TMTuMo VisionDrawing-to-delivery AI
Download freeDownloadBook a demo
Download

Download TuMo Vision for Windows

One self-contained installer: import drawings, detect dimensions, review OCR, produce ballooned drawings and inspection reports. No separate .NET runtime required.

Download installerSee all versions

200 free detections on sign-up

System requirements

  • Windows 10 / 11 (64-bit)x64 installer only; 32-bit and ARM systems are not supported.
  • No .NET runtime neededSelf-contained build; the runtime is included in the installer.
  • Network and accountInstall and local detection work offline; cloud OCR (DeepSeek vision model) needs a network connection and a signed-in account.
  • GPU optionalRuns on CPU; NVIDIA GPUs can use CUDA, some GPUs support DirectML.

Install in three steps

  1. Download the installer

    Save the .exe; close any running older version before installing.

  2. Run the installer

    It installs the VC++ runtime automatically and may ask for UAC confirmation.

  3. Sign in and start

    Sign up or sign in, claim 200 free detections and import your first drawing.

The runtime is included — no separate .NET download. Installation defaults to the current user; installing for all users requires administrator rights.

First run

Three things to prepare after installing

Installation is only the first step. Confirm account, network and inference device, and the first detection run goes much smoother.

Account and quota

Sign-up includes 200 free detections; cloud OCR (DeepSeek vision model) requires a signed-in account.

Network and proxy

Cloud OCR calls the backend service. If the workstation uses a proxy or intranet, confirm it can reach the service first.

Inference device

Settings → Detection switches CPU / GPU CUDA / DirectML; fall back to CPU if the GPU is unavailable.

Installer checksum

Verify the installer after download; version and file information come from the version API.

SHA-25616515e963b36306713bc925c594f84bdf248b2e4455942c676a4d33b9085f53b
Install issues

The four most common install questions

If none of these cover your case, contact us — do not get stuck at installation.

The antivirus blocks the installer.

Some antivirus products flag new installers. Choose to trust and continue, or ask IT to allow it. Enterprise environments can request the offline package and checksum from us.

Can I install without administrator rights?

Installation defaults to the current user and does not require administrator rights. Installing the VC++ runtime or choosing all-users installation will request elevation; the default path avoids it.

The GPU is not available after launch.

Switch to CPU or DirectML under Settings → Detection and finish the work first; then check drivers and CUDA against the GPU vendor documentation. CPU works, just slower.

Cloud OCR reports signed-out or network errors.

Sign in and confirm the machine can reach the backend. Fully offline or intranet environments need a private deployment review — tell us about your network conditions.

Releases

The latest three versions

The full changelog and older versions live on the release notes page. Only the most recent changes are shown here so the download page stays short.

Read the full changelog

Release notes are published in Simplified Chinese by the product team.

v3.3.0图眸 TuMo 3.3.0 更新内容2026-10-28 · 170.7 MB

图眸 TuMo 3.3.0 更新内容 【更快】37 页 PDF 导入 10.6s→最快 0.4s；项目打开 132.7s→5.5s；单次检测约 1.1s→0.8s；45 页导出 PNG 40.7→21.7s。 【更省】37 页导入峰值内存 2.53GB→约 350MB。 【更好用】量具自动识别、检测字段设置、批量检测可取消、未保存保护、PDF 高清缩放、气泡编辑器增强。 【更稳】安装包拒绝降级；设置原子写入并支持损坏恢复；VC++ 运行环境内置；修复云同步、OCR、导出多处问题。 【模板配置】预览支持全列显示、全览、缩放、表头吸顶；修复合并单元格过多导致 AI 配置失败。 建议在 3.2.0 基础上直接升级，项目、设置与报告模板会保留。

Download this version
v3.2.0重写视图分组算法，现在能更好区分视图，一个试图排序完全后 才跳到下一张试图排序2026-10-07 · 113.0 MB

1.重写视图分组算法，现在能更好区分视图，一个试图排序完全后 才跳到下一张试图排序 2.新增繁體中文，界面文案全部支持简体/繁体/英文切换，缺词条自动回退不空白 3.修复英文界面还有很多中文问题。 4.「显示方式」新增"置信度"：在检测框左上角显示置信度，同一张图统一字号 5.检测后切换置信度挡位可快速过滤低置信检测框（手动添加与已复核的保留，可随时切回） 6.品检报告模板配置支持自动识别工作表/表头/明细区与字段列映射，配置升级不再丢失 7.年度会员不再限制 OCR次数 与检测次数。 8.修复记住登录/自动登录失效、清空图纸后内存不释放、模板预览空单元格显示、英文模式启动崩溃等问题 9.随包发布 1280 多行业检测模型，安装即可使用 10.检测模型大更新，增加标注图纸进行训练，由800张图纸增加至8000+张图纸进行训练。提高图纸泛化率 解释：原模型对没经过训练的图纸识别率低是因为训练过拟合了，现重新训练后对图纸大幅提高检测率，即便是复杂图纸识别率高达97%，！！！ 模型名：(1280多行业检测模型). 以前旧模型移入模型市场，如果认为以前模型更好，可以下载使用。

Download this version
v3.1.3新增特征描述字段，优化启动逻辑与项目保存2026-09-13 · 246.0 MB

feat: 新增特征描述字段，优化启动逻辑与项目保存 1. 新增特征描述字段支持OCR识别结果分类 2. 调整启动时加载上次项目的逻辑，默认不自动加载以加快启动 3. 优化项目序列化与反序列化，支持内嵌PDF存储与恢复 4. 完善导出服务对角度公差和形位公差的计算逻辑 5. 更新OCR提示词，新增特征描述字段的规范说明 6.保存项目后 如果点击清空图纸列表则等于新建项目，再次保存时将重新保存文件 点击“品检报告”后会先显示图纸选择列表，默认全部勾选。 快捷操作包括：全选、全不选、反选、只选当前页、只选有检测项；底部会实时显示已选择数量。确认后才进入模板和输出目录选择，最终只导出所选图纸。 在“模板配置 → 字段列映射”中，可将“特征描述列”指定为任意 Excel 列；导出时会写入对应检测项的描述，如“距离、角度、沉头、销孔、垂直度”等。AI 自动配置模板也会识别并建议该列。 feat: 添加特征描述列支持，新增图纸选择导出功能 1. 新增FeatureDescription列配置，支持在报告中导出特征描述信息 2. 新增图纸选择导出弹窗，允许用户选择需要导出的图纸页面 3. 优化OCR提示词，调整特征描述相关规则与示例 4. 调整主数据表格列宽与滚动条显示策略

Download this version
Enterprise

Intranet, offline or volume licensing?

Tell us your device count, concurrency and data requirements; we will assess private OCR, offline models and volume licensing and propose a concrete deployment.

Contact sales Data & security

Run your first drawing in ten minutes

Follow the tutorial from import to export and produce your first ballooned drawing and inspection report.

Read the tutorial Back to features