Update app.py

app.py

@@ -1,372 +1,368 @@
+import os
+import random
+import uuid
+import json
+import time
+import asyncio
+import re
+from threading import Thread
+
 import gradio as gr
+import spaces
 import torch
-from diffusers import FluxFillPipeline
-from diffusers.utils import load_image
-from PIL import Image, ImageDraw
 import numpy as np
-import spaces
-from huggingface_hub import hf_hub_download
-
-pipe = FluxFillPipeline.from_pretrained(
-    "black-forest-labs/FLUX.1-Fill-dev",
-    torch_dtype=torch.bfloat16
-).to("cuda")
-
-def can_expand(source_width, source_height, target_width, target_height, alignment):
-    if alignment in ("Left", "Right") and source_width >= target_width:
-        return False
-    if alignment in ("Top", "Bottom") and source_height >= target_height:
-        return False
-    return True
-
-def prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom):
-    target_size = (width, height)
-
-    scale_factor = min(target_size[0] / image.width, target_size[1] / image.height)
-    new_width = int(image.width * scale_factor)
-    new_height = int(image.height * scale_factor)
-    
-    source = image.resize((new_width, new_height), Image.LANCZOS)
-
-    if resize_option == "Full":
-        resize_percentage = 100
-    elif resize_option == "75%":
-        resize_percentage = 75
-    elif resize_option == "50%":
-        resize_percentage = 50
-    elif resize_option == "33%":
-        resize_percentage = 33
-    elif resize_option == "25%":
-        resize_percentage = 25
-    else:  # Custom
-        resize_percentage = custom_resize_percentage
-
-    # Calculate new dimensions based on percentage
-    resize_factor = resize_percentage / 100
-    new_width = int(source.width * resize_factor)
-    new_height = int(source.height * resize_factor)
-
-    # Ensure minimum size of 64 pixels
-    new_width = max(new_width, 64)
-    new_height = max(new_height, 64)
-
-    # Resize the image
-    source = source.resize((new_width, new_height), Image.LANCZOS)
-
-    # Calculate the overlap in pixels based on the percentage
-    overlap_x = int(new_width * (overlap_percentage / 100))
-    overlap_y = int(new_height * (overlap_percentage / 100))
-
-    # Ensure minimum overlap of 1 pixel
-    overlap_x = max(overlap_x, 1)
-    overlap_y = max(overlap_y, 1)
-
-    # Calculate margins based on alignment
-    if alignment == "Middle":
-        margin_x = (target_size[0] - new_width) // 2
-        margin_y = (target_size[1] - new_height) // 2
-    elif alignment == "Left":
-        margin_x = 0
-        margin_y = (target_size[1] - new_height) // 2
-    elif alignment == "Right":
-        margin_x = target_size[0] - new_width
-        margin_y = (target_size[1] - new_height) // 2
-    elif alignment == "Top":
-        margin_x = (target_size[0] - new_width) // 2
-        margin_y = 0
-    elif alignment == "Bottom":
-        margin_x = (target_size[0] - new_width) // 2
-        margin_y = target_size[1] - new_height
-
-    # Adjust margins to eliminate gaps
-    margin_x = max(0, min(margin_x, target_size[0] - new_width))
-    margin_y = max(0, min(margin_y, target_size[1] - new_height))
-
-    # Create a new background image and paste the resized source image
-    background = Image.new('RGB', target_size, (255, 255, 255))
-    background.paste(source, (margin_x, margin_y))
-
-    # Create the mask
-    mask = Image.new('L', target_size, 255)
-    mask_draw = ImageDraw.Draw(mask)
+from PIL import Image
+import edge_tts
+
+from transformers import (
+    AutoModelForCausalLM,
+    AutoTokenizer,
+    TextIteratorStreamer,
+    Qwen2VLForConditionalGeneration,
+    AutoProcessor,
+)
+from transformers.image_utils import load_image
+from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
+
+DESCRIPTION = """
+# SDXL LoRA DLC 🎃
+"""
 
-    # Calculate overlap areas
-    white_gaps_patch = 2
+css = '''
+h1 {
+  text-align: center;
+  display: block;
+}
 
-    left_overlap = margin_x + overlap_x if overlap_left else margin_x + white_gaps_patch
-    right_overlap = margin_x + new_width - overlap_x if overlap_right else margin_x + new_width - white_gaps_patch
-    top_overlap = margin_y + overlap_y if overlap_top else margin_y + white_gaps_patch
-    bottom_overlap = margin_y + new_height - overlap_y if overlap_bottom else margin_y + new_height - white_gaps_patch
+#duplicate-button {
+  margin: auto;
+  color: #fff;
+  background: #1565c0;
+  border-radius: 100vh;
+}
+'''
+
+MAX_MAX_NEW_TOKENS = 2048
+DEFAULT_MAX_NEW_TOKENS = 1024
+MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+
+# -----------------------
+# 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>
+    '''
+
+# -----------------------
+# Text Generation Setup
+# -----------------------
+model_id = "prithivMLmods/FastThink-0.5B-Tiny"
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    device_map="auto",
+    torch_dtype=torch.bfloat16,
+)
+model.eval()
+
+TTS_VOICES = [
+    "en-US-JennyNeural",  # @tts1
+    "en-US-GuyNeural",    # @tts2
+]
+
+# -----------------------
+# 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()
+
+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 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
+
+# -----------------------
+# Stable Diffusion Image Generation Setup
+# -----------------------
+
+MAX_SEED = np.iinfo(np.int32).max
+USE_TORCH_COMPILE = False
+ENABLE_CPU_OFFLOAD = False
+
+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)
     
-    if alignment == "Left":
-        left_overlap = margin_x + overlap_x if overlap_left else margin_x
-    elif alignment == "Right":
-        right_overlap = margin_x + new_width - overlap_x if overlap_right else margin_x + new_width
-    elif alignment == "Top":
-        top_overlap = margin_y + overlap_y if overlap_top else margin_y
-    elif alignment == "Bottom":
-        bottom_overlap = margin_y + new_height - overlap_y if overlap_bottom else margin_y + new_height
-
-    # Draw the mask
-    mask_draw.rectangle([
-        (left_overlap, top_overlap),
-        (right_overlap, bottom_overlap)
-    ], fill=0)
-
-    return background, mask
+    # 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"),
+    }
+
+    # 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 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
+
+def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+    return seed
+
+@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
 
+# -----------------------
+# Main Chat/Generation Function
+# -----------------------
 @spaces.GPU
-def inpaint(image, width, height, overlap_percentage, num_inference_steps, resize_option, custom_resize_percentage, prompt_input, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom, progress=gr.Progress(track_tqdm=True)):
-    
-    background, mask = prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom)
+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", [])
     
-    if not can_expand(background.width, background.height, width, height, alignment):
-        alignment = "Middle"
-
-    cnet_image = background.copy()
-    cnet_image.paste(0, (0, 0), mask)
-
-    final_prompt = prompt_input
-
-    #generator = torch.Generator(device="cuda").manual_seed(42)
-
-    result = pipe(
-        prompt=final_prompt,
-        height=height,
-        width=width,
-        image=cnet_image,
-        mask_image=mask,
-        num_inference_steps=num_inference_steps,
-        guidance_scale=30,
-    ).images[0]
-   
-    result = result.convert("RGBA")
-    cnet_image.paste(result, (0, 0), mask)
-
-    return cnet_image, background
-
-def preview_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom):
-    background, mask = prepare_image_and_mask(image, width, height, overlap_percentage, resize_option, custom_resize_percentage, alignment, overlap_left, overlap_right, overlap_top, overlap_bottom)
+    # 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 
     
-    preview = background.copy().convert('RGBA')
-    red_overlay = Image.new('RGBA', background.size, (255, 0, 0, 64))
-    red_mask = Image.new('RGBA', background.size, (0, 0, 0, 0))
-    red_mask.paste(red_overlay, (0, 0), mask)
-    preview = Image.alpha_composite(preview, red_mask)
+    # 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)
     
-    return preview
-
-def clear_result():
-    return gr.update(value=None)
-
-def preload_presets(target_ratio, ui_width, ui_height):
-    if target_ratio == "9:16":
-        return 720, 1280, gr.update()
-    elif target_ratio == "16:9":
-        return 1280, 720, gr.update()
-    elif target_ratio == "1:1":
-        return 1024, 1024, gr.update()
-    elif target_ratio == "Custom":
-        return ui_width, ui_height, gr.update(open=True)
-
-def select_the_right_preset(user_width, user_height):
-    if user_width == 720 and user_height == 1280:
-        return "9:16"
-    elif user_width == 1280 and user_height == 720:
-        return "16:9"
-    elif user_width == 1024 and user_height == 1024:
-        return "1:1"
+    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:
-        return "Custom"
-
-def toggle_custom_resize_slider(resize_option):
-    return gr.update(visible=(resize_option == "Custom"))
-
-def update_history(new_image, history):
-    if history is None:
-        history = []
-    history.insert(0, new_image)
-    return history
-
-css = """
-.gradio-container {
-    max-width: 1250px !important;
-}
-"""
-
-title = """<h1 align="center">Flux Outpaint Dev 🤩</h1>"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column():
-        gr.HTML(title)
-
-        with gr.Row():
-            with gr.Column():
-                input_image = gr.Image(
-                    type="pil",
-                    label="Input Image"
-                )
-
-                with gr.Row():
-                    with gr.Column(scale=2):
-                        prompt_input = gr.Textbox(label="Prompt (Optional)")
-                    with gr.Column(scale=1):
-                        run_button = gr.Button("Generate")
-
-                with gr.Row():
-                    target_ratio = gr.Radio(
-                        label="Image Ratio",
-                        choices=["9:16", "16:9", "1:1", "Custom"],
-                        value="9:16",
-                        scale=3
-                    )
-                    alignment_dropdown = gr.Dropdown(
-                        choices=["Middle", "Left", "Right", "Top", "Bottom"],
-                        value="Middle",
-                        label="Alignment",
-                    )
-                resize_option = gr.Radio(
-                    label="Resize input image",
-                    choices=["Full", "75%", "50%", "33%", "25%", "Custom"],
-                    value="75%"
-                )
-                custom_resize_percentage = gr.Slider(
-                    label="Custom resize (%)",
-                    minimum=1,
-                    maximum=100,
-                    step=1,
-                    value=50,
-                    visible=False
-                )
-                with gr.Accordion(label="Advanced settings", open=False) as settings_panel:
-                    with gr.Column():
-                        with gr.Row():
-                            width_slider = gr.Slider(
-                                label="Target Width",
-                                minimum=720,
-                                maximum=1536,
-                                step=8,
-                                value=720,
-                            )
-                            height_slider = gr.Slider(
-                                label="Target Height",
-                                minimum=720,
-                                maximum=1536,
-                                step=8,
-                                value=1280,
-                            )
-                        
-                        num_inference_steps = gr.Slider(label="Steps", minimum=2, maximum=50, step=1, value=28)
-                        with gr.Group():
-                            overlap_percentage = gr.Slider(
-                                label="Mask overlap (%)",
-                                minimum=1,
-                                maximum=50,
-                                value=10,
-                                step=1
-                            )
-                            with gr.Row():
-                                overlap_top = gr.Checkbox(label="Overlap Top", value=True)
-                                overlap_right = gr.Checkbox(label="Overlap Right", value=True)
-                            with gr.Row():
-                                overlap_left = gr.Checkbox(label="Overlap Left", value=True)
-                                overlap_bottom = gr.Checkbox(label="Overlap Bottom", value=True)
-                        
-                        with gr.Column():
-                            preview_button = gr.Button("Preview alignment and mask")
-                            
-            with gr.Column():
-                result = gr.Image(
-                    interactive=False,
-                    label="Generated Image",
-                )
-                use_as_input_button = gr.Button("Use as Input Image", visible=False)
-                with gr.Accordion("History and Mask", open=False):
-                    history_gallery = gr.Gallery(label="History", columns=6, object_fit="contain", interactive=False)
-                    preview_image = gr.Image(label="Mask preview")
-
-    def use_output_as_input(output_image):
-        return output_image
-
-    use_as_input_button.click(
-        fn=use_output_as_input,
-        inputs=[result],
-        outputs=[input_image]
-    )
+        voice = None
+        text = text.replace(tts_prefix, "").strip()
+        conversation = clean_chat_history(chat_history)
+        conversation.append({"role": "user", "content": text})
     
-    target_ratio.change(
-        fn=preload_presets,
-        inputs=[target_ratio, width_slider, height_slider],
-        outputs=[width_slider, height_slider, settings_panel],
-        queue=False
-    )
-
-    width_slider.change(
-        fn=select_the_right_preset,
-        inputs=[width_slider, height_slider],
-        outputs=[target_ratio],
-        queue=False
-    )
-
-    height_slider.change(
-        fn=select_the_right_preset,
-        inputs=[width_slider, height_slider],
-        outputs=[target_ratio],
-        queue=False
-    )
-
-    resize_option.change(
-        fn=toggle_custom_resize_slider,
-        inputs=[resize_option],
-        outputs=[custom_resize_percentage],
-        queue=False
-    )
-    
-    run_button.click(
-        fn=clear_result,
-        inputs=None,
-        outputs=result,
-    ).then(
-        fn=inpaint,
-        inputs=[input_image, width_slider, height_slider, overlap_percentage, num_inference_steps,
-                resize_option, custom_resize_percentage, prompt_input, alignment_dropdown,
-                overlap_left, overlap_right, overlap_top, overlap_bottom],
-        outputs=[result, preview_image],
-    ).then(
-        fn=lambda x, history: update_history(x, history),
-        inputs=[result, history_gallery],
-        outputs=history_gallery,
-    ).then(
-        fn=lambda: gr.update(visible=True),
-        inputs=None,
-        outputs=use_as_input_button,
-    )
-
-    prompt_input.submit(
-        fn=clear_result,
-        inputs=None,
-        outputs=result,
-    ).then(
-        fn=inpaint,
-        inputs=[input_image, width_slider, height_slider, overlap_percentage, num_inference_steps, resize_option, custom_resize_percentage, prompt_input, alignment_dropdown,
-                overlap_left, overlap_right, overlap_top, overlap_bottom],
-        outputs=[result, preview_image],
-    ).then(
-        fn=lambda x, history: update_history(x, history),
-        inputs=[result, history_gallery],
-        outputs=history_gallery,
-    ).then(
-        fn=lambda: gr.update(visible=True),
-        inputs=None,
-        outputs=use_as_input_button,
-    )
-
-    preview_button.click(
-        fn=preview_image_and_mask,
-        inputs=[input_image, width_slider, height_slider, overlap_percentage, resize_option, custom_resize_percentage, alignment_dropdown,
-                overlap_left, overlap_right, overlap_top, overlap_bottom],
-        outputs=preview_image,
-        queue=False
-    )
-
-demo.queue(max_size=12).launch(share=False)
+    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()
+
+        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
+    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()
+
+        outputs = []
+        for new_text in streamer:
+            outputs.append(new_text)
+            yield "".join(outputs)
+
+        final_response = "".join(outputs)
+        yield final_response
+
+        if is_tts and voice:
+            output_file = asyncio.run(text_to_speech(final_response, voice))
+            yield gr.Audio(output_file, autoplay=True)
+
+# -----------------------
+# 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 against a yellow background, in the style of brocore, hyper-realistic'],
+        ["@pixar A young man with light brown wavy hair and light brown eyes sitting in an armchair and looking directly at the camera, pixar style, disney pixar, office background, ultra detailed, 1 man"],
+        ["@realism A futuristic cityscape with neon lights"],
+        ["@photoshoot A portrait of a person with dramatic lighting"],
+        [{"text": "summarize the letter", "files": ["examples/1.png"]}],
+        ["Python Program for Array Rotation"],
+        ["@tts1 Who is Nikola Tesla, and why did he die?"],
+        ["@clothing Fashionable streetwear in an urban environment"],
+        ["@interior A modern living room interior with minimalist design"],
+        ["@fashion A runway model in haute couture"],
+        ["@minimalistic A simple and elegant design of a serene landscape"],
+        ["@modern A contemporary art piece with abstract geometric shapes"],
+        ["@animaliea A cute animal portrait with vibrant colors"],
+        ["@wallpaper A scenic mountain range perfect for a desktop wallpaper"],
+        ["@cars A sleek sports car cruising on a city street"],
+        ["@pencilart A detailed pencil sketch of a historic building"],
+        ["@artminimalistic An artistic minimalist composition with subtle tones"],
+        ["@tts2 What causes rainbows to form?"],
+    ],
+    cache_examples=False,
+    type="messages",
+    description=DESCRIPTION,
+    css=css,
+    fill_height=True,
+    textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="multiple", placeholder="default [text, vision] , scroll down examples to explore more art styles"),
+    stop_btn="Stop Generation",
+    multimodal=True,
+)
+
+if __name__ == "__main__":
+    demo.queue(max_size=20).launch(ssr_mode=False, share=True)