chaiNNer Download: Node-Based AI Image Upscaling

chaiNNer lets you build image pipelines by connecting nodes: load images, run them through AI upscaling models, adjust them and save them, then repeat for a whole folder or a video. It works with ESRGAN-family upscaling models, including Real-ESRGAN, and many other architectures.

What does chaiNNer do?

• AI upscaling: Runs super-resolution models through PyTorch, NCNN, ONNX or TensorRT. PyTorch models are loaded with the Spandrel library, which supports many architectures.
• Node editor: Drag-and-drop nodes with color-coded connections that only link compatible types.
• Batch processing: Process whole image folders with Load Images, and videos with Load Video.
• Other tasks: Background removal with ONNX models and many classic image operations.
• CLI: A command-line mode for automation, documented in the wiki.

Who is it for?

It suits artists, game modders and restorers who upscale many images and want repeatable, customizable pipelines.

Installing chaiNNer

Download the installer for your system from the official releases page with the Download button and run it; the Windows installer is about 150 MB. Python is not required: chaiNNer downloads its own integrated Python on first start. Then open the dependency manager (the button in the upper-right corner) and install the AI framework you want, such as PyTorch, NCNN or ONNX.

GPU support

• NVIDIA: Full support through PyTorch (CUDA), ONNX and TensorRT; TensorRT is fastest on supported models.
• AMD: NCNN on all platforms, PyTorch through ROCm on Linux.
• Apple Silicon: PyTorch with MPS acceleration; ONNX runs on the CPU only, and NCNN may not work on some configurations.
• Intel: NCNN.
• CPU: All frameworks have a CPU fallback.
• Systems: Windows 8.1 and older and macOS 10.x are not supported. On Linux, the clipboard nodes need xclip or wl-copy.

License and models

chaiNNer is free and open source under the GPL-3.0 license. Models you load have their own licenses.