from __future__ import annotations

import json
import os
import tempfile
from pathlib import Path

import gradio as gr

from alpha_core import extract_alpha_video


def run(video, mode, max_frames):
    if not video: raise gr.Error("Upload a video first.")
    work = Path(tempfile.mkdtemp(prefix="simam_alpha_xr_"))
    webm, preview = work / "simam_alpha_xr.webm", work / "checkerboard_preview.mp4"
    try:
        frame, alpha, info = extract_alpha_video(video, str(webm), str(preview), mode, int(max_frames))
    except Exception as exc: raise gr.Error(str(exc))
    return str(webm), str(preview), frame, alpha, json.dumps(info, indent=2)


with gr.Blocks(title="SimamAlphaXR") as demo:
    gr.Markdown("# SimamAlphaXR\n### Transparent video layers for immersive XR\n\nCreate a foreground-only video that can float inside a virtual environment. This release provides a real VP9 alpha-channel compositor and makes its limitations visible.")
    with gr.Row():
        with gr.Column():
            video = gr.Video(label="Input video", sources=["upload", "webcam"], format="mp4")
            mode = gr.Radio(["Green screen", "Blue screen", "Background contrast baseline"], value="Green screen", label="Matte mode")
            max_frames = gr.Slider(8, 240, value=96, step=8, label="Maximum processed frames")
            button = gr.Button("Extract transparent layer", variant="primary")
            gr.Examples([["examples/sample_subject.mp4"]], inputs=[video], label="Try the included synthetic fixture")
        with gr.Column():
            transparent = gr.File(label="Transparent WebM (VP9 alpha)")
            preview = gr.Video(label="Checkerboard preview")
            frame = gr.Image(label="First composited frame")
            alpha = gr.Image(label="Alpha matte", image_mode="L")
            diagnostics = gr.Code(label="Diagnostics", language="json")
    button.click(run, [video, mode, max_frames], [transparent, preview, frame, alpha, diagnostics], api_name="extract_alpha")
    gr.Markdown("**Roadmap:** Wan-Alpha adapter → temporally stable learned mattes → diffusion-generated transparent clips → WebRTC/WebXR playback. Generated alpha video should be evaluated separately from chroma-key compositing.")


if __name__ == "__main__":
    demo.launch(server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"), server_port=int(os.getenv("GRADIO_SERVER_PORT", "7860")))