app.py

import subprocess
import os
import gradio as gr
import torch
from PIL import Image, ImageEnhance
from pygltflib import GLTF2
from pygltflib.utils import ImageFormat, Texture, Material, Image as GLTFImage
import sys
import spaces


if torch.cuda.is_available():
    device = "cuda"
    print("Using GPU")
else:
    device = "cpu"
    print("Using CPU")


subprocess.run(["git", "clone", "https://github.com/Nick088Official/Stable_Diffusion_Finetuned_Minecraft_Skin_Generator.git"])

os.chdir("Stable_Diffusion_Finetuned_Minecraft_Skin_Generator")


@spaces.GPU()
def run_inference(prompt, stable_diffusion_model, num_inference_steps, guidance_scale, model_precision_type, seed, filename, verbose):
    
    if stable_diffusion_model == '2':
        sd_model = "minecraft-skins"
    else:
        sd_model = "minecraft-skins-sdxl"

    inference_command = f"python Scripts/{sd_model}.py '{prompt}' {num_inference_steps} {guidance_scale} {model_precision_type} {seed} {filename} {'--verbose' if verbose else ''}"
    
    os.system(inference_command)

    # view it in 3d
    os.chdir("Scripts")
    command_3d_model = f"python to_3d_model.py '{filename}'"
    os.system(command_3d_model)
    os.chdir("..")
    glb_path = os.path.join(f"output_minecraft_skins/{filename}_3d_model.glb")
    return os.path.join(f"output_minecraft_skins/{filename}"), glb_path


def custom_output(image_path, glb_path):
    if glb_path is None:
        return image_path
    else:
        return [image_path, gr.Model3D(clear_color=[0.0, 0.0, 0.0, 0.0], label="3D Model", path=glb_path)]


with gr.Blocks() as minecraft_skin_generator:
    with gr.Row():
        prompt = gr.Textbox(label="Your Prompt", info="What the Minecraft Skin should look like")
        stable_diffusion_model = gr.Dropdown(['2', 'xl'], value="xl", label="Stable Diffusion Model", info="Choose which Stable Diffusion Model to use, xl understands prompts better")
    num_inference_steps = gr.Number(label="Number of Inference Steps", precision=0, value=25)
    guidance_scale = gr.Number(minimum=0.1, value=7.5, label="Guidance Scale", info="The number of denoising steps of the image. More denoising steps usually lead to a higher quality image at the cost of slower inference")
    model_precision_type = gr.Dropdown(["fp16", "fp32"], value="fp16", label="Model Precision Type", info="The precision type to load the model, like fp16 which is faster, or fp32 which gives better results")
    seed = gr.Number(value=42, label="Seed", info="A starting point to initiate generation, put 0 for a random one")
    filename = gr.Textbox(label="Output Image Name", info="The name of the file of the output image skin, keep the .png", value="output-skin.png")
    verbose = gr.Checkbox(label="Verbose Output", info="Produce more detailed output while running", value=False)
    see_in_3d = gr.Checkbox(label="See in 3D", info="View the generated skin in 3D", value=False)

    image_path, glb_path = run_inference(prompt, stable_diffusion_model, num_inference_steps, guidance_scale, model_precision_type, seed, filename, verbose)

    with gr.Row():
        output = gr.Image(label="Generated Minecraft Skin Image Asset")
        if see_in_3d:
            output.style(height=500)
            output.style(width=500)
            output.style(display="flex")
            output.style(justify_content="center")
            output.style(align_items="center")
            output.style(flex_wrap="wrap")
            output.style(grid_template_columns="repeat(auto-fill, minmax(250px, 1fr))")
            output.style(grid_gap="10px")
            output.style(overflow="auto")
            output.style(padding="10px")
            output.style(box_sizing="border-box")
            output.style(border="1px solid #ccc")
            output.style(border_radius="5px")
            output.style(margin="10px 0")
            output.style(background_color="#f9f9f9")

            output.render(custom_output, inputs=[image_path, glb_path])
        else:
            output.render(image_path)

minecraft_skin_generator.launch(show_api=False, share=True)