app.py

import json
import string
import uuid
import os
import logging
import zipfile
import importlib
import wandb
from contextlib import redirect_stdout, redirect_stderr
import spaces

USE_WANDB = "WANDB_API_KEY" in os.environ
if USE_WANDB:
    wandb.login(key=os.environ["WANDB_API_KEY"])
else:
    print("Warning: WANDB_API_KEY not set. Skipping wandb logging.")

import gradio as gr
import pandas as pd
import time
import sys
from datetime import datetime
import re

# --- Configuration ---
DEFAULT_MATERIALS_CSV = "default_materials.csv"
GRADIO_OUTPUT_BASE_DIR = "output"
os.makedirs(GRADIO_OUTPUT_BASE_DIR, exist_ok=True)

REQUIRED_SCRIPT_COLS = ["Brand", " Name", " TD", " Color"]
DISPLAY_COL_MAP = {
    "Brand": "Brand",
    " Name": "Name",
    " TD": "TD",
    " Color": "Color (Hex)",
}

def exc_text(exc: BaseException) -> str:
    txt = str(exc).strip()
    if txt:
        return txt
    if exc.args:
        return " ".join(str(a) for a in exc.args).strip()
    return exc.__class__.__name__

def ensure_required_cols(df, *, in_display_space):
    target_cols = (
        DISPLAY_COL_MAP if in_display_space else {k: k for k in REQUIRED_SCRIPT_COLS}
    )
    df_fixed = df.copy()
    for col_script, col_display in target_cols.items():
        if col_display not in df_fixed.columns:
            if "TD" in col_display:
                default = 0.0
            elif "Color" in col_display:
                default = "#000000"
            elif "Owned" in col_display:
                default = "false"
            else:
                default = ""
            df_fixed[col_display] = default
    return df_fixed[list(target_cols.values())]

def rgba_to_hex(col: str) -> str:
    if not isinstance(col, str):
        return col
    col = col.strip()
    if col.startswith("#"):
        return col.upper()

    m = re.match(
        r"rgba?\(\s*([\d.]+)\s*,\s*([\d.]+)\s*,\s*([\d.]+)(?:\s*,\s*[\d.]+)?\s*\)",
        col,
    )
    if not m:
        return col

    r, g, b = (int(float(x)) for x in m.groups()[:3])
    return "#{:02X}{:02X}{:02X}".format(r, g, b)

def zip_dir_no_compress(src_dir: str, dest_zip: str) -> str:
    t0 = time.time()
    with zipfile.ZipFile(dest_zip, "w",
                         compression=zipfile.ZIP_STORED,
                         allowZip64=True) as zf:
        for root, _, files in os.walk(src_dir):
            for fname in files:
                fpath = os.path.join(root, fname)
                zf.write(fpath, os.path.relpath(fpath, src_dir))
    print(f"Zipping finished in {time.time() - t0:.1f}s")
    return dest_zip

def get_script_args_info(exclude_args=None):
    if exclude_args is None:
        exclude_args = []

    all_args_info = [
        {
            "name": "--iterations",
            "type": "number",
            "default": 4000,
            "help": "Number of optimization iterations",
        },
        {
            "name": "--layer_height",
            "type": "number",
            "default": 0.04,
            "step": 0.01,
            "help": "Layer thickness in mm",
        },
        {
            "name": "--max_layers",
            "type": "number",
            "default": 75,
            "precision": 0,
            "help": "Maximum number of layers",
        },
        {
            "name": "--learning_rate",
            "type": "number",
            "default": 0.015,
            "step": 0.001,
            "help": "Learning rate for optimization",
        },
        {
            "name": "--background_height",
            "type": "number",
            "default": 0.4,
            "step": 0.01,
            "help": "Height of the background in mm",
        },
        {
            "name": "--background_color",
            "type": "colorpicker",
            "default": "#000000",
            "help": "Background color",
        },
        {
            "name": "--stl_output_size",
            "type": "number",
            "default": 100,
            "precision": 0,
            "help": "Size of the longest dimension of the output STL file in mm",
        },
        {
            "name": "--nozzle_diameter",
            "type": "number",
            "default": 0.4,
            "step": 0.1,
            "help": "Diameter of the printer nozzle in mm",
        },
        {
            "name": "--pruning_max_colors",
            "type": "number",
            "default": 100,
            "precision": 0,
            "help": "Max number of colors allowed after pruning",
        },
        {
            "name": "--pruning_max_swaps",
            "type": "number",
            "default": 50,
            "precision": 0,
            "help": "Max number of swaps allowed after pruning",
        },
        {
            "name": "--pruning_max_layer",
            "type": "number",
            "default": 75,
            "precision": 0,
            "help": "Max number of layers allowed after pruning",
        },
        {
            "name": "--warmup_fraction",
            "type": "slider",
            "default": 1.0,
            "min": 0.0,
            "max": 1.0,
            "step": 0.01,
            "help": "Fraction of iterations for keeping the tau at the initial value",
        },
        {
            "name": "--learning_rate_warmup_fraction",
            "type": "slider",
            "default": 0.01,
            "min": 0.0,
            "max": 1.0,
            "step": 0.01,
            "help": "Fraction of iterations that the learning rate is increasing (warmup)",
        },
        {
            "name": "--early_stopping",
            "type": "number",
            "default": 5000,
            "precision": 0,
            "help": "Number of steps without improvement before stopping",
        },
        {
            "name": "--fast_pruning_percent",
            "type": "slider",
            "default": 0.05,
            "min": 0.0,
            "max": 1.0,
            "step": 0.01,
            "help": "Percentage of increment search for fast pruning.",
        },
        {
            "name": "--random_seed",
            "type": "number",
            "default": 0,
            "precision": 0,
            "help": "Specify the random seed, or use 0 for automatic generation",
        },
        {
            "name": "--num_init_rounds",
            "type": "number",
            "default": 8,
            "precision": 0,
            "help": "Number of rounds to choose the starting height map from.",
        },
        {
            "name": "--flatforge",
            "type": "checkbox",
            "default": False,
            "help": "Enable FlatForge mode to generate separate STL files for each color",
        },
        {
            "name": "--cap_layers",
            "type": "number",
            "default": 0,
            "precision": 0,
            "help": "Number of complete clear/transparent layers to add on top in FlatForge mode",
        },
    ]
    return [arg for arg in all_args_info if arg["name"] not in exclude_args]

# initial data that will be used if no CSV exists
initial_filament_data = {
    "Brand": ["Generic", "Generic", "Generic", "Generic", "Generic", "Generic"],
    " Name": ["PLA Black", "PLA Grey", "PLA White", "PLA Red", "PLA Green", "PLA Blue"],
    " TD": [5.0, 5.0, 5.0, 5.0, 5.0],
    " Color": ["#000000", "#808080", "#FFFFFF", "#FF0000", "#00FF00", "#0000FF"],
    " Owned": ["true", "true", "true", "true", "true", "true"],
}

def normalize_filament_df(df: pd.DataFrame) -> pd.DataFrame:
    df = df.copy()

    df.columns = [c.strip() for c in df.columns]

    rename_map = {
        "Name": " Name",
        "TD": " TD",
        "Color": " Color",
        "Owned": " Owned",
    }
    for src, dst in rename_map.items():
        if src in df.columns and dst not in df.columns:
            df.rename(columns={src: dst}, inplace=True)

    if " TD" in df.columns:
        df[" TD"] = pd.to_numeric(df[" TD"], errors="coerce").fillna(0.0)
    else:
        df[" TD"] = 0.0

    if " Color" in df.columns:
        df[" Color"] = df[" Color"].astype(str)
    else:
        df[" Color"] = "#000000"

    if " Owned" not in df.columns:
        df[" Owned"] = "false"
    else:
        df[" Owned"] = df[" Owned"].astype(str)

    if "Brand" not in df.columns:
        df["Brand"] = ""

    ordered_cols = ["Brand", " Name", " TD", " Color", " Owned"]
    df = df[[c for c in ordered_cols if c in df.columns]]
    return df

# load CSV if present
if os.path.exists(DEFAULT_MATERIALS_CSV):
    try:
        loaded_df = pd.read_csv(DEFAULT_MATERIALS_CSV, index_col=False)
        loaded_df = normalize_filament_df(loaded_df)

        initial_df = loaded_df.copy()

        initial_filament_data = {
            "Brand": initial_df["Brand"].tolist(),
            " Name": initial_df[" Name"].tolist(),
            " TD": initial_df[" TD"].tolist(),
            " Color": initial_df[" Color"].tolist(),
        }
        if " Owned" in initial_df.columns:
            initial_filament_data[" Owned"] = initial_df[" Owned"].astype(str).tolist()
        else:
            initial_filament_data[" Owned"] = ["false"] * len(initial_df)
    except Exception as e:
        print(f"Warning: Could not load {DEFAULT_MATERIALS_CSV}: {e}. Using default.")
        initial_df = pd.DataFrame(initial_filament_data)
else:
    initial_df = pd.DataFrame(initial_filament_data)
    initial_df.to_csv(DEFAULT_MATERIALS_CSV, index=False)

def run_autoforge_process(cmd, log_path):
    from joblib import parallel_backend
    cli_args = cmd[1:]
    autoforge_main = importlib.import_module("autoforge.__main__")

    exit_code = 0
    with open(log_path, "w", buffering=1, encoding="utf-8") as log_f, \
            redirect_stdout(log_f), redirect_stderr(log_f), parallel_backend("threading", n_jobs=-1):
        try:
            sys.argv = ["autoforge"] + cli_args
            autoforge_main.main()
        except SystemExit as e:
            exit_code = e.code
        except Exception as e:
            log_f.write(f"\nERROR: {e}\n")
            exit_code = -1

    return exit_code

def create_empty_error_outputs(log_message=""):
    return (
        log_message,
        None,
        gr.update(visible=False, interactive=False),
    )

def load_filaments_from_json_upload(file_obj):
    if file_obj is None:
        current_script_df = filament_df_state.value
        if current_script_df is not None and not current_script_df.empty:
            return current_script_df.rename(
                columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
            )
        return initial_df.copy().rename(
            columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
        )

    try:
        with open(file_obj.name, "r", encoding="utf-8") as f:
            data = json.load(f)
        if isinstance(data, dict) and "Filaments" in data:
            data = data["Filaments"]

        df_loaded = pd.DataFrame(data)
        df_loaded.columns = [c.strip() for c in df_loaded.columns]

        rename_map = {
            "Name": " Name",
            "Transmissivity": " TD",
            "Color": " Color",
            "Owned": " Owned",
        }
        df_loaded.rename(
            columns={k: v for k, v in rename_map.items() if k in df_loaded.columns},
            inplace=True,
        )

        df_loaded = normalize_filament_df(df_loaded)

        filament_df_state.value = df_loaded.copy()

        return df_loaded.rename(
            columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
        )

    except Exception as e:
        gr.Error(f"Error loading JSON: {e}")
        return filament_table.value

with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# [Autoforge](https://github.com/hvoss-techfak/AutoForge) Web UI")

    filament_df_state = gr.State(initial_df.copy())
    current_run_output_dir = gr.State(None)

    with gr.Tabs():
        with gr.TabItem("Filament Management"):
            gr.Markdown(
                'Manage your filament list here. This list will be used by Autoforge during the optimization process.'
            )
            gr.Markdown(
                'If you have Hueforge, you can export your filaments under "Filaments -> Export" in the Hueforge software. Please make sure to select "CSV" instead of "JSON" during the export dialog.'
            )
            gr.Markdown(
                'If you want to load your personal library of Hueforge filaments, you can also simply paste this path into your explorer address bar: %APPDATA%\\HueForge\\Filaments\\ and import your "personal_library.json" using the "Load Filaments Json" button.'
            )
            gr.Markdown(
                'To remove a filament simply right-click on any of the fields and select "Delete Row"'
            )
            gr.Markdown(
                'Hint: If you have an AMS 3d printer try giving it your entire filament library and then set "pruning_max_colors" under "Autoforge Parameters" in the second tab to your number of AMS slots.'
                ' Autoforge will automatically select the best matching colors for your image.'
            )
            with gr.Row():
                load_csv_button = gr.UploadButton(
                    "Load Filaments CSV", file_types=[".csv"]
                )
                load_json_button = gr.UploadButton(
                    "Load Filaments JSON", file_types=[".json"]
                )
                save_csv_button = gr.Button("Save Current Filaments to CSV")

            filament_table = gr.DataFrame(
                value=ensure_required_cols(
                    initial_df.copy().rename(
                        columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
                    ),
                    in_display_space=True,
                ),
                headers=["Brand", "Name", "TD", "Color (Hex)"],
                datatype=["str", "str", "number", "str"],
                interactive=True,
                label="Filaments",
            )

            gr.Markdown("## Add New Filament")
            with gr.Row():
                new_brand = gr.Textbox(label="Brand")
                new_name = gr.Textbox(label="Name")
            with gr.Row():
                new_td = gr.Number(
                    label="TD (Transmission/Opacity)",
                    value=1.0,
                    minimum=0,
                    maximum=100,
                    step=0.1,
                )
                new_color_hex = gr.ColorPicker(label="Color", value="#FF0000")
            add_filament_button = gr.Button("Add Filament to Table")
            download_csv_trigger = gr.File(
                label="Download Filament CSV", visible=False, interactive=False
            )

            def update_filament_df_state_from_table(display_df):
                display_df = ensure_required_cols(display_df, in_display_space=True)
                if "Color (Hex)" in display_df.columns:
                    display_df["Color (Hex)"] = display_df["Color (Hex)"].apply(
                        rgba_to_hex
                    )

                script_df = display_df.rename(
                    columns={"Name": " Name", "TD": " TD", "Color (Hex)": " Color"}
                )
                script_df = ensure_required_cols(script_df, in_display_space=False)
                filament_df_state.value = script_df

            def add_filament_to_table(current_display_df, brand, name, td, color_hex):
                if not brand or not name:
                    gr.Warning("Brand and Name cannot be empty.")
                    return current_display_df

                color_hex = rgba_to_hex(color_hex)

                new_row = pd.DataFrame(
                    [{"Brand": brand, "Name": name, "TD": td, "Color (Hex)": color_hex}]
                )
                updated_display_df = pd.concat(
                    [current_display_df, new_row], ignore_index=True
                )
                update_filament_df_state_from_table(updated_display_df)
                return updated_display_df

            def load_filaments_from_csv_upload(file_obj):
                if file_obj is None:
                    current_script_df = filament_df_state.value
                    if current_script_df is not None and not current_script_df.empty:
                        return current_script_df.rename(
                            columns={
                                " Name": "Name",
                                " TD": "TD",
                                " Color": "Color (Hex)",
                            }
                        )
                    return initial_df.copy().rename(
                        columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
                    )
                try:
                    loaded_script_df = pd.read_csv(file_obj.name, index_col=False)
                    loaded_script_df = normalize_filament_df(loaded_script_df)

                    expected_cols = ["Brand", " Name", " TD", " Color"]
                    if not all(col in loaded_script_df.columns for col in expected_cols):
                        gr.Error(
                            f"CSV must contain columns: {', '.join(expected_cols)}. Found: {loaded_script_df.columns.tolist()}"
                        )
                        current_script_df = filament_df_state.value
                        if (
                            current_script_df is not None
                            and not current_script_df.empty
                        ):
                            return current_script_df.rename(
                                columns={
                                    " Name": "Name",
                                    " TD": "TD",
                                    " Color": "Color (Hex)",
                                }
                            )
                        return initial_df.copy().rename(
                            columns={
                                " Name": "Name",
                                " TD": "TD",
                                " Color": "Color (Hex)",
                            }
                        )
                    filament_df_state.value = loaded_script_df.copy()
                    return loaded_script_df.rename(
                        columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
                    )
                except Exception as e:
                    gr.Error(f"Error loading CSV: {e}")
                    current_script_df = filament_df_state.value
                    if current_script_df is not None and not current_script_df.empty:
                        return current_script_df.rename(
                            columns={
                                " Name": "Name",
                                " TD": "TD",
                                " Color": "Color (Hex)",
                            }
                        )
                    return initial_df.copy().rename(
                        columns={" Name": "Name", " TD": "TD", " Color": "Color (Hex)"}
                    )

            def save_filaments_to_file_for_download(current_script_df_from_state):
                if (
                    current_script_df_from_state is None
                    or current_script_df_from_state.empty
                ):
                    gr.Warning("Filament table is empty. Nothing to save.")
                    return None
                df_to_save = current_script_df_from_state.copy()
                required_cols = ["Brand", " Name", " TD", " Color"]
                if not all(col in df_to_save.columns for col in required_cols):
                    gr.Error(
                        f"Cannot save. DataFrame missing required script columns. Expected: {required_cols}. Found: {df_to_save.columns.tolist()}"
                    )
                    return None
                temp_dir = os.path.join(GRADIO_OUTPUT_BASE_DIR, "_temp_downloads")
                os.makedirs(temp_dir, exist_ok=True)
                temp_filament_csv_path = os.path.join(
                    temp_dir,
                    f"filaments_{datetime.now().strftime('%Y%m%d_%H%M%S')}.csv",
                )
                try:
                    df_to_save.to_csv(temp_filament_csv_path, index=False)
                    gr.Info("Filaments prepared for download.")
                    return gr.File(
                        value=temp_filament_csv_path,
                        label="Download Filament CSV",
                        interactive=True,
                        visible=True,
                    )
                except Exception as e:
                    gr.Error(f"Error saving CSV for download: {e}")
                    return None

            filament_table.change(
                update_filament_df_state_from_table,
                inputs=[filament_table],
                outputs=None,
                queue=False,
            )
            add_filament_button.click(
                add_filament_to_table,
                inputs=[filament_table, new_brand, new_name, new_td, new_color_hex],
                outputs=[filament_table],
            )
            load_csv_button.upload(
                load_filaments_from_csv_upload,
                inputs=[load_csv_button],
                outputs=[filament_table],
            )
            load_json_button.upload(
                load_filaments_from_json_upload,
                inputs=[load_json_button],
                outputs=[filament_table],
            )
            save_csv_button.click(
                save_filaments_to_file_for_download,
                inputs=[filament_df_state],
                outputs=[download_csv_trigger],
            )

        with gr.TabItem("Run Autoforge"):

            accordion_params_dict = {}
            accordion_params_ordered_names = []

            gr.Markdown(
                'Here you can upload an image, adjust the parameters and run the Autoforge process. The filaments from the "Filament Management" Tab are automatically used.'
            )
            gr.Markdown(
                'Please note: Huggingface recently reduced the number of free seconds that a user can use. Due to this, unregistered users can currently not use Autoforge as they are below the required time limit. Please login/register with a huggingface account to use the demo.'
            )

            with gr.Row():
                with gr.Column(scale=1):
                    gr.Markdown("### Input Image (Required)")
                    input_image_component = gr.Image(
                        type="pil",
                        image_mode="RGBA",
                        label="Upload Image",
                        sources=["upload"],
                        interactive=True,
                    )
                with gr.Column(scale=2):
                    gr.Markdown("### Preview")
                    with gr.Accordion("Progress & Output", open=True):
                        final_image_preview = gr.Image(
                            label="Model Preview",
                            type="filepath",
                            interactive=False,
                        )

            with gr.Row():
                with gr.Accordion("Autoforge Parameters", open=False):
                    args_for_accordion = get_script_args_info(
                        exclude_args=["--input_image"]
                    )

                    for arg in args_for_accordion:
                        label, info, default_val = (
                            f"{arg['name']}",
                            arg["help"],
                            arg.get("default"),
                        )
                        if arg["type"] == "number":
                            accordion_params_dict[arg["name"]] = gr.Number(
                                label=label,
                                value=default_val,
                                info=info,
                                minimum=arg.get("min"),
                                maximum=arg.get("max"),
                                step=arg.get(
                                    "step",
                                    0.001 if isinstance(default_val, float) else 1,
                                ),
                                precision=arg.get("precision", None),
                            )
                        elif arg["type"] == "slider":
                            accordion_params_dict[arg["name"]] = gr.Slider(
                                label=label,
                                value=default_val,
                                info=info,
                                minimum=arg.get("min", 0),
                                maximum=arg.get("max", 1),
                                step=arg.get("step", 0.01),
                            )
                        elif arg["type"] == "checkbox":
                            accordion_params_dict[arg["name"]] = gr.Checkbox(
                                label=label, value=default_val, info=info
                            )
                        elif arg["type"] == "colorpicker":
                            accordion_params_dict[arg["name"]] = gr.ColorPicker(
                                label=label, value=default_val, info=info
                            )
                        else:
                            accordion_params_dict[arg["name"]] = gr.Textbox(
                                label=label, value=str(default_val), info=info
                            )
                        accordion_params_ordered_names.append(arg["name"])

            run_button = gr.Button(
                "Run Autoforge Process",
                variant="primary",
                elem_id="run_button_full_width",
            )

            progress_output = gr.Textbox(
                label="Console Output",
                lines=15,
                autoscroll=True,
                #show_copy_button=False,
            )

            with gr.Row():
                download_results = gr.File(
                    label="Download Results (zip)",
                    file_count="single",
                    interactive=True,
                    visible=False,
                )

    @spaces.GPU(duration=150)
    def execute_autoforge_script(
        current_filaments_df_state_val, input_image, *accordion_param_values
    ):

        log_output = []

        if input_image is None:
            gr.Error("Input Image is required! Please upload an image.")
            return create_empty_error_outputs("Error: Input Image is required!")

        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S") + "_" + str(uuid.uuid4())
        run_output_dir_val = os.path.join(GRADIO_OUTPUT_BASE_DIR, f"run_{timestamp}")
        os.makedirs(run_output_dir_val, exist_ok=True)
        current_run_output_dir.value = run_output_dir_val

        if (
            current_filaments_df_state_val is None
            or current_filaments_df_state_val.empty
        ):
            gr.Error("Filament table is empty. Please add filaments.")
            return create_empty_error_outputs("Error: Filament table is empty.")

        temp_filament_csv = os.path.join(run_output_dir_val, "materials.csv")
        df_to_save = current_filaments_df_state_val.copy()
        required_cols = ["Brand", " Name", " TD", " Color"]
        missing_cols = [col for col in required_cols if col not in df_to_save.columns]
        if missing_cols:
            err_msg = (
                f"Error: Filament data is missing columns: {', '.join(missing_cols)}."
            )
            gr.Error(err_msg)
            return create_empty_error_outputs(err_msg)
        try:
            df_to_save.to_csv(temp_filament_csv, index=False)
        except Exception as e:
            err_msg = f"Error saving temporary filament CSV: {e}"
            gr.Error(err_msg)
            return create_empty_error_outputs(err_msg)

        command = ["autoforge"]
        command.extend(["--csv_file", temp_filament_csv])
        command.extend(["--output_folder", run_output_dir_val])
        command.extend(["--disable_visualization_for_gradio", "1"])

        try:
            script_input_image_path = os.path.join(
                run_output_dir_val, "input_image.png"
            )
            input_image.save(script_input_image_path, format="PNG")
            command.extend(["--input_image", script_input_image_path])
        except Exception as e:
            err_msg = f"Error handling input image: {e}"
            gr.Error(err_msg)
            return create_empty_error_outputs(err_msg)

        param_dict = dict(zip(accordion_params_ordered_names, accordion_param_values))
        for arg_name, arg_widget_val in param_dict.items():
            if arg_widget_val is None or arg_widget_val == "":
                arg_info_list = [
                    item for item in get_script_args_info() if item["name"] == arg_name
                ]
                if (
                    arg_info_list
                    and arg_info_list[0]["type"] == "checkbox"
                    and arg_widget_val is False
                ):
                    continue
                else:
                    continue

            if arg_name == "--background_color":
                arg_widget_val = rgba_to_hex(arg_widget_val)

            if isinstance(arg_widget_val, bool):
                if arg_widget_val:
                    command.append(arg_name)
            else:
                command.extend([arg_name, str(arg_widget_val)])

        log_output = [
            "Starting Autoforge process at ",
            f"{datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\n",
            f"Output directory: {run_output_dir_val}\n",
            f"Command: {' '.join(command)}\n\n",
        ]

        yield create_empty_error_outputs("".join(log_output))

        log_file = os.path.join(run_output_dir_val, "autoforge_live.log")
        open(log_file, "w", encoding="utf-8").close()

        import threading

        class Worker(threading.Thread):
            def __init__(self, cmd, log_path):
                super().__init__(daemon=True)
                self.cmd, self.log_path = cmd, log_path
                self.returncode = None
                self.exc = None

            def run(self):
                try:
                    self.returncode = run_autoforge_process(self.cmd, self.log_path)
                except Exception as e:
                    self.exc = e
                    with open(self.log_path, "a", encoding="utf-8") as lf:
                        lf.write(
                            "\nERROR: {}. This usually means there was no GPU or the process took too long.\n".format(
                                exc_text(e)
                            )
                        )
                    self.returncode = -1

        try:
            worker = Worker(command, log_file)
            worker.start()

            preview_mtime = 0
            last_push = 0
            file_pos = 0

            def _maybe_new_preview():
                nonlocal preview_mtime
                src = os.path.join(run_output_dir_val, "vis_temp.png")
                if not os.path.exists(src):
                    return gr.update()
                mtime = os.path.getmtime(src)
                if mtime <= preview_mtime:
                    return gr.update()
                preview_mtime = mtime
                return src

            while worker.is_alive() or file_pos < os.path.getsize(log_file):
                with open(log_file, "r", encoding="utf-8") as lf:
                    lf.seek(file_pos)
                    new_txt = lf.read()
                    file_pos = lf.tell()
                    log_output.append(new_txt)

                now = time.time()
                if now - last_push >= 1.0:
                    current_preview = _maybe_new_preview()
                    yield (
                        "".join(log_output),
                        current_preview,
                        gr.update(),
                    )
                    last_push = now

                time.sleep(0.05)

            worker.join()
        except RuntimeError as e:
            log_output.append(repr(e))
            gr.Error(str(e))
            with open(log_file, "r", encoding="utf-8") as lf:
                lf.seek(file_pos)
                new_txt = lf.read()
                file_pos = lf.tell()
                log_output.append(new_txt)
            yield (
                "".join(log_output),
                gr.update(),
                gr.update(),
            )
            return create_empty_error_outputs(str(e))

        if getattr(worker, "exc", None) is not None:
            err_msg = f"GPU run failed: {worker.exc}"
            log_output.append(f"\n{err_msg}\n")
            gr.Error(err_msg)
            yield (
                "".join(log_output),
                gr.update(),
                gr.update(),
            )
            return

        with open(log_file, "r", encoding="utf-8") as lf:
            lf.seek(file_pos)
            log_output.append(lf.read())

        return_code = worker.returncode

        # Gather outputs, including all STL files (FlatForge may create multiple)
        static_files = [
            os.path.join(run_output_dir_val, "final_model.png"),
            os.path.join(run_output_dir_val, "swap_instructions.txt"),
            os.path.join(run_output_dir_val, "project_file.hfp"),
        ]
        stl_files = []
        for root, _, files in os.walk(run_output_dir_val):
            for fname in files:
                if fname.lower().endswith(".stl"):
                    stl_files.append(os.path.join(root, fname))

        files_to_offer = [p for p in static_files if os.path.exists(p)] + stl_files
        # Deduplicate while preserving order
        seen = set()
        files_to_offer = [f for f in files_to_offer if not (f in seen or seen.add(f))]
        png_path = os.path.join(run_output_dir_val, "final_model.png")
        out_png = png_path if os.path.exists(png_path) else None

        if return_code != 0:
            err_msg = (
                f"Autoforge exited with code {return_code}\n"
                "See the console output above for details."
            )
            log_output.append(f"\n{err_msg}\n")
            gr.Error(err_msg)
            yield (
                "".join(log_output),
                out_png if out_png else gr.update(),
                gr.update(),
            )
            return

        log_output.append("\nAutoforge process completed successfully!")

        zip_path = None
        if files_to_offer:
            zip_path = os.path.join(run_output_dir_val, "autoforge_results.zip")
            log_output.append(f"\nZipping results to {os.path.basename(zip_path)}...")
            try:
                with zipfile.ZipFile(zip_path, "w", compression=zipfile.ZIP_STORED) as zf:
                    for f in files_to_offer:
                        # Preserve relative paths to avoid collisions when multiple STLs exist
                        arcname = os.path.relpath(f, run_output_dir_val)
                        zf.write(f, arcname)
                log_output.append(" done.")
            except Exception as e:
                log_output.append(f"\nError creating zip file: {e}")
                zip_path = None

        if USE_WANDB:
            run = None
            try:
                run = wandb.init(
                    project="autoforge",
                    name=f"run_{timestamp}",
                    notes="Autoforge Web UI run",
                    tags=["autoforge", "gradio"],
                )
                wlogs = {"input_image": wandb.Image(script_input_image_path)}
                if out_png:
                    wlogs["output_image"] = wandb.Image(out_png)
                material_csv = pd.read_csv(temp_filament_csv)
                table = wandb.Table(dataframe=material_csv)
                wlogs["materials"] = table
                from wandb import Html
                log_text = "".join(log_output).replace("\r", "\n")

                def clean_log_strict(text: str) -> str:
                    allowed = set(string.printable) | {"\n", "\t"}
                    return "".join(ch for ch in text if ch in allowed)

                log_text_cleaned = clean_log_strict(log_text)
                wlogs["log"] = Html(f"<pre>{log_text_cleaned}</pre>")

                wandb.log(wlogs)
            except Exception as e:
                print(e)
            finally:
                if run is not None:
                    run.finish()

        yield (
            "".join(log_output),
            out_png,
            gr.update(
                value=zip_path,
                visible=bool(zip_path),
                interactive=bool(zip_path),
            ),
        )

    run_inputs = [filament_df_state, input_image_component] + [
        accordion_params_dict[name] for name in accordion_params_ordered_names
    ]
    run_outputs = [
        progress_output,
        final_image_preview,
        download_results,
    ]

    run_button.click(execute_autoforge_script, inputs=run_inputs, outputs=run_outputs)

css = """ #run_button_full_width { width: 100%; } """

if __name__ == "__main__":
    if not os.path.exists(DEFAULT_MATERIALS_CSV):
        print(f"Creating default filament file: {DEFAULT_MATERIALS_CSV}")
        try:
            initial_df.to_csv(DEFAULT_MATERIALS_CSV, index=False)
        except Exception as e:
            print(f"Could not write default {DEFAULT_MATERIALS_CSV}: {e}")
    print("To run the UI, execute: python app.py")
    demo.queue(default_concurrency_limit=1).launch(share=False)