mcpField
M
meshy-3d-agent

meshy-3d-agent

github.com/meshy-dev创意与媒体106908 次安装2 周前更新
代码仓库

关于 meshy-3d-agent

meshy-3d-agent 是使用 Meshy AI API 生成、处理并准备打印 3D 资产的技能。它指导创建任务、轮询状态、下载文件及执行多步骤模型工作流,并要求先确认费用且遵循文件管理规则。

主要功能

  1. 01文本和图像生成 3D 模型
  2. 02重新贴图、重构网格和格式转换
  3. 03模型缩放与 UV 展开
  4. 04骨骼绑定和动画任务指导
  5. 05可打印性分析与修复
  6. 06本地文件分类下载及任务历史

适用场景

  1. 01根据提示词或图像生成 3D 模型
  2. 02将模型转换为其他格式
  3. 03为角色绑定骨骼并制作动画
  4. 04检查或修复模型的可打印性

安装

复制下面的命令,在终端中运行。

$npx skills add https://github.com/meshy-dev/meshy-3d-agent --skill meshy-3d-agent

安装后在 Claude Code 中输入/meshy-3d-agent确认已加载。

附带文件reference.md

详细介绍

Meshy 3D — 生成与打印

直接与 Meshy AI API 通信,以生成和打印 3D 资产。涵盖完整生命周期:API 密钥设置、创建任务、指数退避轮询、下载、多步骤流水线,以及通过切片软件集成来准备 3D 打印。


安全清单

访问的环境变量:

  • MESHY_API_KEY — API 身份验证令牌,仅通过 HTTP Authorization: Bearer 标头发送。绝不记录,也不会写入任何文件;只有在用户明确请求时,才会写入当前工作目录中的 .env。

外部网络端点:

  • https://api.meshy.ai — Meshy AI API(创建任务、轮询状态、下载模型和图像)

文件系统访问:

  • 读取:仅当前工作目录中的 .env(用于查找 API 密钥)
  • 写入:仅当前工作目录中的 .env(用于存储 API 密钥,且仅在用户请求时)
  • 写入:当前工作目录中的 ./meshy_output/(下载的模型文件、元数据)
  • 读取:用户明确提供的文件(例如,用于图像转 3D 转换的本地图像);仅在用户指定的确切路径访问
  • 无权访问主目录、Shell 配置文件或上述范围以外的任何路径

离开此设备的数据:

  • 向 api.meshy.ai 发送的 API 请求会在 Authorization 标头中包含 MESHY_API_KEY,以及用户提供的文本提示词或图像 URL。不会传输其他本地数据。下载的模型文件仅保存在本地。

重要:首次使用会话提示

此技能在会话中首次激活时,告知用户:

所有生成的文件都会保存在当前工作目录中的 meshy_output/。每个项目都有自己的文件夹({YYYYMMDD_HHmmss}_{prompt}_{id}/),其中包含模型文件、纹理、缩略图和元数据。历史记录保存在 meshy_output/history.json 中。

每个会话只需说明一次。


重要:文件组织

所有下载的文件都必须放入当前工作目录中结构化的 meshy_output/ 目录。不要将文件随意散落在各处。

  • 每个项目:meshy_output/{YYYYMMDD_HHmmss}_{prompt_slug}_{task_id_prefix}/
  • 链式任务(预览 → 精修 → 绑定)复用同一个 project_dir
  • 在每个项目的 metadata.json 和全局 history.json 中跟踪任务
  • 下载模型时同时自动下载缩略图

重要:Shell 命令规则

仅使用标准 POSIX 工具。不要使用 rg、fd、bat、exa/eza。


重要:正确运行长时间任务

Meshy 生成需要 1–5 分钟。将完整的创建 → 轮询 → 下载流程写入一个 Python 脚本,并通过一次 Bash 调用执行。使用 python3 -u script.py 以禁用输出缓冲。在 99% 停留 30–120 秒属于正常的最终处理过程——不要中断。


步骤 0:检测 API 密钥(始终首先运行)

仅检查当前会话环境和当前工作目录中的 .env 文件。不要扫描主目录或 Shell 配置文件。

echo "=== Meshy API Key Detection ==="

# 1. Check current env var
if [ -n "$MESHY_API_KEY" ]; then
  echo "ENV_VAR: FOUND (${MESHY_API_KEY:0:8}...)"
else
  echo "ENV_VAR: NOT_FOUND"
fi

# 2. Check .env in current working directory only
if [ -f ".env" ] && grep -q "MESHY_API_KEY" ".env" 2>/dev/null; then
  echo "DOTENV(.env): FOUND"
  export MESHY_API_KEY=$(grep "^MESHY_API_KEY=" ".env" | head -1 | cut -d'=' -f2- | tr -d '"'"'" )
fi

# 3. Final status
if [ -n "$MESHY_API_KEY" ]; then
  echo "READY: key=${MESHY_API_KEY:0:8}..."
else
  echo "READY: NO_KEY_FOUND"
fi

# 4. Python requests check
python3 -c "import requests; print('PYTHON_REQUESTS: OK')" 2>/dev/null || echo "PYTHON_REQUESTS: MISSING (run: pip install requests)"

echo "=== Detection Complete ==="
检测后的处理
  • 找到密钥 → 继续步骤 1。
  • 未找到密钥 → 前往步骤 0a。
  • 缺少 Python requests → 运行 pip install requests。

步骤 0a:设置 API 密钥(仅在未找到密钥时)

告知用户:

要使用 Meshy API,你需要一个 API 密钥:

  1. 前往 https://www.meshy.ai/settings/api
  2. 点击 "Create API Key",为密钥命名并复制密钥(以 msy_ 开头)
  3. 密钥只会显示一次——请妥善保存

注意: API 访问要求使用 Pro 计划或更高等级。免费账户无法创建 API 密钥。

用户提供密钥后,将其设置到当前会话中,并可选择持久化到 .env:

# Set for current session only
export MESHY_API_KEY="msy_PASTE_KEY_HERE"

# Verify the key
STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
  -H "Authorization: Bearer $MESHY_API_KEY" \
  https://api.meshy.ai/openapi/v1/balance)

if [ "$STATUS" = "200" ]; then
  BALANCE=$(curl -s -H "Authorization: Bearer $MESHY_API_KEY" https://api.meshy.ai/openapi/v1/balance)
  echo "Key valid. $BALANCE"
else
  echo "Key invalid (HTTP $STATUS). Please check the key and try again."
fi

要持久化密钥(仅限当前项目):

# Write to .env in current working directory
echo 'MESHY_API_KEY=msy_PASTE_KEY_HERE' >> .env
echo "Saved to .env"

# IMPORTANT: add .env to .gitignore to avoid leaking the key
grep -q "^\.env" .gitignore 2>/dev/null || echo ".env" >> .gitignore
echo ".env added to .gitignore"

安全提醒: 密钥仅存储在当前项目目录中的 .env 文件里。切勿将此文件提交到版本控制中。.env 已自动添加到 .gitignore。


步骤 1:先与用户确认计划,再消耗点数

重要:创建任何任务之前,先向用户展示费用摘要并等待确认:

I'll generate a 3D model of "<prompt>" using the following plan:

  1. Preview (mesh generation) — 20 credits
  2. Refine (texturing with PBR) — 10 credits
  3. Download as .glb

  Total cost: 30 credits
  Current balance: <N> credits

  Shall I proceed?

对于多步骤流水线(文本转 3D → 绑定骨骼 → 动画),应提前展示完整流水线的费用。

注意: 自动绑定骨骼会包含行走和跑步动画,不额外收费。只有在需要这些动画之外的自定义动画时,才添加 Animate(3 点)。

意图 → API 映射
用户想要……API端点点数
从文本生成 3D 模型Text to 3DPOST /openapi/v2/text-to-3d5–20(预览)+ 10(精修)
从单张图片生成 3D 模型Image to 3DPOST /openapi/v1/image-to-3d5–30
从多张图片生成 3D 模型Multi-Image to 3DPOST /openapi/v1/multi-image-to-3d5–30
为现有模型添加新纹理RetexturePOST /openapi/v1/retexture10
更改网格格式/拓扑RemeshPOST /openapi/v1/remesh5
将模型转换为其他格式(不重新网格化)ConvertPOST /openapi/v1/convert1
将模型缩放至真实尺寸ResizePOST /openapi/v1/resize1
在外部纹理处理前生成全新的 UV(GLB,≤40k 个面)UV UnwrapPOST /openapi/v1/uv-unwrap5
为角色添加骨架Auto-RiggingPOST /openapi/v1/rigging5
为已绑定骨骼的角色添加动画AnimationPOST /openapi/v1/animations3
根据文本生成 2D 图像(建议作为图像转 3D 的前置步骤)Text to ImagePOST /openapi/v1/text-to-image3 / 6 / 9 / 9
优化/编辑 2D 图像(建议作为图像转 3D 的前置步骤)Image to ImagePOST /openapi/v1/image-to-image3 / 6 / 9 / 12
照片 → 风格化实体产品(手办/台灯/钥匙扣/冰箱贴)Creative LabPOST /openapi/creative-lab/{product}/v1/prototype 然后 .../build6 + 30
检查 FDM 打印适用性Analyze PrintabilityPOST /openapi/v1/print/analyze0(免费)
修复非流形/退化面/孔洞拓扑Repair PrintabilityPOST /openapi/v1/print/repair10
多色 3D 打印Multi-Color PrintPOST /openapi/v1/print/multi-color10(+ 生成费用)
3D 打印模型(白色)→ 请参阅打印流水线部分—20
查看点数余额BalanceGET /openapi/v1/balance0

步骤 2:执行工作流

可复用的脚本模板

将此模板用作所有工作流的基础。它仅从环境变量或当前目录中的 .env 安全读取 API 密钥:

#!/usr/bin/env python3
"""Meshy API task runner. Handles create → poll → download."""
import requests, time, os, sys, re, json
from datetime import datetime

# --- Secure API key loading ---
def load_api_key():
    """Load MESHY_API_KEY from environment, then .env in cwd only."""
    key = os.environ.get("MESHY_API_KEY", "").strip()
    if key:
        return key
    env_path = os.path.join(os.getcwd(), ".env")
    if os.path.exists(env_path):
        with open(env_path) as f:
            for line in f:
                line = line.strip()
                if line.startswith("MESHY_API_KEY=") and not line.startswith("#"):
                    val = line.split("=", 1)[1].strip().strip('"').strip("'")
                    if val:
                        return val
    return ""

API_KEY = load_api_key()
if not API_KEY:
    sys.exit("ERROR: MESHY_API_KEY not set. Run Step 0a to configure it.")

# Never log the full key — only first 8 chars for traceability
print(f"API key loaded: {API_KEY[:8]}...")

BASE = "https://api.meshy.ai"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}
SESSION = requests.Session()
SESSION.trust_env = False  # bypass any system proxy settings

def create_task(endpoint, payload):
    resp = SESSION.post(f"{BASE}{endpoint}", headers=HEADERS, json=payload, timeout=30)
    if resp.status_code == 401:
        sys.exit("ERROR: Invalid API key (401). Re-run Step 0a.")
    if resp.status_code == 402:
        try:
            bal = SESSION.get(f"{BASE}/openapi/v1/balance", headers=HEADERS, timeout=10)
            balance = bal.json().get("balance", "unknown")
            sys.exit(f"ERROR: Insufficient credits (402). Balance: {balance}. Top up at https://www.meshy.ai/pricing")
        except Exception:
            sys.exit("ERROR: Insufficient credits (402). Check balance at https://www.meshy.ai/pricing")
    if resp.status_code == 429:
        sys.exit("ERROR: Rate limited (429). Wait and retry.")
    resp.raise_for_status()
    task_id = resp.json()["result"]
    print(f"TASK_CREATED: {task_id}")
    return task_id

def poll_task(endpoint, task_id, timeout=300):
    """Poll with exponential backoff (5s→30s, fixed 15s at 95%+)."""
    elapsed, delay, max_delay, backoff, finalize_delay, poll_count = 0, 5, 30, 1.5, 15, 0
    while elapsed < timeout:
        poll_count += 1
        resp = SESSION.get(f"{BASE}{endpoint}/{task_id}", headers=HEADERS, timeout=30)
        resp.raise_for_status()
        task = resp.json()
        status = task["status"]
        progress = task.get("progress", 0)
        bar = f"[{'█' * int(progress/5)}{'░' * (20 - int(progress/5))}] {progress}%"
        print(f"  {bar} — {status} ({elapsed}s, poll #{poll_count})", flush=True)
        if status == "SUCCEEDED":
            return task
        if status in ("FAILED", "CANCELED"):
            msg = task.get("task_error", {}).get("message", "Unknown")
            sys.exit(f"TASK_{status}: {msg}")
        current_delay = finalize_delay if progress >= 95 else delay
        time.sleep(current_delay)
        elapsed += current_delay
        if progress < 95:
            delay = min(delay * backoff, max_delay)
    sys.exit(f"TIMEOUT after {timeout}s ({poll_count} polls)")

def download(url, filepath):
    """Download a file into a project directory (within cwd/meshy_output/)."""
    os.makedirs(os.path.dirname(filepath), exist_ok=True)
    print(f"Downloading {filepath}...", flush=True)
    resp = SESSION.get(url, timeout=300, stream=True)
    resp.raise_for_status()
    with open(filepath, "wb") as f:
        for chunk in resp.iter_content(chunk_size=8192):
            f.write(chunk)
    print(f"DOWNLOADED: {filepath} ({os.path.getsize(filepath)/1024/1024:.1f} MB)")

# --- File organization helpers ---
OUTPUT_ROOT = os.path.join(os.getcwd(), "meshy_output")
os.makedirs(OUTPUT_ROOT, exist_ok=True)
HISTORY_FILE = os.path.join(OUTPUT_ROOT, "history.json")

def get_project_dir(task_id, prompt="", task_type="model"):
    slug = re.sub(r'[^a-z0-9]+', '-', (prompt or task_type).lower())[:30].strip('-')
    folder = f"{datetime.now().strftime('%Y%m%d_%H%M%S')}_{slug}_{task_id[:8]}"
    project_dir = os.path.join(OUTPUT_ROOT, folder)
    os.makedirs(project_dir, exist_ok=True)
    return project_dir

def record_task(project_dir, task_id, task_type, stage, prompt="", files=None):
    meta_path = os.path.join(project_dir, "metadata.json")
    meta = json.load(open(meta_path)) if os.path.exists(meta_path) else {
        "project_name": prompt or task_type, "folder": os.path.basename(project_dir),
        "root_task_id": task_id, "created_at": datetime.now().isoformat(), "tasks": []
    }
    meta["tasks"].append({"task_id": task_id, "task_type": task_type, "stage": stage,
                          "files": files or [], "created_at": datetime.now().isoformat()})
    meta["updated_at"] = datetime.now().isoformat()
    json.dump(meta, open(meta_path, "w"), indent=2)
    history = json.load(open(HISTORY_FILE)) if os.path.exists(HISTORY_FILE) else {"version": 1, "projects": []}
    folder = os.path.basename(project_dir)
    entry = next((p for p in history["projects"] if p["folder"] == folder), None)
    if entry:
        entry.update({"task_count": len(meta["tasks"]), "updated_at": meta["updated_at"]})
    else:
        history["projects"].append({"folder": folder, "prompt": prompt, "task_type": task_type,
            "root_task_id": task_id, "created_at": meta["created_at"],
            "updated_at": meta["updated_at"], "task_count": len(meta["tasks"])})
    json.dump(history, open(HISTORY_FILE, "w"), indent=2)

def save_thumbnail(project_dir, url):
    path = os.path.join(project_dir, "thumbnail.png")
    if os.path.exists(path): return
    try:
        r = SESSION.get(url, timeout=15); r.raise_for_status()
        open(path, "wb").write(r.content)
    except Exception: pass

文本转 3D(预览 + 精修)

将以下内容附加到上方模板:

PROMPT = "USER_PROMPT"

# Preview
preview_id = create_task("/openapi/v2/text-to-3d", {
    "mode": "preview",
    "prompt": PROMPT,
    "ai_model": "latest",
    # "pose_mode": "t-pose",   # Use "t-pose" if rigging/animating later
    # "hd_texture": True,      # 4K base color on refine (meshy-6/latest only)
    # "target_formats": ["glb", "3mf"],  # 3mf must be explicitly requested
    # NOTE: symmetry_mode / art_style / is_a_t_pose are deprecated (symmetry_mode & art_style ignored; use pose_mode)
})
task = poll_task("/openapi/v2/text-to-3d", preview_id)
project_dir = get_project_dir(preview_id, prompt=PROMPT)
download(task["model_urls"]["glb"], os.path.join(project_dir, "preview.glb"))
record_task(project_dir, preview_id, "text-to-3d", "preview", prompt=PROMPT, files=["preview.glb"])
if task.get("thumbnail_url"):
    save_thumbnail(project_dir, task["thumbnail_url"])
print(f"\nPREVIEW COMPLETE — Task: {preview_id} | Project: {project_dir}")

# Refine
refine_id = create_task("/openapi/v2/text-to-3d", {
    "mode": "refine",
    "preview_task_id": preview_id,
    "enable_pbr": True,
    "ai_model": "latest",
})
task = poll_task("/openapi/v2/text-to-3d", refine_id)
download(task["model_urls"]["glb"], os.path.join(project_dir, "refined.glb"))
record_task(project_dir, refine_id, "text-to-3d", "refined", prompt=PROMPT, files=["refined.glb"])
print(f"\nREFINE COMPLETE — Task: {refine_id} | Formats: {', '.join(task['model_urls'].keys())}")

精修兼容性: 精修支持 meshy-5、meshy-6 或 latest(即 Meshy 6)——为保持一致,请选择与预览相同的系列。无论使用哪种模型,精修均需 10 点。(meshy-4 已停用,调用会返回 400。)


###(可选但强烈建议)2D 优化预处理

优先选择图像转 3D,而非直接文本转 3D——这种方式质量更高,也更可控。因此,对于纯文本请求,请先制作设计图,再转换为 3D。

图像质量直接决定 3D 模型质量。在调用 /openapi/v1/image-to-3d 或 /openapi/v1/multi-image-to-3d 之前,请先评估用户输入,并主动建议进行 2D 处理:

用户输入推荐的预处理步骤
只有文字描述,没有参考图使用 nano-banana-pro 调用 /openapi/v1/text-to-image。对于角色,添加 generate_multi_view: True 和 pose_mode: "a-pose" 或 "t-pose",以生成便于绑定骨骼的输出。
参考图分辨率低 / 背景杂乱 / 主体不清晰 / 光线不佳使用 nano-banana-pro 调用 /openapi/v1/image-to-image 进行清理。
用户想调整风格 / 颜色 / 细节调用 /openapi/v1/image-to-image 进行风格迁移,然后转换为 3D。

额外花费 3-9 个点数通常能显著提升质量。如果用户已经提供干净的影棚风格图片,则跳过此步骤。对于 Creative Lab 产品(摆件 / 台灯 / 钥匙扣 / 冰箱贴)也跳过此步骤:它们有自己的内置风格化处理——将原始照片(或台灯产品的文本)直接输入 Creative Lab,而不要先经过 text-to-image / image-to-image。

图像转 3D
import base64

# For local files: convert to data URI
# with open("photo.jpg", "rb") as f:
#     image_url = "data:image/jpeg;base64," + base64.b64encode(f.read()).decode()

task_id = create_task("/openapi/v1/image-to-3d", {
    "image_url": "IMAGE_URL_OR_DATA_URI",
    "should_texture": True,
    "enable_pbr": True,
    "ai_model": "latest",
})
task = poll_task("/openapi/v1/image-to-3d", task_id)
project_dir = get_project_dir(task_id, task_type="image-to-3d")
download(task["model_urls"]["glb"], os.path.join(project_dir, "model.glb"))
record_task(project_dir, task_id, "image-to-3d", "complete", files=["model.glb"])

多图转 3D
task_id = create_task("/openapi/v1/multi-image-to-3d", {
    "image_urls": ["URL_1", "URL_2", "URL_3"],  # 1–4 images
    "should_texture": True,
    "enable_pbr": True,
    "ai_model": "latest",
})
task = poll_task("/openapi/v1/multi-image-to-3d", task_id)
project_dir = get_project_dir(task_id, task_type="multi-image-to-3d")
download(task["model_urls"]["glb"], os.path.join(project_dir, "model.glb"))

重新贴图

重要:先询问用户的纹理风格——text_style_prompt 或 image_style_url(必须提供其中一个;如果两者都提供,则优先使用图像)。

# REQUIRED: ask user for text_style_prompt OR image_style_url
task_id = create_task("/openapi/v1/retexture", {
    "input_task_id": "PREVIOUS_TASK_ID",
    "text_style_prompt": "wooden texture",     # REQUIRED if no image_style_url
    # "image_style_url": "URL",               # REQUIRED if no prompt (takes precedence)
    "enable_pbr": True,
    # "target_formats": ["glb", "3mf"],  # 3mf must be explicitly requested
})
task = poll_task("/openapi/v1/retexture", task_id)
project_dir = get_project_dir(task_id, task_type="retexture")
download(task["model_urls"]["glb"], os.path.join(project_dir, "retextured.glb"))

网格重建 / 格式转换
task_id = create_task("/openapi/v1/remesh", {
    "input_task_id": "TASK_ID",
    "target_formats": ["glb", "fbx", "obj"],
    "topology": "quad",
    "target_polycount": 10000,
})
task = poll_task("/openapi/v1/remesh", task_id)
project_dir = get_project_dir(task_id, task_type="remesh")
for fmt, url in task["model_urls"].items():
    download(url, os.path.join(project_dir, f"remeshed.{fmt}"))

网格实用工具(转换 / 调整尺寸 / UV 展开)

对已完成的模型进行轻量级后处理(通过 input_task_id 或 model_url):

# Convert to other formats without remeshing (1 credit). Cheapest way to get 3MF/STL.
conv_id = create_task("/openapi/v1/convert", {
    "input_task_id": "TASK_ID",         # or "model_url": "URL"
    "target_formats": ["stl", "3mf"],   # required: glb/fbx/obj/usdz/blend/stl/3mf
})
poll_task("/openapi/v1/convert", conv_id)

# Resize to a real-world size (1 credit). Give EXACTLY ONE resize mode.
resize_id = create_task("/openapi/v1/resize", {
    "input_task_id": "TASK_ID",         # or "model_url": "URL"
    "resize_height": 0.15,              # meters — OR "resize_longest_side": 0.2  OR "auto_size": True
    # "origin_at": "bottom",            # "bottom" | "center"
})
poll_task("/openapi/v1/resize", resize_id)

# UV Unwrap a GLB (5 credits). GLB only, ≤ 40,000 faces (else 400 → remesh down first).
# Output: a GLB "UV white model" (fresh UVs + placeholder grey material) for external texturing.
uv_id = create_task("/openapi/v1/uv-unwrap", {
    "input_task_id": "TASK_ID",         # or "model_url": "GLB_URL"
})
poll_task("/openapi/v1/uv-unwrap", uv_id)

Creative Lab 消费级产品

将照片转换为风格化、可打印的实体产品。产品包括:摆件、台灯、钥匙扣、冰箱贴。分为两个阶段(将 {product} 替换为其中一种产品):

  1. 原型(6 个点数):照片 → 风格化概念图。
  2. 构建(30 个点数):对状态为 SUCCEEDED 的原型运行 image-to-3d → 带纹理的 GLB / OBJ+MTL。
PRODUCT = "figure"  # figure | lamp | keychain | fridge-magnet

# Stage 1 — prototype (6 credits)
proto_id = create_task(f"/openapi/creative-lab/{PRODUCT}/v1/prototype", {
    "image_url": "PHOTO_URL_OR_DATA_URI",   # jpg/jpeg/png/webp
    # "name": "My figure",                  # optional, ≤ 100 chars
})
poll_task(f"/openapi/creative-lab/{PRODUCT}/v1/prototype", proto_id)

# Stage 2 — build (30 credits). Must reference a SUCCEEDED prototype of the SAME product.
# Web-app prototypes are rejected with 404 — only API-created prototypes are accepted.
build_id = create_task(f"/openapi/creative-lab/{PRODUCT}/v1/build", {
    "input_task_id": proto_id,
})
build = poll_task(f"/openapi/creative-lab/{PRODUCT}/v1/build", build_id)
project_dir = get_project_dir(build_id, task_type=f"creative-lab-{PRODUCT}")
download(build["model_urls"]["glb"], os.path.join(project_dir, "creative-lab.glb"))

# To MULTICOLOR a Creative Lab result: a Creative Lab model can only be sent to multi-color
# as model_url — pass the build's GLB URL (or a data URI of the downloaded GLB):
#   create_task("/openapi/v1/print/multi-color", {"model_url": build["model_urls"]["glb"], "max_colors": 4})

自动绑定骨骼 + 动画

用户要求绑定骨骼或制作动画时,生成步骤必须使用 pose_mode: "t-pose"。

# Pre-rig check: polycount must be ≤ 300,000
source_endpoint = "/openapi/v2/text-to-3d"  # adjust to match source task endpoint
source_task_id = "TASK_ID"
check = SESSION.get(f"{BASE}{source_endpoint}/{source_task_id}", headers=HEADERS, timeout=30)
check.raise_for_status()
face_count = check.json().get("face_count", 0)
if face_count > 300000:
    sys.exit(f"ERROR: {face_count:,} faces exceeds 300,000 limit. Remesh first.")

# Rig
rig_id = create_task("/openapi/v1/rigging", {
    "input_task_id": source_task_id,
    "height_meters": 1.7,
})
rig_task = poll_task("/openapi/v1/rigging", rig_id)
project_dir = get_project_dir(rig_id, task_type="rigging")
download(rig_task["result"]["rigged_character_glb_url"], os.path.join(project_dir, "rigged.glb"))
download(rig_task["result"]["basic_animations"]["walking_glb_url"], os.path.join(project_dir, "walking.glb"))
download(rig_task["result"]["basic_animations"]["running_glb_url"], os.path.join(project_dir, "running.glb"))

# Custom animation (optional, 3 credits — only if user needs beyond walking/running)
# anim_id = create_task("/openapi/v1/animations", {"rig_task_id": rig_id, "action_id": 1})
# anim_task = poll_task("/openapi/v1/animations", anim_id)
# download(anim_task["result"]["animation_glb_url"], os.path.join(project_dir, "animated.glb"))

文本转图像 / 图像转图像
# Text to Image
task_id = create_task("/openapi/v1/text-to-image", {
    "ai_model": "nano-banana-pro",
    "prompt": "a futuristic spaceship",
})
task = poll_task("/openapi/v1/text-to-image", task_id)
# Result URL: task["image_url"]

# Image to Image
task_id = create_task("/openapi/v1/image-to-image", {
    "ai_model": "nano-banana-pro",
    "prompt": "make it look cyberpunk",
    "reference_image_urls": ["URL"],
})
task = poll_task("/openapi/v1/image-to-image", task_id)

3D 打印工作流

重要:当用户的请求涉及 3D 打印时,整个工作流(包括模型生成)都必须使用本节。不要先运行上述生成工作流,再切换到此处。本节从一开始就会控制 target_formats 和其他打印专用参数。

当用户提到以下内容时触发:打印、3D 打印、切片软件、切片、bambu、orca、prusa、cura、多色、multicolor、multi-color、3mf、手办、微缩模型、雕像、实体模型、桌面玩具、手机支架。

决策:白色模型还是多色模型
  1. 先检测已安装的切片软件(见下方脚本)
  2. 询问用户:“白色模型(单色)还是多色模型?”
  3. 如果选择多色模型:检查是否有支持多色打印的切片软件(OrcaSlicer、Bambu Studio、Creality Print、Elegoo Slicer、Anycubic Slicer Next),询问 max_colors(1-16,默认值为 4)和 max_depth(3-6,默认值为 4),并确认费用:40 个点数(如果需要修复,则额外加 10 个点数)
  4. (推荐)生成后运行可打印性分析(POST /openapi/v1/print/analyze,免费)。仅当状态为 error 时才运行**POST /openapi/v1/print/repair**(10 个点数)。
可打印性分析与修复
# After the textured/final mesh is ready:
INPUT_TASK_ID = refine_id  # or whatever produced the print-ready mesh

# input_task_id MUST refer to a Meshy 6 / Preview task. For Meshy 4/5 outputs,
# pass `model_url` (the GLB download URL) instead.
analyze_id = create_task("/openapi/v1/print/analyze", {
    "input_task_id": INPUT_TASK_ID,
})
analyze_task = poll_task("/openapi/v1/print/analyze", analyze_id)

p = analyze_task.get("printability") or {}
metrics = p.get("metrics", {})
print(f"Printability: {p.get('status')}: {metrics}")

if p.get("status") == "error":
    repair_id = create_task("/openapi/v1/print/repair", {
        "input_task_id": INPUT_TASK_ID,   # output is GLB
    })
    repair_task = poll_task("/openapi/v1/print/repair", repair_id)
    repaired_url = next((u for u in repair_task["model_urls"].values() if u), None)
    # Use repaired_url for the next step (download / multicolor / slicer)

状态含义:healthy(可直接打印)| warning(存在退化面 / 孔洞——可选择修复)| error(非水密 / 非流形——建议修复)| unknown(分析失败)。修复只会保留几何形状,不会保留纹理——如果需要多色打印,请重新贴图。

切片软件检测与打开
import subprocess, shutil, platform, os, glob as glob_mod

SLICER_MAP = {
    "OrcaSlicer":           {"mac_app": "OrcaSlicer",          "win_exe": "orca-slicer.exe",         "win_dir": "OrcaSlicer",          "linux_exe": "orca-slicer"},
    "Bambu Studio":         {"mac_app": "BambuStudio",         "win_exe": "bambu-studio.exe",        "win_dir": "BambuStudio",         "linux_exe": "bambu-studio"},
    "Creality Print":       {"mac_app": "Creality Print",      "win_exe": "CrealityPrint.exe",       "win_dir": "Creality Print*",     "linux_exe": None},
    "Elegoo Slicer":        {"mac_app": "ElegooSlicer",        "win_exe": "elegoo-slicer.exe",       "win_dir": "ElegooSlicer",        "linux_exe": None},
    "Anycubic Slicer Next": {"mac_app": "AnycubicSlicerNext",  "win_exe": "AnycubicSlicerNext.exe",  "win_dir": "AnycubicSlicerNext",  "linux_exe": None},
    "PrusaSlicer":          {"mac_app": "PrusaSlicer",         "win_exe": "prusa-slicer.exe",        "win_dir": "PrusaSlicer",         "linux_exe": "prusa-slicer"},
    "UltiMaker Cura":       {"mac_app": "UltiMaker Cura",      "win_exe": "UltiMaker-Cura.exe",     "win_dir": "UltiMaker Cura*",     "linux_exe": None},
}
MULTICOLOR_SLICERS = {"OrcaSlicer", "Bambu Studio", "Creality Print", "Elegoo Slicer", "Anycubic Slicer Next"}

def detect_slicers():
    found = []
    system = platform.system()
    for name, info in SLICER_MAP.items():
        path = None
        if system == "Darwin":
            app = info.get("mac_app")
            if app and os.path.exists(f"/Applications/{app}.app"):
                path = f"/Applications/{app}.app"
        elif system == "Windows":
            win_dir, win_exe = info.get("win_dir", ""), info.get("win_exe", "")
            for base in [os.environ.get("ProgramFiles", r"C:\Program Files"),
                         os.environ.get("ProgramFiles(x86)", r"C:\Program Files (x86)")]:
                if "*" in win_dir:
                    matches = glob_mod.glob(os.path.join(base, win_dir, win_exe))
                    if matches: path = matches[0]; break
                else:
                    candidate = os.path.join(base, win_dir, win_exe)
                    if os.path.exists(candidate): path = candidate; break
        else:
            exe = info.get("linux_exe")
            if exe: path = shutil.which(exe)
        if path:
            found.append({"name": name, "path": path, "multicolor": name in MULTICOLOR_SLICERS})
    return found

def open_in_slicer(file_path, slicer_name):
    info = SLICER_MAP.get(slicer_name, {})
    system, abs_path = platform.system(), os.path.abspath(file_path)
    if system == "Darwin":
        subprocess.run(["open", "-a", info.get("mac_app", slicer_name), abs_path])
    elif system == "Windows":
        exe_path = shutil.which(info.get("win_exe", ""))
        (subprocess.Popen([exe_path, abs_path]) if exe_path else os.startfile(abs_path))
    else:
        exe_path = shutil.which(info.get("linux_exe", ""))
        (subprocess.Popen([exe_path, abs_path]) if exe_path else subprocess.run(["xdg-open", abs_path]))
    print(f"Opened {abs_path} in {slicer_name}")

# Several SEPARATE/unrelated models (e.g. results from different tasks)? Call this once per
# model with a short gap — Bambu Studio especially may respond to only one if fired
# back-to-back. (Parts of ONE model belong in a single project — not spaced.)
#     import time
#     for f in [model_a, model_b, model_c]:
#         open_in_slicer(f, slicer_name); time.sleep(2)

slicers = detect_slicers()
for s in slicers:
    mc = " [multicolor]" if s["multicolor"] else ""
    print(f"  - {s['name']}{mc}: {s['path']}")
白色模型工作流
步骤操作点数
1生成无纹理模型20
2下载 OBJ0
3修复 OBJ(fix_obj_for_printing)0
4在切片软件中打开0

生成时在 target_formats 中包含 "obj",然后修复模型以便打印:

# --- Generate for white model printing ---
# Text to 3D:
task_id = create_task("/openapi/v2/text-to-3d", {
    "mode": "preview", "prompt": "USER_PROMPT", "ai_model": "latest",
    "target_formats": ["obj"],  # Only OBJ for white model printing
})
# OR Image to 3D:
# task_id = create_task("/openapi/v1/image-to-3d", {
#     "image_url": "URL", "should_texture": False,
#     "target_formats": ["glb", "obj"],
# })
task = poll_task("/openapi/v2/text-to-3d", task_id)
project_dir = get_project_dir(task_id, "print")

obj_url = task["model_urls"].get("obj") or task["model_urls"].get("glb")
obj_path = os.path.join(project_dir, "model.obj")
download(obj_url, obj_path)

def fix_obj_for_printing(input_path, output_path=None, target_height_mm=75.0):
    if output_path is None: output_path = input_path
    lines = open(input_path, "r").readlines()
    rotated, min_x, max_x, min_y, max_y, min_z, max_z = [], float("inf"), float("-inf"), float("inf"), float("-inf"), float("inf"), float("-inf")
    for line in lines:
        if line.startswith("v "):
            parts = line.split()
            x, y, z = float(parts[1]), float(parts[2]), float(parts[3])
            rx, ry, rz = x, -z, y
            min_x, max_x = min(min_x, rx), max(max_x, rx)
            min_y, max_y = min(min_y, ry), max(max_y, ry)
            min_z, max_z = min(min_z, rz), max(max_z, rz)
            rotated.append(("v", rx, ry, rz, parts[4:]))
        elif line.startswith("vn "):
            parts = line.split()
            rotated.append(("vn", float(parts[1]), -float(parts[3]), float(parts[2]), []))
        else:
            rotated.append(("line", line))
    h = max_z - min_z
    s = target_height_mm / h if h > 1e-6 else 1.0
    xo, yo, zo = -(min_x+max_x)/2*s, -(min_y+max_y)/2*s, -(min_z*s)
    with open(output_path, "w") as f:
        for item in rotated:
            if item[0] == "v":
                _, rx, ry, rz, extra = item
                e = " "+" ".join(extra) if extra else ""
                f.write(f"v {rx*s+xo:.6f} {ry*s+yo:.6f} {rz*s+zo:.6f}{e}\n")
            elif item[0] == "vn":
                f.write(f"vn {item[1]:.6f} {item[2]:.6f} {item[3]:.6f}\n")
            else:
                f.write(item[1])
    print(f"OBJ fixed: scaled to {target_height_mm:.0f}mm, Z-up, centered")

fix_obj_for_printing(obj_path, target_height_mm=75.0)
if slicers: open_in_slicer(obj_path, slicers[0]["name"])
多色工作流
步骤操作点数
1生成并添加纹理30
2多色处理10
3下载 3MF0
4在多色切片软件中打开0
mc_slicers = [s for s in slicers if s["multicolor"]]
if not mc_slicers:
    print("WARNING: No multicolor slicer detected. Install: OrcaSlicer, Bambu Studio, etc.")

# --- Generate + texture with target_formats including 3mf ---
preview_id = create_task("/openapi/v2/text-to-3d", {
    "mode": "preview", "prompt": "USER_PROMPT", "ai_model": "latest",
    # No target_formats needed — 3MF comes from multi-color API
})
poll_task("/openapi/v2/text-to-3d", preview_id)

refine_id = create_task("/openapi/v2/text-to-3d", {
    "mode": "refine", "preview_task_id": preview_id, "enable_pbr": True,
})
poll_task("/openapi/v2/text-to-3d", refine_id)
project_dir = get_project_dir(preview_id, "multicolor-print")

# --- Multi-color processing ---
mc_task_id = create_task("/openapi/v1/print/multi-color", {
    "input_task_id": refine_id,
    "max_colors": 4,   # 1-16, ask user
    "max_depth": 4,     # 3-6, ask user
})
task = poll_task("/openapi/v1/print/multi-color", mc_task_id)
threemf_path = os.path.join(project_dir, "multicolor.3mf")
download(task["model_urls"]["3mf"], threemf_path)
if mc_slicers: open_in_slicer(threemf_path, mc_slicers[0]["name"])
可打印性检查清单
检查项建议
壁厚FDM 至少 1.2mm,树脂至少 0.8mm
悬垂控制在 45° 以下,或添加支撑
流形网格水密、无孔洞
最小细节FDM 为 0.4mm,树脂为 0.05mm
底座稳定性使用平底,或在切片软件中添加裙边/筏层
悬空部件所有部件都应连接,或分开打印

步骤 3:报告结果

任务成功后:

  1. 已下载文件的路径和大小
  2. 任务 ID(用于后续操作:细化、绑定、重新贴图)
  3. 可用格式(列出 model_urls 的键)
  4. 已消耗点数及当前余额
  5. 建议的后续操作:
    • 预览完成 → “要细化模型(添加纹理)吗?”
    • 模型完成 → “要为这个角色绑定骨骼吗?”
    • 已绑定骨骼 → “要应用自定义动画吗?”
    • 任何带纹理的模型 → “要 3D 打印这个模型吗?支持多色打印!”
    • 任何模型 → “要 3D 打印这个模型吗?”

错误恢复

HTTP 状态含义操作
401API 密钥无效重新执行步骤 0;请用户检查密钥
402点数不足显示余额,并提供链接 https://www.meshy.ai/pricing
422无法处理说明原因(例如,用于绑定的模型不是人形模型)
429请求频率受限5s 后自动重试(最多 3 次)
5xx服务端错误10s 后自动重试(仅 1 次)

任务 FAILED 消息:

  • "The server is busy..." → 采用退避策略重试(5s、10s、20s)
  • "Internal server error." → 简化提示词,然后重试一次

已知行为与限制

  • 99% 卡住:这是正常的最终处理过程(30–120s)。请勿中断。
  • 资源保留期限:文件将在 3 天后删除(非 Enterprise)。请立即下载。
  • PBR 贴图:必须显式设置 enable_pbr: true。
  • 细化:支持 meshy-5、meshy-6 或 latest——为保持一致,请选择与预览相同的系列。无论模型为何,均消耗 10 点。(meshy-4 已退役 → 400。)
  • 已弃用参数:symmetry_mode 不再影响输出;Meshy-6 会忽略 art_style;请使用 pose_mode 替代旧的 is_a_t_pose 标志。
  • consumed_credits:每个任务 GET 响应都包含 consumed_credits——读取此值以报告实际消耗的点数,而非估算。
  • 绑定骨骼:仅支持双足人形模型,面数 ≤ 300,000。
  • 打印格式:白色模型 → 使用 fix_obj_for_printing() 处理 OBJ。多色模型 → 使用多色打印 API 生成 3MF。请务必先检测切片软件。
  • 下载格式:下载前询问用户需要哪种格式。GLB(查看)、OBJ(打印)、3MF(多色打印)、FBX(游戏)、USDZ(AR)。请勿下载所有格式。
  • 多色打印的 3MF:多色打印 API 会直接输出 3MF——无需在生成/细化时请求 3MF。对于需要 3MF 的非打印用途,请在 target_formats 中传入 "3mf"。
  • 时间戳:所有 API 时间戳均为 Unix 纪元毫秒。

执行检查清单

  • 已运行 API 密钥检测(步骤 0)——仅检查环境变量和 .env
  • 已验证 API 密钥(从未完整打印)
  • 已展示费用摘要并获得用户确认
  • 已将完整工作流写为单个 Python 脚本
  • 已使用 python3 -u 运行,以启用无缓冲输出
  • 已报告文件路径、格式、任务 ID 和余额
  • 已建议后续操作

其他资源

如需查看包含所有参数、响应架构和错误代码的完整 API 端点参考,请阅读 reference.md。

评论

支持 Markdown