A visual wiring kit for image-processing pipelines

What it does

ChaiNNer is a cross-platform visual node editor for building image-processing workflows. You arrange nodes on a canvas, connect them with color-coded wires, and execute the chain. It began as an AI upscaling application but has grown into a general-purpose programmatic image processor.

The interesting bit

The application carries its own isolated Python build, so it will not touch your system Python environment. It supports multiple neural-network backends—PyTorch, NCNN, ONNX, and TensorRT—and only exposes compatible connections between nodes, which keeps the visual programming from devolving into guesswork.

Key highlights

• Broad GPU support: Nvidia via CUDA and TensorRT, AMD via ROCm on Linux or NCNN everywhere, Apple Silicon via PyTorch MPS, Intel via NCNN, plus CPU fallback for all frameworks.
• Batch processing for folders of images or video files, though the README notes you cannot mix batch-iteration nodes such as Load Images and Load Video in the same chain.
• Background removal via ONNX models including u2net and isnet, alongside super-resolution tasks.
• PyTorch model support is handled through the project’s own Spandrel package.
• Clipboard nodes on Linux require xclip or wl-copy to be present.

Caveats

• macOS 10.x and Windows 8.1 or below are not supported.
• On Apple Silicon, ONNX is restricted to the CPU Execution Provider and NCNN may not work properly on some configurations.
• Some NCNN users with non-Nvidia GPUs report all-black outputs, which the README attributes to graphics-driver memory limits; manual tiling may mitigate it.

Verdict

Try it if you want a node-based alternative to scripting image pipelines, particularly for AI upscaling or batch conversion. Avoid it if you are on an older OS or need the flexibility of handwritten imperative code.