What is Image Pipes?
Image Pipes is a visual editor for classical computer vision and data augmentation. You drag nodes such as resize, blur, threshold or find contours onto a canvas, connect them, and see each node's output image, bounding boxes, masks, keypoints and run time as you change parameters. When the pipeline looks right, it exports standalone Python that calls OpenCV and Albumentations directly.
The backend is FastAPI and Python; the editor is React 19 with React Flow, wrapped in Electron. The graph runs as a directed acyclic graph with caching and partial recomputation, so changing one node reruns only what depends on it. A Custom Python node takes a process(image, seed=0) function written in a built-in Monaco editor, and saved scripts are versioned on disk so old pipelines keep their code.
It is MIT-licensed, with desktop installers for Windows, macOS and Linux on the releases page. It runs no neural networks yet; ONNX and PyTorch nodes, video and batch processing are on the roadmap.
Is Image Pipes worth it?
Image Pipes does one job well: it removes the edit, run, save, open loop from tuning classical image operations. Every node shows its output live, and the exported code is ordinary cv2 and Albumentations calls you can paste into a training script. That export is the main reason to prefer it to a notebook full of imshow calls; nothing you build is tied to the app.
Augmentation is the stronger half. Albumentations transforms run inside the graph and carry bounding boxes, segmentation masks and keypoints along with the image, with seeded, repeatable results. That makes it a quick way to check that a rotation or crop keeps labels aligned before a long training run, which is where augmentation bugs usually hide.
The gaps are significant. It processes one image at a time, with batch processing, video and model inference all still on the roadmap, so it prepares code rather than running datasets. Custom Python nodes execute arbitrary code; a trust prompt blocks workflows containing it until you confirm, and that trust is not saved into exported JSON, which is the right default for shared workflow files.
Activity has slowed. Six releases came out between 31 July and 15 August 2026, ending at v0.3.0; since then the commits are a minimap feature on 20 September and dependency bumps through 29 October. The project has 130 stars, 20 forks and three contributors, and the v0.3.0 installers have been downloaded 80 times in total. ComfyUI covers far more ground for generative work, but it is not built around OpenCV or around exporting a script you own.
Where it earns its place
An ML engineer training an object detector on warehouse photos
Check that my augmentations keep bounding boxes on the right objects
Load a labelled sample, chain flips, rotations and cutout, and watch the boxes move with the image at each node; export the Albumentations code into the training loop. Cost is nothing beyond a desktop install.
A university lecturer teaching an introductory computer vision module
Show students what each OpenCV step does to an image
Build grayscale, blur, Canny and contour pipelines live in class and change kernel sizes or thresholds while students watch; the exported Python becomes the lab handout.
A quality-inspection engineer at a small manufacturer
Tune a threshold-and-contour defect check before writing the production script
Adjust thresholds against a handful of real part photos with per-node previews, then export the cv2 calls. Running it over thousands of images still happens in your own script, since batch mode is not built yet.
Before you commit
- Setup
- Download the .exe, .dmg or .AppImage from Releases with no Python or Node needed, or clone and run npm run install:all then npm run desktop; the Electron app starts the backend itself. Source builds need Python 3.12+.
- Integrations
- OpenCV, Albumentations, NumPy and Pillow on the backend; REST and WebSockets between editor and backend; workflows save as JSON; new backend nodes appear in the editor automatically from their metadata.
- Data and privacy
- Runs locally; uploads, outputs, cache and versioned scripts are stored on disk, under Electron userData in the desktop app. The README describes no telemetry or hosted calls. MIT.
- Limits
- Single-image execution only; no GPU acceleration; 17 open issues and pull requests; no release since 15 August 2026.
What’s good
- Live output, run time and logs at every node
- Exports plain OpenCV and Albumentations Python with no runtime
- Augmentations keep boxes, masks and keypoints aligned
- Trust prompt before running custom code; not saved in exports
- Installers for Windows, macOS and Linux
What’s not
- One image at a time; no batch, video or model inference yet
- No release since v0.3.0 on 15 August 2026
- Small project: 130 stars, three contributors
- Electron desktop app; no web or headless mode
Best for
Computer vision engineers and students who tune OpenCV preprocessing or Albumentations augmentation by eye and want the result as plain Python.
Image Pipes alternatives
Image Pipes: common questions
- Is Image Pipes worth it?
- Image Pipes takes the guesswork out of tuning OpenCV and augmentation steps, but it is a young one-person playground with no batch mode, not a pipeline tool. We rate it 3.5 out of 5.
- How much does Image Pipes cost?
- Free under MIT; no hosted or paid edition.
- What are the downsides of Image Pipes?
- One image at a time; no batch, video or model inference yet; no release since v0.3.0 on 15 August 2026; small project: 130 stars, three contributors; electron desktop app; no web or headless mode.
- Who is Image Pipes best for?
- Computer vision engineers and students who tune OpenCV preprocessing or Albumentations augmentation by eye and want the result as plain Python.
- Is Image Pipes safe for company data?
- Runs locally; uploads, outputs, cache and versioned scripts are stored on disk, under Electron userData in the desktop app. The README describes no telemetry or hosted calls. MIT.
- What are the best Image Pipes alternatives?
- The closest alternatives we have reviewed are ComfyUI.
Answers checked Nov 5, 2026.