app.py

#!/usr/bin/env python3
"""MiniMax All-in-One GUI — wraps mmx CLI with Gradio web interface."""

import os
os.environ["HF_HUB_OFFLINE"] = "1"
os.environ["GRADIO_ANALYTICS_ENABLED"] = "false"

import subprocess
import tempfile
import os
import gradio as gr

OUT_DIR = os.path.expanduser("~/Desktop/mmx_outputs")
os.makedirs(OUT_DIR, exist_ok=True)


def run_mmx(args, timeout=120):
    """Run mmx CLI and return parsed result."""
    cmd = ["mmx"] + args + ["--output", "json"]
    try:
        r = subprocess.run(cmd, capture_output=True, text=True, timeout=timeout)
        stdout = r.stdout.strip()
        stderr = r.stderr.strip()
        if r.returncode != 0:
            return None, stderr or stdout or "Unknown error"
        return stdout, None
    except subprocess.TimeoutExpired:
        return None, "请求超时"
    except Exception as e:
        return None, str(e)


# ─── Tab 1: Text Chat ────────────────────────────────────────────

def text_chat(message, history):
    history = history or []
    history.append({"role": "user", "content": message})

    # Build multi-message input
    args = ["text", "chat"]
    for h in history[-10:]:  # last 10 turns
        args += ["--message", f"{h['role']}:{h['content']}"]

    output, err = run_mmx(args, timeout=60)
    if err:
        history.append({"role": "assistant", "content": f"❌ {err}"})
    else:
        try:
            import json
            data = json.loads(output)
            reply = data.get("content", output)
        except:
            reply = output
        history.append({"role": "assistant", "content": reply})
    return "", history


# ─── Tab 2: Image Generation ─────────────────────────────────────

def image_generate(prompt, num_images):
    args = ["image", prompt, "--n", str(num_images)]
    out_path = os.path.join(OUT_DIR, "image")
    args += ["--out", out_path]

    output, err = run_mmx(args, timeout=90)
    if err:
        return [None] * num_images, f"❌ {err}"

    # Find generated files
    import glob
    files = sorted(glob.glob(out_path + "*"))
    if not files:
        files = sorted(glob.glob(out_path + "_*"))
    if not files:
        # Try current dir
        files = sorted(glob.glob("mmx_*.jpg") + glob.glob("mmx_*.png"))

    images = []
    for i in range(num_images):
        if i < len(files):
            images.append(files[i])
        else:
            images.append(None)

    return images, f"✅ 已保存到 {OUT_DIR}/"


# ─── Tab 3: Image Understanding ──────────────────────────────────

def vision_describe(image):
    if image is None:
        return "请上传图片"
    args = ["vision", image]
    output, err = run_mmx(args, timeout=30)
    if err:
        return f"❌ {err}"
    try:
        import json
        data = json.loads(output)
        return data.get("content", output)
    except:
        return output


# ─── Tab 4: Speech Synthesis ─────────────────────────────────────

def speech_synthesize(text, voice, speed):
    if not text.strip():
        return None, "请输入文本"
    out_path = os.path.join(OUT_DIR, "speech.mp3")
    args = ["speech", "synthesize", "--text", text, "--voice", voice,
            "--speed", str(speed), "--out", out_path]
    output, err = run_mmx(args, timeout=60)
    if err:
        return None, f"❌ {err}"
    return out_path, f"✅ 已保存 {out_path}"


# ─── Tab 5: Music Generation ─────────────────────────────────────

def music_generate(prompt, lyrics, vocals, instrumental, lyrics_optimizer):
    out_path = os.path.join(OUT_DIR, "music.mp3")
    args = ["music", "generate", "--prompt", prompt, "--out", out_path]

    if instrumental:
        args.append("--instrumental")
    elif lyrics_optimizer:
        args.append("--lyrics-optimizer")
    elif lyrics.strip():
        args += ["--lyrics", lyrics]

    if vocals.strip():
        args += ["--vocals", vocals]

    output, err = run_mmx(args, timeout=120)
    if err:
        return None, f"❌ {err}"
    return out_path, f"✅ 已保存 {out_path}"


# ─── Tab 6: Music Cover ──────────────────────────────────────────

def music_cover(audio_file, style_prompt):
    if audio_file is None:
        return None, "请上传参考音频"
    out_path = os.path.join(OUT_DIR, "cover.mp3")
    args = ["music", "cover", "--prompt", style_prompt,
            "--audio-file", audio_file, "--out", out_path]
    output, err = run_mmx(args, timeout=180)
    if err:
        return None, f"❌ {err}"
    return out_path, f"✅ 已保存 {out_path}"


# ─── Build UI ────────────────────────────────────────────────────

VOICES = [
    "male-qn-qingse", "male-qn-jingying", "male-qn-badao", "male-qn-daxuesheng",
    "female-shaonv", "female-yujie", "female-chengshu", "female-tianmei",
    "clever_boy", "cute_boy", "lovely_girl", "junlang_nanyou",
    "tianxin_xiaoling", "qiaopi_mengmei", "wumei_yujie",
]

with gr.Blocks(title="MiniMax Studio") as app:
    gr.Markdown("# 🎨 MiniMax Studio")
    gr.Markdown("AI 文本 · 图片 · 语音 · 音乐 — 一站式创作")

    with gr.Tabs():
        # ── Text ──
        with gr.Tab("💬 文本对话"):
            gr.Markdown("M2.7 多轮对话（每次独立请求）")
            txt_msg = gr.Textbox(placeholder="输入消息...", label="你的消息", lines=2)
            txt_system = gr.Textbox(label="系统提示词（可选）", 
                                    placeholder="你是一个有帮助的助手", lines=1)
            txt_btn = gr.Button("发送", variant="primary")
            txt_output = gr.Textbox(label="回复", lines=20)
            
            def text_single(message, system_prompt):
                args = ["text", "chat", "--message", message]
                if system_prompt.strip():
                    args += ["--system", system_prompt]
                output, err = run_mmx(args, timeout=60)
                if err:
                    return f"❌ {err}"
                try:
                    import json
                    data = json.loads(output)
                    # Handle array response (thinking + text blocks)
                    if isinstance(data, list):
                        texts = []
                        for block in data:
                            if isinstance(block, dict) and block.get("type") == "text":
                                texts.append(block.get("text", ""))
                        return "\n".join(texts) if texts else output
                    # Handle object response
                    if isinstance(data, dict):
                        return data.get("content", output)
                    return output
                except:
                    return output
            
            txt_btn.click(text_single, [txt_msg, txt_system], [txt_output])
            txt_msg.submit(text_single, [txt_msg, txt_system], [txt_output])

        # ── Image Gen ──
        with gr.Tab("🖼️ 图片生成"):
            with gr.Row():
                img_prompt = gr.Textbox(label="描述", placeholder="一只猫在太空站里喝咖啡")
                img_num = gr.Slider(1, 4, value=1, step=1, label="数量")
            img_btn = gr.Button("生成", variant="primary")
            img_gallery = gr.Gallery(label="生成结果", columns=2, height=400)
            img_msg = gr.Markdown("")
            img_btn.click(image_generate, [img_prompt, img_num], [img_gallery, img_msg])

        # ── Vision ──
        with gr.Tab("👁️ 图片理解"):
            with gr.Row():
                vis_image = gr.Image(type="filepath", label="上传图片")
                vis_output = gr.Textbox(label="分析结果", lines=15)
            vis_btn = gr.Button("分析", variant="primary")
            vis_btn.click(vision_describe, [vis_image], [vis_output])

        # ── Speech ──
        with gr.Tab("🔊 语音合成"):
            sp_text = gr.Textbox(label="文本", placeholder="输入要朗读的文字...", lines=3)
            with gr.Row():
                sp_voice = gr.Dropdown(VOICES, value="female-yujie", label="音色")
                sp_speed = gr.Slider(0.5, 2.0, value=1.0, label="语速")
            sp_btn = gr.Button("合成", variant="primary")
            sp_audio = gr.Audio(label="播放")
            sp_msg = gr.Markdown("")
            sp_btn.click(speech_synthesize, [sp_text, sp_voice, sp_speed], [sp_audio, sp_msg])

        # ── Music ──
        with gr.Tab("🎵 音乐生成"):
            mu_prompt = gr.Textbox(label="风格描述", placeholder="轻快流行，温暖人声",
                                   value="轻快的电子流行音乐")
            mu_lyrics = gr.Textbox(label="歌词（留空则自动生成）", lines=5,
                                   placeholder="[verse] 歌词...\n[chorus] 副歌...")
            mu_vocals = gr.Textbox(label="人声描述（可选）",
                                   placeholder="温暖男声 / 明亮女高音 / 男女对唱")
            with gr.Row():
                mu_instrumental = gr.Checkbox(label="纯音乐（无歌词）")
                mu_lyrics_opt = gr.Checkbox(label="AI 自动写歌词")
            mu_btn = gr.Button("生成音乐", variant="primary")
            mu_audio = gr.Audio(label="播放")
            mu_msg = gr.Markdown("")
            mu_btn.click(music_generate, [mu_prompt, mu_lyrics, mu_vocals,
                                          mu_instrumental, mu_lyrics_opt],
                         [mu_audio, mu_msg])

        # ── Cover ──
        with gr.Tab("🎤 翻唱"):
            cv_audio = gr.Audio(type="filepath", label="上传原曲")
            cv_style = gr.Textbox(label="目标风格", placeholder="摇滚 / 爵士 / 民谣 / ...",
                                  value="硬摇滚，电吉他，强力鼓点")
            cv_btn = gr.Button("翻唱", variant="primary")
            cv_output = gr.Audio(label="翻唱结果")
            cv_msg = gr.Markdown("")
            cv_btn.click(music_cover, [cv_audio, cv_style], [cv_output, cv_msg])

if __name__ == "__main__":
    app.launch(server_port=7860, share=False, inbrowser=True)