diffusers_empr.py

import torch
import gradio as gr
from PIL import Image
from diffusers import DiffusionPipeline
from transformers import pipeline

# Diffuser
# diffuser = DiffusionPipeline.from_pretrained(
#     "bb1070/catvton-acne-4000-full",
#     torch_dtype=torch.float16
# ).to("cuda")

# Clasificador de acné (GRADE REAL)
classifier = pipeline(
    "image-classification",
    model="imfarzanansari/skintelligent-acne",
    device=0
)

GRADE_MAP = {
    "Clear Skin": 0,
    "Mild Acne": 1,
    "Moderate Acne": 2,
    "Severe Acne": 3
}

# def analyze_skin(image, use_diffuser):
#     if image is None:
#         return None, "No image", "-", "-"

#     prompt = "face with acne, dermatology photo, neutral lighting"
#     image = diffuser(prompt, image=image).images[0]

#     # Clasificación
#     result = classifier(image)
#     top = result[0]

#     label = top["label"]
#     confidence = round(top["score"], 3)
#     grade = GRADE_MAP.get(label, "N/A")

#     return image, label, grade, confidence

with gr.Blocks(title="Skin Acne AI") as demo:
    gr.Markdown("## Análisis de Acné con IA")
    gr.Markdown(
        "Sube una imagen de la piel. "
        "El sistema devuelve un **GRADE dermatológico no clínico**."
    )

    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Imagen de la piel")
            btn = gr.Button("Analizar")

        with gr.Column():
            output_image = gr.Image(label="Imagen procesada")
            label_out = gr.Textbox(label="Diagnóstico IA")
            grade_out = gr.Textbox(label="Grade (0–3)")
            conf_out = gr.Textbox(label="Confianza")

    btn.click(
        # analyze_skin,
        inputs=[input_image],
        outputs=[output_image, label_out, grade_out, conf_out]
    )

demo.launch()