformers import AutoProcessor, AutoModelForCausalLM, pipeline
-from diffusers import DiffusionPipeline
 import random
-import numpy as np
-import os
-import subprocess
-
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+from diffusers import DiffusionPipeline
 
 # Initialize models
 device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -17,74 +10,36 @@ dtype = torch.bfloat16
 
 huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
 
-# FLUX.1-schnell model
-pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=dtype, revision="refs/pr/1", token=huggingface_token).to(device)
-
-# Initialize Florence model
-florence_model = AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to(device).eval()
-florence_processor = AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True)
+# Initialize the base model and move it to GPU
+base_model = "black-forest-labs/FLUX.1-dev"
+pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16, token=huggingface_token).to("cuda")
 
-# Prompt Enhancer
-enhancer_long = pipeline("summarization", model="gokaygokay/Lamini-Prompt-Enchance-Long", device=device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 2048
-
-# Florence caption function
-def florence_caption(image):
-    if not isinstance(image, Image.Image):
-        image = Image.fromarray(image)
-    
-    inputs = florence_processor(text="<MORE_DETAILED_CAPTION>", images=image, return_tensors="pt").to(device)
-    generated_ids = florence_model.generate(
-        input_ids=inputs["input_ids"],
-        pixel_values=inputs["pixel_values"],
-        max_new_tokens=1024,
-        early_stopping=False,
-        do_sample=False,
-        num_beams=3,
-    )
-    generated_text = florence_processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
-    parsed_answer = florence_processor.post_process_generation(
-        generated_text,
-        task="<MORE_DETAILED_CAPTION>",
-        image_size=(image.width, image.height)
-    )
-    return parsed_answer["<MORE_DETAILED_CAPTION>"]
+# Load LoRA weights
+pipe.load_lora_weights("gokaygokay/Flux-Detailer-LoRA")
 
-# Prompt Enhancer function
-def enhance_prompt(input_prompt):
-    result = enhancer_long("Enhance the description: " + input_prompt)
-    enhanced_text = result[0]['summary_text']
-    return enhanced_text
+MAX_SEED = 2**32-1
 
 @spaces.GPU(duration=75)
-def process_workflow(image, text_prompt, use_enhancer, seed, randomize_seed, width, height, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
-    if image is not None:
-        if not isinstance(image, Image.Image):
-            image = Image.fromarray(image)
-        prompt = florence_caption(image)
-    else:
-        prompt = text_prompt
-    
-    if use_enhancer:
-        prompt = enhance_prompt(prompt)
-    
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-    
-    generator = torch.Generator(device=device).manual_seed(seed)
+def generate_image(prompt, steps, seed, cfg_scale, width, height, lora_scale):
+    generator = torch.Generator(device="cuda").manual_seed(seed)
     
     image = pipe(
         prompt=prompt,
-        generator=generator,
-        num_inference_steps=num_inference_steps,
+        num_inference_steps=steps,
+        guidance_scale=cfg_scale,
         width=width,
         height=height,
-        guidance_scale=0.0
+        generator=generator,
+        joint_attention_kwargs={"scale": lora_scale},
     ).images[0]
+    return image
+
+def run_lora(prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_scale):
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
     
-    return image, prompt, seed
+    image = generate_image(prompt, steps, seed, cfg_scale, width, height, lora_scale)
+    return image, seed
 
 custom_css = """
 .input-group, .output-group {
@@ -103,47 +58,39 @@ custom_css = """
 }
 """
 
-title = """<h1 align="center">FLUX.1-schnell with Florence-2 Captioner and Prompt Enhancer</h1>
-<p><center>
-<a href="https://huggingface.co/black-forest-labs/FLUX.1-schnell" target="_blank">[FLUX.1-schnell Model]</a>
-<a href="https://huggingface.co/microsoft/Florence-2-base" target="_blank">[Florence-2 Model]</a>
-<a href="https://huggingface.co/gokaygokay/Lamini-Prompt-Enchance-Long" target="_blank">[Prompt Enhancer Long]</a>
-<p align="center">Create long prompts from images or enhance your short prompts with prompt enhancer</p>
-</center></p>
+title = """<h1 align="center">FLUX Creativity LoRA</h1>
 """
 
-with gr.Blocks(css=custom_css, theme=gr.themes.Soft(primary_hue="blue", secondary_hue="gray")) as demo:
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="gray"), css=custom_css) as app:
     gr.HTML(title)
     
     with gr.Row():
-        with gr.Column(scale=1):
-            with gr.Group(elem_classes="input-group"):
-                input_image = gr.Image(label="Input Image (Florence-2 Captioner)")
-            
-            with gr.Accordion("Advanced Settings", open=False):
-                text_prompt = gr.Textbox(label="Text Prompt (optional, used if no image is uploaded)")
-                use_enhancer = gr.Checkbox(label="Use Prompt Enhancer", value=False)
-                seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
-                randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
-                width = gr.Slider(label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024)
-                height = gr.Slider(label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024)
-                num_inference_steps = gr.Slider(label="Inference Steps", minimum=1, maximum=50, step=1, value=4)
-            
-            generate_btn = gr.Button("Generate Image", elem_classes="submit-btn")
-        
-        with gr.Column(scale=1):
-            with gr.Group(elem_classes="output-group"):
-                output_image = gr.Image(label="Result", elem_id="gallery", show_label=False)
-                final_prompt = gr.Textbox(label="Final Prompt Used")
-                used_seed = gr.Number(label="Seed Used")
+        prompt = gr.Textbox(label="Prompt", lines=3, placeholder="Type your prompt here")
+    
+    with gr.Row():
+        generate_button = gr.Button("Generate", variant="primary")
     
-    generate_btn.click(
-        fn=process_workflow,
-        inputs=[
-            input_image, text_prompt, use_enhancer, seed, randomize_seed,
-            width, height, num_inference_steps
-        ],
-        outputs=[output_image, final_prompt, used_seed]
-    )
+    with gr.Row():
+        result = gr.Image(label="Generated Image")
+    
+    with gr.Accordion("Advanced Settings", open=False):
+        with gr.Row():
+            cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=3.5)
+            steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=28)
+        
+        with gr.Row():
+            width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=1024)
+            height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=1024)
+        
+        with gr.Row():
+            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+            seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
+            lora_scale = gr.Slider(label="LoRA Scale", minimum=0, maximum=1, step=0.01, value=0.95)
+
+    inputs = [prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_scale]
+    outputs = [result, seed]
+
+    generate_button.click(fn=run_lora, inputs=inputs, outputs=outputs)
+    prompt.submit(fn=run_lora, inputs=inputs, outputs=outputs)
 
-demo.launch(debug=True)
+app.launch(debug=True)