import gradio as gr
from gradio_client import Client
import os 
import numpy as np
import random

hf_token = os.environ.get("HF_TKN")
MAX_SEED = np.iinfo(np.int32).max

def get_caption(image_in):
    client = Client("https://fffiloni-moondream1.hf.space/", hf_token=hf_token)
    result = client.predict(
		image_in,	# filepath  in 'image' Image component
		"Describe the image",	# str  in 'Question' Textbox component
		api_name="/predict"
    )
    print(result)
    return result

def get_lcm(prompt):
    client = Client("https://latent-consistency-lcm-lora-for-sdxl.hf.space/")
    result = client.predict(
        prompt,	# str  in 'parameter_5' Textbox component
        0.3,	# float (numeric value between 0.0 and 5) in 'Guidance' Slider component
        8,	# float (numeric value between 2 and 10) in 'Steps' Slider component
        0,	# float (numeric value between 0 and 12013012031030) in 'Seed' Slider component
        True,	# bool  in 'Randomize' Checkbox component
        api_name="/predict"
    )
    print(result)
    return result[0]

def get_sdxl_lightning(prompt):
    client = Client("AP123/SDXL-Lightning")
    result = client.predict(
        prompt,	# str  in 'parameter_1' Textbox component
        "4-Step",
        api_name="/generate_image"
    )
    print(result)
    return result

def get_turbo(prompt):
    seed = random.randint(0, MAX_SEED)
    print(f"SEED: {seed}")
    client = Client("https://diffusers-unofficial-sdxl-turbo-i2i-t2i.hf.space/")
    result = client.predict(
        None,	# filepath  in 'Webcam' Image component
        prompt,	# str  in 'parameter_5' Textbox component
        0.7,	# float (numeric value between 0.0 and 1.0) in 'Strength' Slider component
        8,	# float (numeric value between 1 and 10) in 'Steps' Slider component
        seed,	# float (numeric value between 0 and MAX_SEED) in 'Seed' Slider component
        api_name="/predict"
    )
    print(result)
    return result

def infer(image_in, chosen_method):
    caption = get_caption(image_in)
    if chosen_method == "LCM" :
        img_var = get_lcm(caption)
    elif chosen_method == "SDXL Lightning" :
        img_var = get_sdxl_lightning(caption)
    elif chosen_method == "SDXL Turbo" :
        img_var = get_turbo(caption)
    return img_var

gr.Interface(
    title = "Supa Fast Image Variation",
    description = "Get quick image variation from image input, using <a href='https://huggingface.co/vikhyatk/moondream1' target='_blank'>moondream1</a> for caption, and <a href='https://huggingface.co/spaces/latent-consistency/lcm-lora-for-sdxl' target='_blank'>LCM SDXL</a>, <a href='https://huggingface.co/spaces/AP123/SDXL-Lightning' target='_blank'>SDXL Lightning</a> or <a href='https://huggingface.co/spaces/diffusers/unofficial-SDXL-Turbo-i2i-t2i' target='_blank'>SDXL Turbo</a> for image generation",
    fn = infer,
    inputs = [
        gr.Image(type="filepath", label="Image input"),
        gr.Dropdown(label="Choose a model", choices=["LCM", "SDXL Lightning", "SDXL Turbo"], value="SDXL Lightning")
    ],
    outputs = [
        gr.Image(label="Image variation")
    ],
    examples = [
        ["examples/frog_clean.jpg", "LCM"],
        ["examples/martin_pecheur.jpeg", "SDXL Turbo"],
        ["examples/forest_deer.png", "SDXL Lightning"]
    ],
    cache_examples = False,
    concurrency_limit = 2
).queue(max_size=25).launch(show_api=False, show_error=True)