import gradio as gr
from transformers import pipeline
from diffusers import AutoPipelineForText2Image
import torch

# -------- TEXT MODEL --------
text_model = pipeline("text-generation", model="distilgpt2")

# -------- IMAGE MODEL --------
pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/sd-turbo",
    torch_dtype=torch.float32
)
pipe = pipe.to("cpu")

# -------- FUNCTION --------
def respond(message, history):
    user_input = message

    if any(word in user_input.lower() for word in ["draw", "image", "generate", "picture"]):
        image = pipe(
            user_input,
            num_inference_steps=1,
            guidance_scale=0.0
        ).images[0]

        history.append({"role": "user", "content": user_input})
        history.append({"role": "assistant", "content": "🖼️ Image generated"})

        return history, image

    else:
        result = text_model(user_input, max_length=80)[0]["generated_text"]

        history.append({"role": "user", "content": user_input})
        history.append({"role": "assistant", "content": result})

        return history, None

# -------- UI --------
with gr.Blocks() as demo:
    gr.Markdown("# AI Chat + Image (Gradio 6 Fixed)")

    chatbot = gr.Chatbot(type="messages")
    image_output = gr.Image()

    msg = gr.Textbox(placeholder="Ask or generate image...")
    state = gr.State([])

    msg.submit(respond, [msg, state], [chatbot, image_output])

demo.launch()