app.py

import gradio as gr
import numpy as np
import random
import os
import spaces
import torch
from diffusers import DiffusionPipeline
from transformers import pipeline
from huggingface_hub import login

hf_token = os.getenv("hf_token")
login(token=hf_token)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

_text_gen_pipeline = None
_image_gen_pipeline = None

@spaces.GPU()
def get_image_gen_pipeline():
    global _image_gen_pipeline
    if _image_gen_pipeline is None:
        try:
            device = "cuda" if torch.cuda.is_available() else "cpu"
            dtype = torch.bfloat16
            _image_gen_pipeline = DiffusionPipeline.from_pretrained(
                "black-forest-labs/FLUX.1-schnell",
                torch_dtype=dtype
            ).to(device)
        except Exception as e:
            print(f"Error loading image generation model: {e}")
            return None
    return _image_gen_pipeline

@spaces.GPU()
def get_text_gen_pipeline():
    global _text_gen_pipeline
    if _text_gen_pipeline is None:
         try:
              device = "cuda" if torch.cuda.is_available() else "cpu"
              _text_gen_pipeline = pipeline(
                   "text-generation",
                   model="mistralai/Mistral-7B-Instruct-v0.3",
                   max_new_tokens=2048,
                   device=device,
              )
         except Exception as e:
              print(f"Error loading text generation model: {e}")
              return None
    return _text_gen_pipeline

@spaces.GPU()
def refine_prompt(prompt):
    text_gen = get_text_gen_pipeline()
    if text_gen is None:
        return "Text generation model is unavailable."
    try:
        messages = [
            {"role": "system", "content": "You are a product designer specialized in text-to-image generation. You will get a basic prompt of product request and you need to imagine a new product design to satisfy that need. Generate a visual product description that will then be used by a text-to-image AI (Flux) to suggest a visual. The prompt should be in the form of a strict product description, not a story, maximum 2048 tokens. You need to mention explicitly the visual aesthetics (ex: photo realistic, high quality, etc). Background should be a full white background."},           {"role": "system", "content": "You are a product designer specialized in text-to-image generation. You will get a basic prompt of product request and you need to imagine a new product design to satisfy that need. Genrerate a product description of product front view that will be used by a text-to-image AI (Flux) to generate a visual. The prompt should be in the form of a product description, not a story, maximum 2048 tokens. You need to mention explicitly the visual aesthetics (ex: photo realistic, high quality, etc). Background should be a full white background."},
            {"role": "user", "content": prompt},
        ]
        refined_prompt = text_gen(messages)
        
        # Extract just the assistant's content from the response
        try:
            messages = refined_prompt[0]['generated_text']
            # Find the last message with role 'assistant'
            assistant_messages = [msg for msg in messages if msg['role'] == 'assistant']
            if not assistant_messages:
                return "Error: No assistant response found"
            assistant_content = assistant_messages[-1]['content']
            return assistant_content
        except (KeyError, IndexError):
            return "Error: Unexpected response format from the model"
    except Exception as e:
        return f"Error refining prompt: {str(e)}"

def validate_dimensions(width, height):
    if width * height > MAX_IMAGE_SIZE * MAX_IMAGE_SIZE:
        return False, "Image dimensions too large"
    return True, None

@spaces.GPU()
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, progress=gr.Progress(track_tqdm=True)):
    try:
        progress(0, desc="Starting generation...")
        
        pipe = get_image_gen_pipeline()
        if pipe is None:
            return None, "Image generation model is unavailable."
        
        # Validate that prompt is not empty
        if not prompt or prompt.strip() == "":
            return None, "Please provide a valid prompt."

        # Validate width/height dimensions
        is_valid, error_msg = validate_dimensions(width, height)
        if not is_valid:
            return None, error_msg

        if randomize_seed:
            seed = random.randint(0, MAX_SEED)
        
        progress(0.2, desc="Setting up generator...")
        generator = torch.Generator().manual_seed(seed)
        
        progress(0.4, desc="Generating image...")
        with torch.cuda.amp.autocast():
            image = pipe(
                    prompt=prompt, 
                    width=width,
                    height=height,
                    num_inference_steps=num_inference_steps, 
                    generator=generator,
                    guidance_scale=5.0,
                    max_sequence_length=2048
            ).images[0]

        torch.cuda.empty_cache()  # Clean up GPU memory after generation
        progress(1.0, desc="Done!")
        return image, seed
    except Exception as e:
        return None, f"Error generating image: {str(e)}"
 
examples = [
    "a tiny astronaut hatching from an egg on the moon",
    "a cat holding a sign that says hello world",
    "an anime illustration of a wiener schnitzel",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    # Compute the model loading status message ahead of creating the Info component.
    model_status = "Models loaded successfully!"

    info = gr.Info(model_status)

    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# Text to Product
        Using Mistral + Flux + Trellis
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            prompt_button = gr.Button("Refine prompt", scale=0)
        
        refined_prompt = gr.Text(
            label="Refined Prompt", 
            show_label=False,
            max_lines=10, 
            placeholder="Prompt refined by Mistral", 
            container=False,
            max_length=2048,
            )
        
        
        run_button = gr.Button("Create visual", scale=0)

        generated_image = gr.Image(label="Generated Image", show_label=False)
        
        with gr.Accordion("Advanced Settings Mistral", open=False):
            gr.Slider(
                label="Temperature",
                value=0.9,
                minimum=0.0,
                maximum=1.0,
                step=0.05,
                interactive=True,
                info="Higher values produce more diverse outputs",
            ),
            gr.Slider(
                label="Max new tokens",
                value=256,
                minimum=0,
                maximum=1048,
                step=64,
                interactive=True,
                info="The maximum numbers of new tokens",
            ),
            gr.Slider(
                label="Top-p (nucleus sampling)",
                value=0.90,
                minimum=0.0,
                maximum=1,
                step=0.05,
                interactive=True,
                info="Higher values sample more low-probability tokens",
            ),
            gr.Slider(
                label="Repetition penalty",
                value=1.2,
                minimum=1.0,
                maximum=2.0,
                step=0.05,
                interactive=True,
                info="Penalize repeated tokens",
            )
        
        with gr.Accordion("Advanced Settings Flux", 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():
                
                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,
                )
            
            with gr.Row():
                
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=4,
                )
        
        gr.Examples(
            examples = examples,
            fn = infer,
            inputs = [prompt],
            outputs = [generated_image, seed],
            cache_examples="lazy"
        )


    gr.on(
        triggers=[prompt_button.click, prompt.submit],
        fn = refine_prompt,
        inputs = [prompt],
        outputs = [refined_prompt]
    )

    gr.on(
        triggers=[run_button.click],
        fn = infer,
        inputs = [refined_prompt, seed, randomize_seed, width, height, num_inference_steps],
        outputs = [generated_image, seed]
    )

demo.launch()