app.py

import streamlit as st
from huggingface_hub import InferenceClient
from gradio_client import Client
import re

# Load custom CSS
with open('style.css') as f:
    st.markdown(f'<style>{f.read()}</style>', unsafe_allow_html=True)

# Initialize the HuggingFace Inference Client
text_client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.1")
image_client = Client("Boboiazumi/animagine-xl-3.1")

def format_prompt_for_description(name, hair_color, personality, outfit_style, hobbies, favorite_food, background_story):
    prompt = f"Create a waifu character named {name} with {hair_color} hair, a {personality} personality, and wearing a {outfit_style}. "
    prompt += f"Her hobbies include {hobbies}. Her favorite food is {favorite_food}. Here is her background story: {background_story}."
    return prompt

def format_prompt_for_image(name, hair_color, personality, outfit_style):
    prompt = f"Generate an image prompt for a waifu character named {name} with {hair_color} hair, a {personality} personality, and wearing a {outfit_style}."
    return prompt

def clean_generated_text(text):
    # Remove any unwanted trailing tags or characters like </s>
    clean_text = re.sub(r'</s>$', '', text).strip()
    return clean_text

def generate_text(prompt, temperature=0.9, max_new_tokens=512, top_p=0.95, repetition_penalty=1.0):
    temperature = max(temperature, 1e-2)
    generate_kwargs = dict(
        temperature=temperature,
        max_new_tokens=max_new_tokens,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        seed=42,
    )
    try:
        stream = text_client.text_generation(prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
        output = ""
        for response in stream:
            output += response.token.text
        return clean_generated_text(output)
    except Exception as e:
        st.error(f"Error generating text: {e}")
        return ""

def generate_image(prompt):
    try:
        result = image_client.predict(
            prompt,  # Image prompt
            "",  # Negative prompt
            0,  # Seed
            1024,  # Width
            1024,  # Height
            7.0,  # Guidance scale
            28,  # Number of inference steps
            'Euler a',  # Sampler
            '896 x 1152',  # Aspect Ratio
            '(None)',  # Style Preset
            'Standard v3.1',  # Quality Tags Presets
            False,  # Use Upscaler
            0.55,  # Upscaler strength
            1.5,  # Upscale by
            True,  # Add Quality Tags
            api_name="/run"
        )
        # Check if result is not empty and contains the expected structure
        if result and isinstance(result, list) and len(result) > 0 and 'image' in result[0]:
            return result[0]['image']
        else:
            st.error("Unexpected result format from the Gradio API.")
            return None
    except Exception as e:
        st.error(f"Error generating image: {e}")
        st.write("Full error details:", e)
        return None

def main():
    st.title("Enhanced Waifu Character Generator")

    # User inputs
    col1, col2 = st.columns(2)
    with col1:
        name = st.text_input("Name of the Waifu")
        hair_color = st.selectbox("Hair Color", ["Blonde", "Brunette", "Red", "Black", "Blue", "Pink"])
        personality = st.selectbox("Personality", ["Tsundere", "Yandere", "Kuudere", "Dandere", "Genki", "Normal"])
        outfit_style = st.selectbox("Outfit Style", ["School Uniform", "Maid Outfit", "Casual", "Kimono", "Gothic Lolita"])
    with col2:
        hobbies = st.text_input("Hobbies")
        favorite_food = st.text_input("Favorite Food")
        background_story = st.text_area("Background Story")
        system_prompt = st.text_input("Optional System Prompt", "")

    # Advanced settings
    with st.expander("Advanced Settings"):
        temperature = st.slider("Temperature", 0.0, 1.0, 0.9, step=0.05)
        max_new_tokens = st.slider("Max new tokens", 0, 8192, 512, step=64)
        top_p = st.slider("Top-p (nucleus sampling)", 0.0, 1.0, 0.95, step=0.05)
        repetition_penalty = st.slider("Repetition penalty", 1.0, 2.0, 1.0, step=0.05)

    # Initialize session state for generated text and image prompt
    if "character_description" not in st.session_state:
        st.session_state.character_description = ""
    if "image_prompt" not in st.session_state:
        st.session_state.image_prompt = ""
    if "image_path" not in st.session_state:
        st.session_state.image_path = ""

    # Generate button
    if st.button("Generate Waifu"):
        with st.spinner("Generating waifu character..."):
            description_prompt = format_prompt_for_description(name, hair_color, personality, outfit_style, hobbies, favorite_food, background_story)
            image_prompt = format_prompt_for_image(name, hair_color, personality, outfit_style)

            # Generate character description
            st.session_state.character_description = generate_text(description_prompt, temperature, max_new_tokens, top_p, repetition_penalty)
            
            # Generate image prompt
            st.session_state.image_prompt = generate_text(image_prompt, temperature, max_new_tokens, top_p, repetition_penalty)
            
            # Generate image from image prompt
            st.session_state.image_path = generate_image(st.session_state.image_prompt)
            
            st.success("Waifu character generated!")

    # Display the generated character and image prompt
    if st.session_state.character_description:
        st.subheader("Generated Waifu Character")
        st.write(st.session_state.character_description)
    if st.session_state.image_prompt:
        st.subheader("Image Prompt")
        st.write(st.session_state.image_prompt)
    if st.session_state.image_path:
        st.subheader("Generated Image")
        st.image(st.session_state.image_path, caption="Generated Waifu Image")

if __name__ == "__main__":
    main()