app.py

from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
import numpy as np
import tensorflow as tf
from tensorflow.keras import layers
from huggingface_hub import hf_hub_download
from PIL import Image
import io

app = FastAPI()

LEAKY_ALPHA = 0.2
Z_DIM       = 100
NUM_CLASSES = 32

CLASSES = [
    'ain','al','aleff','bb','dal','dha','dhad','fa',
    'gaaf','ghain','ha','haa','jeem','kaaf','khaa','la',
    'laam','meem','nun','ra','saad','seen','sheen','ta',
    'taa','thaa','thal','toot','waw','ya','yaa','zay'
]

def build_generator():
    noise_input = tf.keras.Input(shape=(Z_DIM,), dtype='float32', name="noise")
    label_input = tf.keras.Input(shape=(1,),     dtype='int32',   name="label")
    label_emb   = layers.Embedding(NUM_CLASSES, Z_DIM)(label_input)
    label_emb   = layers.Flatten()(label_emb)
    x = layers.Concatenate()([noise_input, label_emb])
    x = layers.Dense(4 * 4 * 512, use_bias=False)(x)
    x = layers.BatchNormalization()(x)
    x = layers.LeakyReLU(LEAKY_ALPHA)(x)
    x = layers.Reshape((4, 4, 512))(x)
    for filters in [512, 256, 128, 64, 64, 32]:
        x = layers.Conv2DTranspose(filters, 4, strides=2, padding='same', use_bias=False)(x)
        x = layers.BatchNormalization()(x)
        x = layers.LeakyReLU(LEAKY_ALPHA)(x)
    x      = layers.Conv2D(1, 3, padding='same')(x)
    output = layers.Activation('tanh', dtype='float32')(x)
    return tf.keras.Model([noise_input, label_input], output, name="generator")

# Load model once at startup
print("Loading generator weights...")
weights_path = hf_hub_download(
    repo_id="zainebrahem/my-cgan",
    filename="generator.weights.h5"
)
generator = build_generator()
generator.load_weights(weights_path)
print("Generator ready!")

class GenerateRequest(BaseModel):
    label: int  # 0 to 31

@app.get("/")
def root():
    return {"status": "CGAN API is running", "classes": CLASSES}

@app.post("/generate")
def generate(req: GenerateRequest):
    if req.label < 0 or req.label >= NUM_CLASSES:
        return {"error": f"Label must be between 0 and {NUM_CLASSES - 1}"}
    noise        = tf.random.normal([1, Z_DIM])
    label_tensor = tf.constant([[req.label]], dtype=tf.int32)
    image        = generator([noise, label_tensor], training=False)
    image        = (image[0].numpy() * 127.5 + 127.5).clip(0, 255).astype(np.uint8)
    img_pil      = Image.fromarray(image[:, :, 0], mode='L')
    buf          = io.BytesIO()
    img_pil.save(buf, format='PNG')
    buf.seek(0)
    return StreamingResponse(buf, media_type="image/png")

@app.get("/classes")
def get_classes():
    return {i: c for i, c in enumerate(CLASSES)}