chaiNNer
Node-based desktop app that chains image tasks such as AI upscaling, using PyTorch, NCNN, ONNX or TensorRT models.
Connect nodes to upscale, batch-process and edit images or video with neural network models.
chaiNNer is a node-based image processing app that started as an AI upscaler. It suits photographers, restorers and game-mod artists who want to chain upscaling, background removal and other steps into repeatable graphs on their own computer.
What it is
You drag nodes into an editor, connect colour-coded handles and press run. It loads upscaling models through PyTorch (via the Spandrel package), NCNN, ONNX and TensorRT, and has Load Images and Load Video nodes for batch work. It carries its own isolated Python so you need not set one up.
Who it's for
Good for people upscaling or restoring many images with community models and who like visual pipelines. If you only need a one-off enlargement, a plain upscaler app is simpler, and you must install a neural-network framework through the dependency manager first.
Key features
Node-graph editor
Drag nodes in, connect typed handles and run the chain with the green button.
Four inference backends
PyTorch, NCNN, ONNX and TensorRT, with TensorRT noted as fastest on supported Nvidia models.
Batch images and video
Load Images and Load Video nodes process folders and clips, though you cannot combine two batch nodes in one chain.
Integrated Python
chaiNNer downloads its own isolated Python build and installs dependencies from an in-app manager.
Background removal nodes
ONNX background-removal models such as u2net and isnet are supported.
Wide GPU support
Nvidia, AMD via ROCm or NCNN, Apple Silicon via MPS, Intel via NCNN, and CPU fallback.
Up and running in 3 steps
Our honest verdict
What's great
- Installers exist for Windows, macOS and Linux, and the README says you do not need Python installed.
- Chains are reusable, so one graph can handle a whole folder the same way every time.
- It reads many community super-resolution architectures through Spandrel, with OpenModelDB listed as a model source.
- Works on AMD, Intel and Apple GPUs as well as Nvidia, which many AI image tools do not.
Where it falls short
- You must install at least one neural-network framework through the dependency manager before a first upscale, and those packages are large.
- The licence is GPL-3.0, which matters if you want to build chaiNNer into closed software.
- README compatibility notes list quirks: some NCNN users with non-Nvidia GPUs see all-black output, and ONNX uses CPU only on Apple Silicon.
- Node graphs have a learning curve, and the README admits the interface can look intimidating at first.
Pricing & license
chaiNNer is GPL-3.0 and free to download and use. The README shows a Ko-fi page for donations and no paid plan. Your costs are the computer and GPU you run it on.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Do I need to install Python?
No. The README says chaiNNer downloads an isolated integrated Python build on startup. You can switch to your system Python (3.10 or later) if you prefer.
Which GPUs work?
Nvidia via PyTorch, ONNX and TensorRT; AMD via ROCm on Linux or NCNN; Apple Silicon via MPS; Intel via NCNN; and CPU as a fallback.
Can it process video?
Yes, with the Load Video node, but you cannot combine it with Load Images or any other batch-iteration node in one chain.
