app.py

import gradio as gr
from gradio_client import Client
import random
import time
import os

# --- 1. CONFIGURATION ---
TARGET_URL = "https://black-forest-labs-flux-2-dev.hf.space"

# Token List (Aman dengan trik +)
HF_TOKENS = [
    "hf_" + "PiRCDDtPcPFMLWkTkVaZmzoleHOunXnLIA", "hf_" + "BHvZXGICstaktSwycmwNmzHGrTNmKxnlRZ",
    "hf_" + "ZdgawyTPzXIpwhnRYIteUKSMsWnEDtGKtM", "hf_" + "nMiFYAFsINxAJWPwiCQlaunmdgmrcxKoaT",
    "hf_" + "PccpUIbTckCiafwErDLkRlsvqhgtfZaBHL", "hf_" + "faGyXBPfBkaHXDMUSJtxEggonhhZbomFIz",
    "hf_" + "SndsPaRWsevDXCgZcSjTUlBYUJqOkSfFmn", "hf_" + "CqobFdUpeVCeuhUaiuXwvdczBUmoUHXRGa",
    "hf_" + "JKCQYUhhHPPkpucegqkNSyureLdXpmeXRF", "hf_" + "tBYfslUwHNiNMufzwAYIlrDVovEWmOQulC",
    "hf_" + "LKLdrdUxyUyKODSUthmqHXqDMfHrQueera", "hf_" + "ivSBboJYQVcifWkCNcOTOnxUQrZOtOglnU"
]

def get_valid_token():
    return random.choice(HF_TOKENS)

# --- 2. GENERATION LOGIC ---
def process_generation(prompt, width, height, guidance, steps, seed):
    """
    Logic persis seperti contohmu:
    1. Ambil token
    2. Set Header Authorization
    3. Tembak client.predict dengan api_name='/infer'
    """
    max_retries = 5
    attempt = 0
    
    # Validasi input (cegah error type)
    width = int(width)
    height = int(height)
    guidance = float(guidance)
    steps = int(steps)
    seed = int(seed)

    while attempt < max_retries:
        try:
            current_token = get_valid_token()
            print(f"🚀 Generating with token ending ...{current_token[-5:]}")
            
            # --- KUNCI: Client dengan Header Manual ---
            client = Client(
                TARGET_URL, 
                headers={"Authorization": f"Bearer {current_token}"}
            )
            
            # Request ke Flux
            result = client.predict(
                prompt=prompt,
                seed=seed if seed != -1 else 0,
                randomize_seed=(seed == -1),
                width=width,
                height=height,
                guidance_scale=guidance,
                num_inference_steps=steps,
                api_name="/infer"
            )
            
            # Handle return tuple/list dari Flux
            if isinstance(result, (list, tuple)):
                return result[0]
            return result

        except Exception as e:
            err_msg = str(e)
            print(f"❌ Error (Attempt {attempt+1}): {err_msg}")
            
            # Retry jika error kuota/queue
            if "429" in err_msg or "quota" in err_msg.lower() or "queue" in err_msg.lower():
                print("🔄 Token limit, switching token...")
            else:
                pass # Tetap retry siapa tau network glitch
            
            attempt += 1
            time.sleep(1)

    return None # Gagal total

# --- 3. UI BLOCKS (Solusi Error 'allow_flagging') ---
# Menggunakan gr.Blocks seperti contohmu, BUKAN gr.Interface
with gr.Blocks() as app:
    # Invisible inputs untuk API endpoint
    with gr.Row(visible=False):
        prompt = gr.Textbox(label="Prompt")
        width = gr.Number(value=1024, label="Width")
        height = gr.Number(value=1024, label="Height")
        guidance = gr.Number(value=3.5, label="Guidance")
        steps = gr.Number(value=28, label="Steps")
        seed = gr.Number(value=-1, label="Seed")
    
    # Output Image
    out_image = gr.Image(label="Result")
    
    # Tombol Trigger (Hidden juga gapapa, yg penting ada event click)
    btn_gen = gr.Button("Generate", visible=False)
    
    # --- API ENDPOINT DEFINITION ---
    # api_name="generate" membuat endpoint: /api/generate
    btn_gen.click(
        process_generation, 
        inputs=[prompt, width, height, guidance, steps, seed], 
        outputs=out_image, 
        api_name="generate"
    )

    # Tambahan: Load event agar bisa testing lewat GUI jika perlu
    # Tapi utamanya ini buat API
    with gr.Row():
        gr.Markdown("## Flux Load Balancer API Ready")

# Launch
if __name__ == "__main__":
    app.queue(max_size=20).launch(server_name="0.0.0.0", server_port=7860)