Update app.py

app.py

@@ -1,349 +1,415 @@
 import os
-import random
-import uuid
+import gradio as gr
 import json
+import logging
+import torch
+from PIL import Image
+import spaces
+from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
+from diffusers.utils import load_image
+from huggingface_hub import hf_hub_download, HfFileSystem, ModelCard
+import copy
+import random
 import time
-import asyncio
 import re
-from threading import Thread
 
-import gradio as gr
-import spaces
-import torch
-import numpy as np
-from PIL import Image
-import edge_tts
+# Load LoRAs from JSON file
+with open('loras.json', 'r') as f:
+    loras = json.load(f)
 
-from transformers import (
-    AutoModelForCausalLM,
-    AutoTokenizer,
-    TextIteratorStreamer,
-    Qwen2VLForConditionalGeneration,
-    AutoProcessor,
-)
-from transformers.image_utils import load_image
-from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
+# Initialize the base model for SDXL
+dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+device = "cuda" if torch.cuda.is_available() else "cpu"
+base_model = "stabilityai/stable-diffusion-xl-base-1.0"
 
-MAX_MAX_NEW_TOKENS = 2048
-DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+# Load SDXL pipelines
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    base_model,
+    torch_dtype=dtype,
+    use_safetensors=True
+).to(device)
 
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+pipe_i2i = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+    base_model,
+    torch_dtype=dtype,
+    use_safetensors=True
+).to(device)
 
-# -----------------------
-# Progress Bar Helper
-# -----------------------
-def progress_bar_html(label: str) -> str:
-    """
-    Returns an HTML snippet for a thin progress bar with a label.
-    The progress bar is styled as a dark red animated bar.
-    """
-    return f'''
-<div style="display: flex; align-items: center;">
-    <span style="margin-right: 10px; font-size: 14px;">{label}</span>
-    <div style="width: 110px; height: 5px; background-color: #DDA0DD; border-radius: 2px; overflow: hidden;">
-        <div style="width: 100%; height: 100%; background-color: #FF00FF; animation: loading 1.5s linear infinite;"></div>
-    </div>
-</div>
-<style>
-@keyframes loading {{
-    0% {{ transform: translateX(-100%); }}
-    100% {{ transform: translateX(100%); }}
-}}
-</style>
-    '''
+MAX_SEED = 2**32 - 1
 
-# -----------------------
-# Text Generation Setup
-# -----------------------
-model_id = "prithivMLmods/DeepHermes-3-Llama-3-3B-Preview-abliterated"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    device_map="auto",
-    torch_dtype=torch.bfloat16,
-)
-model.eval()
+# Custom SDXL generation function for live preview
+@torch.inference_mode()
+def generate_sdxl_images(
+    pipe,
+    prompt: str,
+    height: int = 1024,
+    width: int = 1024,
+    num_inference_steps: int = 50,
+    guidance_scale: float = 7.5,
+    generator: Optional[torch.Generator] = None,
+    output_type: str = "pil",
+):
+    # Encode prompt
+    prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds = pipe.encode_prompt(
+        prompt=prompt,
+        num_images_per_prompt=1,
+        do_classifier_free_guidance=True,
+    )
+    # Prepare latents
+    latents = pipe.prepare_latents(
+        batch_size=1,
+        num_channels_latents=pipe.unet.config.in_channels,
+        height=height,
+        width=width,
+        dtype=prompt_embeds.dtype,
+        device=pipe.device,
+        generator=generator,
+    )
+    # Prepare timesteps
+    pipe.scheduler.set_timesteps(num_inference_steps, device=pipe.device)
+    timesteps = pipe.scheduler.timesteps
+    # Prepare guidance
+    do_classifier_free_guidance = guidance_scale > 1.0
+    if do_classifier_free_guidance:
+        prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+        pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds])
+    # Denoising loop
+    for i, t in enumerate(timesteps):
+        # Expand latents for guidance
+        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+        # Predict noise
+        noise_pred = pipe.unet(
+            latent_model_input,
+            t,
+            encoder_hidden_states=prompt_embeds,
+            added_cond_kwargs={"text_embeds": pooled_prompt_embeds},
+        ).sample
+        # Perform guidance
+        if do_classifier_free_guidance:
+            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+            noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+        # Step scheduler
+        latents = pipe.scheduler.step(noise_pred, t, latents).prev_sample
+        # Decode latents to image every step
+        image = pipe.vae.decode(latents / pipe.vae.config.scaling_factor, return_dict=False)[0]
+        yield pipe.image_processor.postprocess(image, output_type=output_type)[0]
+    # Final image
+    image = pipe.vae.decode(latents / pipe.vae.config.scaling_factor, return_dict=False)[0]
+    yield pipe.image_processor.postprocess(image, output_type=output_type)[0]
 
-TTS_VOICES = [
-    "en-US-JennyNeural",  # @tts1
-    "en-US-GuyNeural",    # @tts2
-]
+class calculateDuration:
+    def __init__(self, activity_name=""):
+        self.activity_name = activity_name
 
-# -----------------------
-# Multimodal OCR Setup
-# -----------------------
-MODEL_ID = "prithivMLmods/Qwen2-VL-OCR2-2B-Instruct" 
-processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
-model_m = Qwen2VLForConditionalGeneration.from_pretrained(
-    MODEL_ID,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to("cuda").eval()
+    def __enter__(self):
+        self.start_time = time.time()
+        return self
+    
+    def __exit__(self, exc_type, exc_value, traceback):
+        self.end_time = time.time()
+        self.elapsed_time = self.end_time - self.start_time
+        if self.activity_name:
+            print(f"Elapsed time for {self.activity_name}: {self.elapsed_time:.6f} seconds")
+        else:
+            print(f"Elapsed time: {self.elapsed_time:.6f} seconds")
 
-async def text_to_speech(text: str, voice: str, output_file="output.mp3"):
-    """Convert text to speech using Edge TTS and save as MP3"""
-    communicate = edge_tts.Communicate(text, voice)
-    await communicate.save(output_file)
-    return output_file
+def update_selection(evt: gr.SelectData, width, height):
+    selected_lora = loras[evt.index]
+    new_placeholder = f"Type a prompt for {selected_lora['title']}"
+    lora_repo = selected_lora["repo"]
+    updated_text = f"### Selected: [{lora_repo}](https://huggingface.co/{lora_repo}) ✨"
+    if "aspect" in selected_lora:
+        if selected_lora["aspect"] == "portrait":
+            width = 768
+            height = 1024
+        elif selected_lora["aspect"] == "landscape":
+            width = 1024
+            height = 768
+        else:
+            width = 1024
+            height = 1024
+    return (
+        gr.update(placeholder=new_placeholder),
+        updated_text,
+        evt.index,
+        width,
+        height,
+    )
 
-def clean_chat_history(chat_history):
-    """
-    Filter out any chat entries whose "content" is not a string.
-    """
-    cleaned = []
-    for msg in chat_history:
-        if isinstance(msg, dict) and isinstance(msg.get("content"), str):
-            cleaned.append(msg)
-    return cleaned
+@spaces.GPU(duration=70)
+def generate_image(prompt_mash, steps, seed, cfg_scale, width, height, progress):
+    pipe.to("cuda")
+    generator = torch.Generator(device="cuda").manual_seed(seed)
+    with calculateDuration("Generating image"):
+        for img in generate_sdxl_images(
+            pipe,
+            prompt=prompt_mash,
+            num_inference_steps=steps,
+            guidance_scale=cfg_scale,
+            width=width,
+            height=height,
+            generator=generator,
+            output_type="pil",
+        ):
+            yield img
 
-# -----------------------
-# Stable Diffusion Image Generation Setup
-# -----------------------
+def generate_image_to_image(prompt_mash, image_input_path, image_strength, steps, cfg_scale, width, height, seed):
+    generator = torch.Generator(device="cuda").manual_seed(seed)
+    pipe_i2i.to("cuda")
+    image_input = load_image(image_input_path)
+    final_image = pipe_i2i(
+        prompt=prompt_mash,
+        image=image_input,
+        strength=image_strength,
+        num_inference_steps=steps,
+        guidance_scale=cfg_scale,
+        width=width,
+        height=height,
+        generator=generator,
+        output_type="pil",
+    ).images[0]
+    return final_image 
 
-MAX_SEED = np.iinfo(np.int32).max
-USE_TORCH_COMPILE = False
-ENABLE_CPU_OFFLOAD = False
+@spaces.GPU(duration=70)
+def run_lora(prompt, image_input, image_strength, cfg_scale, steps, selected_index, randomize_seed, seed, width, height, lora_scale, progress=gr.Progress(track_tqdm=True)):
+    if selected_index is None:
+        raise gr.Error("You must select a LoRA before proceeding.")
+    selected_lora = loras[selected_index]
+    lora_path = selected_lora["repo"]
+    trigger_word = selected_lora["trigger_word"]
+    if trigger_word:
+        if "trigger_position" in selected_lora and selected_lora["trigger_position"] == "prepend":
+            prompt_mash = f"{trigger_word} {prompt}"
+        else:
+            prompt_mash = f"{prompt} {trigger_word}"
+    else:
+        prompt_mash = prompt
 
-if torch.cuda.is_available():
-    pipe = StableDiffusionXLPipeline.from_pretrained(
-        "SG161222/RealVisXL_V4.0_Lightning",
-        torch_dtype=torch.float16,
-        use_safetensors=True,
-    )
-    pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-    
-    # LoRA options with one example for each.
-    LORA_OPTIONS = {
-        "Realism": ("prithivMLmods/Canopus-Realism-LoRA", "Canopus-Realism-LoRA.safetensors", "rlms"),
-        "Pixar": ("prithivMLmods/Canopus-Pixar-Art", "Canopus-Pixar-Art.safetensors", "pixar"),
-        "Photoshoot": ("prithivMLmods/Canopus-Photo-Shoot-Mini-LoRA", "Canopus-Photo-Shoot-Mini-LoRA.safetensors", "photo"),
-        "Clothing": ("prithivMLmods/Canopus-Clothing-Adp-LoRA", "Canopus-Dress-Clothing-LoRA.safetensors", "clth"),
-        "Interior": ("prithivMLmods/Canopus-Interior-Architecture-0.1", "Canopus-Interior-Architecture-0.1δ.safetensors", "arch"),
-        "Fashion": ("prithivMLmods/Canopus-Fashion-Product-Dilation", "Canopus-Fashion-Product-Dilation.safetensors", "fashion"),
-        "Minimalistic": ("prithivMLmods/Pegasi-Minimalist-Image-Style", "Pegasi-Minimalist-Image-Style.safetensors", "minimalist"),
-        "Modern": ("prithivMLmods/Canopus-Modern-Clothing-Design", "Canopus-Modern-Clothing-Design.safetensors", "mdrnclth"),
-        "Animaliea": ("prithivMLmods/Canopus-Animaliea-Artism", "Canopus-Animaliea-Artism.safetensors", "Animaliea"),
-        "Wallpaper": ("prithivMLmods/Canopus-Liquid-Wallpaper-Art", "Canopus-Liquid-Wallpaper-Minimalize-LoRA.safetensors", "liquid"),
-        "Cars": ("prithivMLmods/Canes-Cars-Model-LoRA", "Canes-Cars-Model-LoRA.safetensors", "car"),
-        "PencilArt": ("prithivMLmods/Canopus-Pencil-Art-LoRA", "Canopus-Pencil-Art-LoRA.safetensors", "Pencil Art"),
-        "ArtMinimalistic": ("prithivMLmods/Canopus-Art-Medium-LoRA", "Canopus-Art-Medium-LoRA.safetensors", "mdm"),
-    }
+    # Unload previous LoRA weights
+    with calculateDuration("Unloading LoRA"):
+        pipe.unload_lora_weights()
+        pipe_i2i.unload_lora_weights()
+        
+    # Load LoRA weights and set adapter scale
+    with calculateDuration(f"Loading LoRA weights for {selected_lora['title']}"):
+        weight_name = selected_lora.get("weights", None)
+        adapter_name = "lora"
+        pipe.load_lora_weights(lora_path, weight_name=weight_name, adapter_name=adapter_name)
+        pipe.set_adapters([adapter_name], [lora_scale])
+        pipe_i2i.load_lora_weights(lora_path, weight_name=weight_name, adapter_name=adapter_name)
+        pipe_i2i.set_adapters([adapter_name], [lora_scale])
+                
+    # Set random seed
+    with calculateDuration("Randomizing seed"):
+        if randomize_seed:
+            seed = random.randint(0, MAX_SEED)
+            
+    if image_input is not None:
+        final_image = generate_image_to_image(prompt_mash, image_input, image_strength, steps, cfg_scale, width, height, seed)
+        yield final_image, seed, gr.update(visible=False)
+    else:
+        image_generator = generate_image(prompt_mash, steps, seed, cfg_scale, width, height, progress)
+        final_image = None
+        step_counter = 0
+        for image in image_generator:
+            step_counter += 1
+            final_image = image
+            progress_bar = f'<div class="progress-container"><div class="progress-bar" style="--current: {step_counter}; --total: {steps};"></div></div>'
+            yield image, seed, gr.update(value=progress_bar, visible=True)
+        yield final_image, seed, gr.update(value=progress_bar, visible=False)
 
-    # Load all LoRA weights
-    for model_name, weight_name, adapter_name in LORA_OPTIONS.values():
-        pipe.load_lora_weights(model_name, weight_name=weight_name, adapter_name=adapter_name)
-    pipe.to("cuda")
-else:
-    pipe = StableDiffusionXLPipeline.from_pretrained(
-        "SG161222/RealVisXL_V4.0_Lightning",
-        torch_dtype=torch.float32,
-        use_safetensors=True,
-    ).to(device)
+def get_huggingface_safetensors(link):
+    split_link = link.split("/")
+    if len(split_link) != 2:
+        raise Exception("Invalid Hugging Face repository link format.")
 
-def save_image(img: Image.Image) -> str:
-    """Save a PIL image with a unique filename and return the path."""
-    unique_name = str(uuid.uuid4()) + ".png"
-    img.save(unique_name)
-    return unique_name
+    # Load model card
+    model_card = ModelCard.load(link)
+    base_model = model_card.data.get("base_model")
+    print(base_model)
 
-def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-    return seed
+    # Validate model type for SDXL
+    if base_model != "stabilityai/stable-diffusion-xl-base-1.0":
+        raise Exception("Not an SDXL LoRA!")
 
-@spaces.GPU(duration=180, enable_queue=True)
-def generate_image(
-    prompt: str,
-    negative_prompt: str = "",
-    seed: int = 0,
-    width: int = 1024,
-    height: int = 1024,
-    guidance_scale: float = 3.0,
-    randomize_seed: bool = True,
-    lora_model: str = "Realism",
-    progress=gr.Progress(track_tqdm=True),
-):
-    seed = int(randomize_seed_fn(seed, randomize_seed))
-    effective_negative_prompt = negative_prompt  # Use provided negative prompt if any
-    model_name, weight_name, adapter_name = LORA_OPTIONS[lora_model]
-    pipe.set_adapters(adapter_name)
-    outputs = pipe(
-         prompt=prompt,
-         negative_prompt=effective_negative_prompt,
-         width=width,
-         height=height,
-         guidance_scale=guidance_scale,
-         num_inference_steps=28,
-         num_images_per_prompt=1,
-         cross_attention_kwargs={"scale": 0.65},
-         output_type="pil",
-    )
-    images = outputs.images
-    image_paths = [save_image(img) for img in images]
-    return image_paths, seed
+    # Extract image and trigger word
+    image_path = model_card.data.get("widget", [{}])[0].get("output", {}).get("url", None)
+    trigger_word = model_card.data.get("instance_prompt", "")
+    image_url = f"https://huggingface.co/{link}/resolve/main/{image_path}" if image_path else None
 
-# -----------------------
-# Main Chat/Generation Function
-# -----------------------
-@spaces.GPU
-def generate(
-    input_dict: dict,
-    chat_history: list[dict],
-    max_new_tokens: int = 1024,
-    temperature: float = 0.6,
-    top_p: float = 0.9,
-    top_k: int = 50,
-    repetition_penalty: float = 1.2,
-):
-    """
-    Generates chatbot responses with support for multimodal input, TTS, and image generation.
-    Special commands:
-      - "@tts1" or "@tts2": triggers text-to-speech.
-      - "@<lora_command>": triggers image generation using the new LoRA pipeline.
-         Available commands (case-insensitive): @realism, @pixar, @photoshoot, @clothing, @interior, @fashion, 
-         @minimalistic, @modern, @animaliea, @wallpaper, @cars, @pencilart, @artminimalistic.
-    """
-    text = input_dict["text"]
-    files = input_dict.get("files", [])
-    
-    # Check for image generation command based on LoRA tags.
-    lora_mapping = { key.lower(): key for key in LORA_OPTIONS }
-    for key_lower, key in lora_mapping.items():
-        command_tag = "@" + key_lower
-        if text.strip().lower().startswith(command_tag):
-            prompt_text = text.strip()[len(command_tag):].strip()
-            yield progress_bar_html(f"Processing Image Generation ({key} style)")
-            image_paths, used_seed = generate_image(
-                prompt=prompt_text,
-                negative_prompt="",
-                seed=1,
-                width=1024,
-                height=1024,
-                guidance_scale=3,
-                randomize_seed=True,
-                lora_model=key,
-            )
-            yield progress_bar_html("Finalizing Image Generation")
-            yield gr.Image(image_paths[0])
-            return 
-    
-    # Check for TTS command (@tts1 or @tts2)
-    tts_prefix = "@tts"
-    is_tts = any(text.strip().lower().startswith(f"{tts_prefix}{i}") for i in range(1, 3))
-    voice_index = next((i for i in range(1, 3) if text.strip().lower().startswith(f"{tts_prefix}{i}")), None)
-    
-    if is_tts and voice_index:
-        voice = TTS_VOICES[voice_index - 1]
-        text = text.replace(f"{tts_prefix}{voice_index}", "").strip()
-        conversation = [{"role": "user", "content": text}]
-    else:
-        voice = None
-        text = text.replace(tts_prefix, "").strip()
-        conversation = clean_chat_history(chat_history)
-        conversation.append({"role": "user", "content": text})
-    
-    if files:
-        if len(files) > 1:
-            images = [load_image(image) for image in files]
-        elif len(files) == 1:
-            images = [load_image(files[0])]
-        else:
-            images = []
-        messages = [{
-            "role": "user",
-            "content": [
-                *[{"type": "image", "image": image} for image in images],
-                {"type": "text", "text": text},
-            ]
-        }]
-        prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-        inputs = processor(text=[prompt], images=images, return_tensors="pt", padding=True).to("cuda")
-        streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-        generation_kwargs = {**inputs, "streamer": streamer, "max_new_tokens": max_new_tokens}
-        thread = Thread(target=model_m.generate, kwargs=generation_kwargs)
-        thread.start()
+    # Initialize Hugging Face file system
+    fs = HfFileSystem()
+    try:
+        list_of_files = fs.ls(link, detail=False)
+        safetensors_name = None
+        highest_trained_file = None
+        highest_steps = -1
+        last_safetensors_file = None
+        step_pattern = re.compile(r"_0{3,}\d+")  # Detects step count `_000...`
+
+        for file in list_of_files:
+            filename = file.split("/")[-1]
+            if filename.endswith(".safetensors"):
+                last_safetensors_file = filename
+                match = step_pattern.search(filename)
+                if not match:
+                    safetensors_name = filename
+                    break
+                else:
+                    steps = int(match.group().lstrip("_"))
+                    if steps > highest_steps:
+                        highest_trained_file = filename
+                        highest_steps = steps
+            if not image_url and filename.lower().endswith((".jpg", ".jpeg", ".png", ".webp")):
+                image_url = f"https://huggingface.co/{link}/resolve/main/{filename}"
+
+        if not safetensors_name:
+            safetensors_name = highest_trained_file if highest_trained_file else last_safetensors_file
+        if not safetensors_name:
+            raise Exception("No valid *.safetensors file found in the repository.")
+    except Exception as e:
+        print(e)
+        raise Exception("You didn't include a valid Hugging Face repository with a *.safetensors LoRA")
 
-        buffer = ""
-        yield progress_bar_html("Processing with Qwen2VL Ocr")
-        for new_text in streamer:
-            buffer += new_text
-            buffer = buffer.replace("<|im_end|>", "")
-            time.sleep(0.01)
-            yield buffer
+    return split_link[1], link, safetensors_name, trigger_word, image_url
+
+def check_custom_model(link):
+    if link.startswith("https://"):
+        if link.startswith("https://huggingface.co") or link.startswith("https://www.huggingface.co"):
+            link_split = link.split("huggingface.co/")
+            return get_huggingface_safetensors(link_split[1])
+    else: 
+        return get_huggingface_safetensors(link)
+
+def add_custom_lora(custom_lora):
+    global loras
+    if custom_lora:
+        try:
+            title, repo, path, trigger_word, image = check_custom_model(custom_lora)
+            print(f"Loaded custom LoRA: {repo}")
+            card = f'''
+            <div class="custom_lora_card">
+              <span>Loaded custom LoRA:</span>
+              <div class="card_internal">
+                <img src="{image}" />
+                <div>
+                    <h3>{title}</h3>
+                    <small>{"Using: <code><b>"+trigger_word+"</code></b> as the trigger word" if trigger_word else "No trigger word found. If there's a trigger word, include it in your prompt"}<br></small>
+                </div>
+              </div>
+            </div>
+            '''
+            existing_item_index = next((index for (index, item) in enumerate(loras) if item['repo'] == repo), None)
+            if not existing_item_index:
+                new_item = {
+                    "image": image,
+                    "title": title,
+                    "repo": repo,
+                    "weights": path,
+                    "trigger_word": trigger_word
+                }
+                print(new_item)
+                existing_item_index = len(loras)
+                loras.append(new_item)
+            return gr.update(visible=True, value=card), gr.update(visible=True), gr.Gallery(selected_index=None), f"Custom: {path}", existing_item_index, trigger_word
+        except Exception as e:
+            gr.Warning(f"Invalid LoRA: either you entered an invalid link, or a non-SDXL LoRA")
+            return gr.update(visible=True, value=f"Invalid LoRA: either you entered an invalid link, a non-SDXL LoRA"), gr.update(visible=True), gr.update(), "", None, ""
     else:
-        input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
-        if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
-            input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
-            gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
-        input_ids = input_ids.to(model.device)
-        streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
-        generation_kwargs = {
-            "input_ids": input_ids,
-            "streamer": streamer,
-            "max_new_tokens": max_new_tokens,
-            "do_sample": True,
-            "top_p": top_p,
-            "top_k": top_k,
-            "temperature": temperature,
-            "num_beams": 1,
-            "repetition_penalty": repetition_penalty,
-        }
-        t = Thread(target=model.generate, kwargs=generation_kwargs)
-        t.start()
+        return gr.update(visible=False), gr.update(visible=False), gr.update(), "", None, ""
+
+def remove_custom_lora():
+    return gr.update(visible=False), gr.update(visible=False), gr.update(), "", None, ""
 
-        outputs = []
-        for new_text in streamer:
-            outputs.append(new_text)
-            yield "".join(outputs)
+run_lora.zerogpu = True
 
-        final_response = "".join(outputs)
-        yield final_response
+css = '''
+#gen_btn{height: 100%}
+#gen_column{align-self: stretch}
+#title{text-align: center}
+#title h1{font-size: 3em; display:inline-flex; align-items:center}
+#title img{width: 100px; margin-right: 0.5em}
+#gallery .grid-wrap{height: 10vh}
+#lora_list{background: var(--block-background-fill);padding: 0 1em .3em; font-size: 90%}
+.card_internal{display: flex;height: 100px;margin-top: .5em}
+.card_internal img{margin-right: 1em}
+.styler{--form-gap-width: 0px !important}
+#progress{height:30px}
+#progress .generating{display:none}
+.progress-container {width: 100%;height: 30px;background-color: #f0f0f0;border-radius: 15px;overflow: hidden;margin-bottom: 20px}
+.progress-bar {height: 100%;background-color: #4f46e5;width: calc(var(--current) / var(--total) * 100%);transition: width 0.5s ease-in-out}
+'''
+font = [gr.themes.GoogleFont("Source Sans Pro"), "Arial", "sans-serif"]
+with gr.Blocks(theme=gr.themes.Soft(font=font), css=css, delete_cache=(60, 60)) as app:
+    title = gr.HTML(
+        """<h1>SDXL LoRA DLC</h1>""",
+        elem_id="title",
+    )
+    selected_index = gr.State(None)
+    with gr.Row():
+        with gr.Column(scale=3):
+            prompt = gr.Textbox(label="Prompt", lines=1, placeholder="Type a prompt after selecting a LoRA")
+        with gr.Column(scale=1, elem_id="gen_column"):
+            generate_button = gr.Button("Generate", variant="primary", elem_id="gen_btn")
+    with gr.Row():
+        with gr.Column():
+            selected_info = gr.Markdown("")
+            gallery = gr.Gallery(
+                [(item["image"], item["title"]) for item in loras],
+                label="LoRA Gallery",
+                allow_preview=False,
+                columns=3,
+                elem_id="gallery",
+                show_share_button=False
+            )
+            with gr.Group():
+                custom_lora = gr.Textbox(label="Custom LoRA", info="LoRA Hugging Face path", placeholder="username/sdxl-lora-model")
+                gr.Markdown("[Check the list of SDXL LoRAs](https://huggingface.co/models?other=base_model:stabilityai/stable-diffusion-xl-base-1.0)", elem_id="lora_list")
+            custom_lora_info = gr.HTML(visible=False)
+            custom_lora_button = gr.Button("Remove custom LoRA", visible=False)
+        with gr.Column():
+            progress_bar = gr.Markdown(elem_id="progress", visible=False)
+            result = gr.Image(label="Generated Image")
 
-        if is_tts and voice:
-            output_file = asyncio.run(text_to_speech(final_response, voice))
-            yield gr.Audio(output_file, autoplay=True)
+    with gr.Row():
+        with gr.Accordion("Advanced Settings", open=False):
+            with gr.Row():
+                input_image = gr.Image(label="Input image", type="filepath")
+                image_strength = gr.Slider(label="Denoise Strength", info="Lower means more image influence", minimum=0.1, maximum=1.0, step=0.01, value=0.75)
+            with gr.Column():
+                with gr.Row():
+                    cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=7.5)
+                    steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=30)
+                
+                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(True, label="Randomize seed")
+                    seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0, randomize=True)
+                    lora_scale = gr.Slider(label="LoRA Scale", minimum=0, maximum=3, step=0.01, value=1.0)
 
-# -----------------------
-# Gradio Chat Interface
-# -----------------------
-demo = gr.ChatInterface(
-    fn=generate,
-    additional_inputs=[
-        gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS),
-        gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6),
-        gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
-        gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50),
-        gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2),
-    ],
-    examples=[
-        ["@realism Chocolate dripping from a donut, hyper-realistic"],
-        ["@pixar Young man with wavy hair in an armchair, Pixar style"],
-        ["@realism Futuristic cityscape with neon lights"],
-        ["Write the python array rotation program"],
-        ["@photoshoot Dramatic portrait lighting"],
-        [{"text": "Summarize the letter", "files": ["examples/1.png"]}],
-        ["@tts1 Who was Nikola Tesla and how did he die?"],
-        ["@clothing Fashionable streetwear in the city"],
-        ["@interior Modern minimalist living room"],
-        ["@fashion Runway model in haute couture"],
-        ["@minimalistic Elegant serene landscape"],
-        ["@modern Abstract geometric art"],
-        ["@animaliea Cute animal portrait, vibrant colors"],
-        ["@wallpaper Scenic mountain desktop wallpaper"],
-        ["@cars Sleek sports car on city streets"],
-        ["@pencilart Detailed historic building sketch"],
-        ["@artminimalistic Subtle minimalist artwork"],
-        ["@tts2 What causes rainbows?"],
-    ],
-    cache_examples=False,
-    type="messages",
-    description="# **Gen Vision Sdxl** `tts: @tts1 @tts2` \n `image-tags: @realism, @pixar, @photoshoot, @clothing, @interior, @fashion, @minimalistic, @modern, @animaliea, @wallpaper, @cars, @pencilart, @artminimalistic` \n \n `default: chat, image inference`",
-    fill_height=True,
-    textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="multiple", placeholder="text, image-infer, image-generation, tts"),
-    stop_btn="Stop Generation",
-    multimodal=True,
-)
+    gallery.select(
+        update_selection,
+        inputs=[width, height],
+        outputs=[prompt, selected_info, selected_index, width, height]
+    )
+    custom_lora.input(
+        add_custom_lora,
+        inputs=[custom_lora],
+        outputs=[custom_lora_info, custom_lora_button, gallery, selected_info, selected_index, prompt]
+    )
+    custom_lora_button.click(
+        remove_custom_lora,
+        outputs=[custom_lora_info, custom_lora_button, gallery, selected_info, selected_index, custom_lora]
+    )
+    gr.on(
+        triggers=[generate_button.click, prompt.submit],
+        fn=run_lora,
+        inputs=[prompt, input_image, image_strength, cfg_scale, steps, selected_index, randomize_seed, seed, width, height, lora_scale],
+        outputs=[result, seed, progress_bar]
+    )
 
-if __name__ == "__main__":
-    demo.queue(max_size=20).launch(share=True)
+app.queue()
+app.launch()