app.py

import gradio as gr
import numpy as np
import random
import torch
import spaces
import os
import json

from PIL import Image
from diffusers import QwenImageEditPipeline, FlowMatchEulerDiscreteScheduler
from huggingface_hub import InferenceClient
import math

# --- Prompt Enhancement using Hugging Face InferenceClient ---
def polish_prompt_hf(original_prompt, system_prompt):
    """
    Rewrites the prompt using a Hugging Face InferenceClient.
    """
    # Ensure HF_TOKEN is set
    api_key = os.environ.get("HF_TOKEN")
    if not api_key:
        print("Warning: HF_TOKEN not set. Falling back to original prompt.")
        return original_prompt

    try:
        # Initialize the client
        client = InferenceClient(
            provider="cerebras",
            api_key=api_key,
        )

        # Format the messages for the chat completions API
        messages = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": original_prompt}
        ]

        # Call the API
        completion = client.chat.completions.create(
            model="Qwen/Qwen3-235B-A22B-Instruct-2507",
            messages=messages,
        )
        
        # Parse the response
        result = completion.choices[0].message.content
        
        # Try to extract JSON if present
        if '{"Rewritten"' in result:
            try:
                # Clean up the response
                result = result.replace('```json', '').replace('```', '')
                result_json = json.loads(result)
                polished_prompt = result_json.get('Rewritten', result)
            except:
                polished_prompt = result
        else:
            polished_prompt = result
            
        polished_prompt = polished_prompt.strip().replace("\n", " ")
        return polished_prompt
        
    except Exception as e:
        print(f"Error during API call to Hugging Face: {e}")
        # Fallback to original prompt if enhancement fails
        return original_prompt


def polish_prompt(prompt, img):
    """
    Main function to polish prompts for image editing using HF inference.
    """
    SYSTEM_PROMPT = '''
# Edit Instruction Rewriter
You are a professional edit instruction rewriter. Your task is to generate a precise, concise, and visually achievable professional-level edit instruction based on the user-provided instruction and the image to be edited.  

Please strictly follow the rewriting rules below:

## 1. General Principles
- Keep the rewritten prompt **concise**. Avoid overly long sentences and reduce unnecessary descriptive language.  
- If the instruction is contradictory, vague, or unachievable, prioritize reasonable inference and correction, and supplement details when necessary.  
- Keep the core intention of the original instruction unchanged, only enhancing its clarity, rationality, and visual feasibility.  
- All added objects or modifications must align with the logic and style of the edited input image's overall scene.  

## 2. Task Type Handling Rules
### 1. Add, Delete, Replace Tasks
- If the instruction is clear (already includes task type, target entity, position, quantity, attributes), preserve the original intent and only refine the grammar.  
- If the description is vague, supplement with minimal but sufficient details (category, color, size, orientation, position, etc.). For example:  
    > Original: "Add an animal"  
    > Rewritten: "Add a light-gray cat in the bottom-right corner, sitting and facing the camera"  
- Remove meaningless instructions: e.g., "Add 0 objects" should be ignored or flagged as invalid.  
- For replacement tasks, specify "Replace Y with X" and briefly describe the key visual features of X.  

### 2. Text Editing Tasks
- All text content must be enclosed in English double quotes " ". Do not translate or alter the original language of the text, and do not change the capitalization.  
- **For text replacement tasks, always use the fixed template:**
    - Replace "xx" to "yy".  
    - Replace the xx bounding box to "yy".  
- If the user does not specify text content, infer and add concise text based on the instruction and the input image's context. For example:  
    > Original: "Add a line of text" (poster)  
    > Rewritten: "Add text "LIMITED EDITION" at the top center with slight shadow"  
- Specify text position, color, and layout in a concise way.  

### 3. Human Editing Tasks
- Maintain the person's core visual consistency (ethnicity, gender, age, hairstyle, expression, outfit, etc.).  
- If modifying appearance (e.g., clothes, hairstyle), ensure the new element is consistent with the original style.  
- **For expression changes, they must be natural and subtle, never exaggerated.**  
- If deletion is not specifically emphasized, the most important subject in the original image (e.g., a person, an animal) should be preserved.
    - For background change tasks, emphasize maintaining subject consistency at first.  
- Example:  
    > Original: "Change the person's hat"  
    > Rewritten: "Replace the man's hat with a dark brown beret; keep smile, short hair, and gray jacket unchanged"  

### 4. Style Transformation or Enhancement Tasks
- If a style is specified, describe it concisely with key visual traits. For example:  
    > Original: "Disco style"  
    > Rewritten: "1970s disco: flashing lights, disco ball, mirrored walls, colorful tones"  
- If the instruction says "use reference style" or "keep current style," analyze the input image, extract main features (color, composition, texture, lighting, art style), and integrate them concisely.  
- **For coloring tasks, including restoring old photos, always use the fixed template:** "Restore old photograph, remove scratches, reduce noise, enhance details, high resolution, realistic, natural skin tones, clear facial features, no distortion, vintage photo restoration"  
- If there are other changes, place the style description at the end.

## 3. Rationality and Logic Checks
- Resolve contradictory instructions: e.g., "Remove all trees but keep all trees" should be logically corrected.  
- Add missing key information: if position is unspecified, choose a reasonable area based on composition (near subject, empty space, center/edges).  

# Output Format
Return only the rewritten instruction text directly, without JSON formatting or any other wrapper.
'''
    
    # Note: We're not actually using the image in the HF version, 
    # but keeping the interface consistent
    full_prompt = f"{SYSTEM_PROMPT}\n\nUser Input: {prompt}\n\nRewritten Prompt:"
    
    return polish_prompt_hf(full_prompt, SYSTEM_PROMPT)

# --- Model Loading ---
dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

# Scheduler configuration for Lightning
scheduler_config = {
    "base_image_seq_len": 256,
    "base_shift": math.log(3),
    "invert_sigmas": False,
    "max_image_seq_len": 8192,
    "max_shift": math.log(3),
    "num_train_timesteps": 1000,
    "shift": 1.0,
    "shift_terminal": None,
    "stochastic_sampling": False,
    "time_shift_type": "exponential",
    "use_beta_sigmas": False,
    "use_dynamic_shifting": True,
    "use_exponential_sigmas": False,
    "use_karras_sigmas": False,
}

# Initialize scheduler with Lightning config
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)

# Load the edit pipeline with Lightning scheduler
pipe = QwenImageEditPipeline.from_pretrained(
    "Qwen/Qwen-Image-Edit", 
    scheduler=scheduler,
    torch_dtype=dtype
).to(device)

# Load Lightning LoRA weights for acceleration
try:
    pipe.load_lora_weights(
        "lightx2v/Qwen-Image-Lightning", 
        weight_name="Qwen-Image-Lightning-8steps-V1.1.safetensors"
    )
    pipe.fuse_lora()
    print("Successfully loaded Lightning LoRA weights")
except Exception as e:
    print(f"Warning: Could not load Lightning LoRA weights: {e}")
    print("Continuing with base model...")

# --- UI Constants and Helpers ---
MAX_SEED = np.iinfo(np.int32).max

# Illumination options mapping
ILLUMINATION_OPTIONS = {
# Natural Daylight
    "natural lighting": "Neutral white color temperature with balanced exposure and soft shadows",
    "sunshine from window": "Bright directional sunlight with hard shadows and visible light rays",
    "golden time": "Warm golden hour lighting with enhanced warm colors and soft shadows",
    "sunrise in the mountains": "Warm backlighting with atmospheric haze and lens flare",
    "afternoon light filtering through trees": "Dappled sunlight patterns with green color cast from foliage",
    "early morning rays, forest clearing": "God rays through trees with warm color temperature",
    "golden sunlight streaming through trees": "Golden god rays with atmospheric particles in light beams",
    
    # Sunset & Evening
    "sunset over sea": "Warm sunset light with soft diffused lighting and gentle gradients",
    "golden hour in a meadow": "Golden backlighting with lens flare and rim lighting",
    "golden hour on a city skyline": "Golden lighting on buildings with silhouette effects",
    "evening glow in the desert": "Warm directional lighting with long shadows",
    "dusky evening on a beach": "Cool backlighting with horizon silhouettes",
    "mellow evening glow on a lake": "Warm lighting with water reflections",
    "warm sunset in a rural village": "Golden hour lighting with peaceful warm tones",
    
    # Night & Moonlight
    "moonlight through curtains": "Cool blue lighting with curtain shadow patterns",
    "moonlight in a dark alley": "Cool blue lighting with deep urban shadows",
    "midnight in the forest": "Very low brightness with minimal ambient lighting",
    "midnight sky with bright starlight": "Cool blue lighting with star point sources",
    "fireflies lighting up a summer night": "Small glowing points with warm ambient lighting",
    
    # Indoor & Cozy
    "warm atmosphere, at home, bedroom": "Very warm lighting with soft diffused glow",
    "home atmosphere, cozy bedroom illumination": "Warm table lamp lighting with pools of light",
    "cozy candlelight": "Warm orange flickering light with dramatic shadows",
    "candle-lit room, rustic vibe": "Multiple warm candlelight sources with atmospheric shadows",
    "night, cozy warm light from fireplace": "Warm orange-red firelight with flickering effects",
    "campfire light": "Warm orange flickering light from below with dancing shadows",
    
    # Urban & Neon
    "neon night, city": "Vibrant blue, magenta, and green neon lights with reflections",
    "blue neon light, urban street": "Blue neon lighting with urban glow effects",
    "neon, Wong Kar-wai, warm": "Warm amber and red neon with moody selective lighting",
    "red and blue police lights in rain": "Alternating red and blue strobing with wet reflections",
    "red glow, emergency lights": "Red emergency lighting with harsh shadows and high contrast",

     # Sci-Fi & Fantasy
    "sci-fi RGB glowing, cyberpunk": "Electric blue, pink, and green RGB lighting with glowing effects",
    "rainbow reflections, neon": "Chromatic rainbow patterns with prismatic reflections",
    "magic lit": "Colored rim lighting in purple and blue with soft ethereal glow",
    "mystical glow, enchanted forest": "Supernatural green and blue glowing with floating particles",
    "ethereal glow, magical forest": "Supernatural lighting with blue-green rim lighting",
    "underwater glow, deep sea": "Blue-green lighting with caustic patterns and particles",
    "underwater luminescence": "Blue-green bioluminescent glow with caustic light patterns",
    "aurora borealis glow, arctic landscape": "Green and purple dancing sky lighting",
    "crystal reflections in a cave": "Sparkle effects with prismatic light dispersion",
    
    # Weather & Atmosphere
    "foggy forest at dawn": "Volumetric fog with cool god rays through trees",
    "foggy morning, muted light": "Soft fog effects with reduced contrast throughout",
    "soft, diffused foggy glow": "Heavy fog with soft lighting and no harsh shadows",
    "stormy sky lighting": "Dramatic lighting with high contrast and rim lighting",
    "lightning flash in storm": "Brief intense white light with stark shadows",
    "rain-soaked reflections in city lights": "Wet surface reflections with streaking light effects",
    "gentle snowfall at dusk": "Cool blue lighting with snowflake particle effects",
    "hazy light of a winter morning": "Neutral lighting with atmospheric haze",
    "mysterious twilight, heavy mist": "Heavy fog with cool lighting and atmospheric depth",
    
    # Seasonal & Nature
    "vibrant autumn lighting in a forest": "Enhanced warm autumn colors with dappled sunlight",
    "purple and pink hues at twilight": "Warm lighting with soft purple and pink color grading",
    "desert sunset with mirage-like glow": "Warm orange lighting with heat distortion effects",
    "sunrise through foggy mountains": "Warm lighting through mist with atmospheric perspective",
    
    # Professional & Studio
    "soft studio lighting": "Multiple diffused sources with even illumination and minimal shadows",
    "harsh, industrial lighting": "Bright fluorescent lighting with hard shadows",
    "fluorescent office lighting": "Cool white overhead lighting with slight green tint",
    "harsh spotlight in dark room": "Single intense directional light with dramatic shadows",
    
    # Special Effects & Drama
    "light and shadow": "Maximum contrast with sharp shadow boundaries",
    "shadow from window": "Window frame shadow patterns with geometric shapes",
    "apocalyptic, smoky atmosphere": "Orange-red fire tint with smoke effects",
    "evil, gothic, in a cave": "Low brightness with cool lighting and deep shadows",
    "flickering light in a haunted house": "Unstable flickering with cool and warm mixed lighting",
    "golden beams piercing through storm clouds": "Dramatic god rays with high contrast",
    "dim candlelight in a gothic castle": "Warm orange candlelight with stone texture enhancement",

    # Festival & Celebration
    "colorful lantern light at festival": "Multiple colored lantern sources with bokeh effects",
    "golden glow at a fairground": "Warm carnival lighting with colorful bulb effects",
    "soft glow through stained glass": "Colored light filtering with rainbow surface patterns",
    "glowing embers from a forge": "Orange-red glowing particles with intense heat effects"

    }

# Lighting direction options
DIRECTION_OPTIONS = {
    "auto": "",  
    "left side": "Position the light source from the left side of the frame, creating shadows falling to the right.",
    "right side": "Position the light source from the right side of the frame, creating shadows falling to the left.",
    "top": "Position the light source from directly above, creating downward shadows.",
    "top left": "Position the light source from the top left corner, creating diagonal shadows falling down and to the right.",
    "top right": "Position the light source from the top right corner, creating diagonal shadows falling down and to the left.",
    "bottom": "Position the light source from below, creating upward shadows and dramatic under-lighting.",
    "front": "Position the light source from the front, minimizing shadows and creating even illumination.",
    "back": "Position the light source from behind the subject, creating silhouette effects and rim lighting."
}


# --- Main Inference Function ---
@spaces.GPU(duration=60)
def infer(
    image,
    prompt,
    illumination_dropdown, direction_dropdown, 
    seed=42,
    randomize_seed=False,
    true_guidance_scale=1.0,
    num_inference_steps=8,  # Default to 8 steps for fast inference
    rewrite_prompt=True,
    progress=gr.Progress(track_tqdm=True),
):
    """
    Generates an edited image using the Qwen-Image-Edit pipeline with Lightning acceleration.
    """
    # Hardcode the negative prompt as in the original
    negative_prompt = " "
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    # Set up the generator for reproducibility
    generator = torch.Generator(device=device).manual_seed(seed)
    
    print(f"Original prompt: '{prompt}'")
    print(f"Negative Prompt: '{negative_prompt}'")
    print(f"Seed: {seed}, Steps: {num_inference_steps}, Guidance: {true_guidance_scale}")
    
    #If the dropdown isn't custom, and the user didn't specify a prompt, fill the prompt with the correct one from the illumination options
    if illumination_dropdown != "custom" and (prompt == "" or prompt == ILLUMINATION_OPTIONS[illumination_dropdown]):
        prompt = f"change the lighting. add {ILLUMINATION_OPTIONS[illumination_dropdown]}"


    # If direction isn't auto, add the direction suffix
    if direction_dropdown != "auto":
        prompt_with_template = prompt+ f" coming from the {direction_dropdown}"
    else:
        prompt_with_template= prompt
    
    if rewrite_prompt:
        final_prompt = polish_prompt(prompt_with_template, input_image)
    else:
        final_prompt = prompt_with_template

    print(f"Calling pipeline with prompt: '{final_prompt}'")
    

    # Generate the edited image - always generate just 1 image
    try:
        images = pipe(
            image,
            prompt=final_prompt,
            negative_prompt=negative_prompt,
            num_inference_steps=num_inference_steps,
            generator=generator,
            true_cfg_scale=true_guidance_scale,
            num_images_per_prompt=1  # Always generate only 1 image
        ).images
        
        # Return the first (and only) image
        return images[0], seed
        
    except Exception as e:
        print(f"Error during inference: {e}")
        raise e

def update_prompt_from_dropdown(illumination_option):
    """Update the prompt textbox based on dropdown selection"""
    if illumination_option == "custom":
        return ""  # Clear the prompt for custom input
    else:
        return ILLUMINATION_OPTIONS[illumination_option]

# --- Examples and UI Layout ---
examples = [
    # You can add example pairs of [image_path, prompt] here
    # ["path/to/image1.jpg", "Replace the background with a beach scene"],
    # ["path/to/image2.jpg", "Add a red hat to the person"],
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 1024px;
}
#logo-title {
    text-align: center;
}
#logo-title img {
    width: 400px;
}
#edit_text{margin-top: -62px !important}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.HTML("""
        <div id="logo-title">
            <img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/qwen_image_edit_logo.png" alt="Qwen-Image Edit Logo" width="400" style="display: block; margin: 0 auto;">
            <h2 style="font-style: italic;color: #5b47d1;margin-top: -27px !important;margin-left: 133px;">Relight</h2>
        </div>
        """)
        gr.Markdown("""
        [Learn more](https://github.com/QwenLM/Qwen-Image) about the Qwen-Image series. 
        This demo uses the [Qwen-Image-Lightning](https://huggingface.co/lightx2v/Qwen-Image-Lightning) LoRA for accelerated 8-step inference.
        Try on [Qwen Chat](https://chat.qwen.ai/), or [download model](https://huggingface.co/Qwen/Qwen-Image-Edit) to run locally with ComfyUI or diffusers.
        """)
        
        with gr.Row():
            with gr.Column():
                input_image = gr.Image(
                    label="Input Image", 
                    show_label=True, 
                    type="pil"
                )
                with gr.Row():
                    
                    illumination_dropdown = gr.Dropdown(
                        choices=["custom"] + list(ILLUMINATION_OPTIONS.keys()),
                        value="sunshine from window",
                        label="Choose Lighting Style",
                        scale=2
                    )
                    
                    direction_dropdown = gr.Dropdown(
                        choices=list(DIRECTION_OPTIONS.keys()),
                        value="auto",
                        label="Light Direction",
                        scale=1
                    )
            # Changed from Gallery to Image
            result = gr.Image(
                label="Result", 
                show_label=True, 
                type="pil"
            )
            
        with gr.Row():
            prompt = gr.Text(
                label="Edit Instruction",
                show_label=False,
                placeholder="Describe the edit instruction (e.g., 'Replace the background with a sunset', 'Add a red hat', 'Remove the person')",
                container=False,
            )
            run_button = gr.Button("Edit!", variant="primary")

        with gr.Accordion("Advanced Settings", open=False):
            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():
                true_guidance_scale = gr.Slider(
                    label="True guidance scale",
                    minimum=1.0,
                    maximum=10.0,
                    step=0.1,
                    value=1.0
                )

                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=4,
                    maximum=28,
                    step=1,
                    value=8
                )
                
            # Removed num_images_per_prompt slider entirely
            rewrite_prompt = gr.Checkbox(
                label="Enhance prompt (using HF Inference)", 
                value=True
            )

        
        gr.Examples(
                examples=[
                    ["./assets/pexels-creationhill-1681010.jpg", "Add multiple colored light sources from lanterns. Create warm festival lighting. Set varied color temperatures. Add bokeh effects.", "colorful lantern light at festival", "auto"],
                    ["./assets/pexels-creationhill-1681010.jpg",  "add futuristic RGB lighting with electric blues, hot pinks, and neon greens creating a high-tech atmosphere with dramatic color separation and glowing effects", "sci-fi RGB glowing, cyberpunk", "left side"],
                    ["./assets/pexels-moose-photos-170195-1587009.jpg",  "Set blue-green color temperature. Add volumetric lighting effects. Reduce red channel significantly. Create particle effects in light beams. Add caustic light patterns.", "underwater glow, deep sea", "top"],
                    ["./assets/pexels-moose-photos-170195-1587009.jpg", "Replace lighting with red sources. Add flashing strobing effects. Increase contrast. Create harsh shadows. Set monochromatic red color scheme.", "red glow, emergency lights", "right side"],
                    ["./assets/pexels-simon-robben-55958-614810.jpg",  "Add directional sunlight from window source. Increase brightness on lit areas. Create hard shadows with sharp edges. Set warm white color temperature. Add visible light rays and dust particles in beams.", "sunshine from window", "top right"],
                    ["./assets/pexels-simon-robben-55958-614810.jpg", "add vibrant neon lights in electric blues, magentas, and greens casting colorful reflections on surfaces, creating a cyberpunk urban atmosphere with dramatic color contrasts", "neon night, city", "top left"],
                    ["./assets/pexels-freestockpro-1227513.jpg",  "warm lighting with soft purple and pink color grading", "purple and pink hues at twilight", "auto"],
                    ["./assets/pexels-pixabay-158827.jpg", "Soft fog effects with reduced contrast throughout", "foggy morning, muted light", "auto"],
                    ["./assets/pexels-pixabay-355465.jpg", "daylight, bright sunshine", "custom", "auto" ]           
                ],
                inputs=[input_image, prompt, illumination_dropdown, direction_dropdown],
                outputs=[result, seed],
                fn=infer,
                cache_examples="lazy"
            )
    # update prompt when dropdown changes
    illumination_dropdown.change(
        fn=update_prompt_from_dropdown,
        inputs=[illumination_dropdown],
        outputs=[prompt]
    )
    
    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn=infer,
        inputs=[
            input_image,
            prompt,
            illumination_dropdown, direction_dropdown, 
            seed,
            randomize_seed,
            true_guidance_scale,
            num_inference_steps,
            rewrite_prompt,
            # Removed num_images_per_prompt from inputs
        ],
        outputs=[result, seed],
    )

if __name__ == "__main__":
    demo.launch()