chaiNNer is a node-based visual programming environment for Mac that lets you design complex image-processing pipelines by wiring together modular operations — no Python scripts required.
What is chaiNNer?
chaiNNer is a free, open-source desktop application that replaces image-processing scripts with an interactive node graph. You build a pipeline visually: drag in a node for loading images, chain it through an upscaling model or colour-correction block, and pipe the result into an export node. What would otherwise require a hand-rolled Python environment and a pile of dependencies becomes a canvas you can reason about at a glance.
The killer feature is first-class support for AI-based upscaling models — ESRGAN variants, Real-ESRGAN, SwinIR, and friends. Drop a trained .pth model file onto the graph and chaiNNer handles inference, tiling for VRAM-constrained hardware, and batch processing across entire folders, all without a single line of code.
What does chaiNNer do best?
chaiNNer shines at batch AI upscaling and non-destructive image pipeline authorship. Once you wire up a graph, you can point it at a folder of hundreds of images and walk away — it iterates, applies every node in sequence, and writes output without you babysitting each file.
- Node-based composition: every operation is a discrete block; swap, duplicate, or bypass nodes without rewriting anything.
- AI model integration: load ONNX, PyTorch, and NCNN models directly; the app resolves backend dependencies through its own package manager so you don't wrestle with conda or pip.
- Tiling: large images are split into overlapping tiles for inference, then stitched — critical if you're upscaling 4K source material on consumer GPU memory.
- Iterators: a dedicated iterator node loops a sub-graph over every file in a directory, making bulk jobs trivial to set up.
- Portable graphs: save and share your pipeline as a .chn file, which colleagues can open and run without reconstructing anything.
I've personally used it to upscale sprite sheets for a retro-game preservation project. What Topaz Gigapixel does for photographers, chaiNNer does for people who need control, repeatability, and zero subscription fees.
Is chaiNNer free?
Yes — chaiNNer is completely free to download and use, with no feature tiers, no paywalled models, and no usage caps. The project is open-source on GitHub under the GPL licence, actively maintained by its community, and available via Homebrew Cask for painless Mac installation. The only costs are your time assembling good upscaling models (most are free to download from community repositories) and the compute your GPU provides.
Who should use chaiNNer?
chaiNNer is made for people who need systematic, repeatable image transformation at scale. That means digital artists batch-enhancing textures, game modders upscaling low-resolution sprites, archivists restoring scanned documents, and photo enthusiasts who want AI upscaling without a monthly Topaz or Adobe bill.
If your workflow is occasional one-off edits, Photoshop or Pixelmator Pro will serve you better — their direct-manipulation tools are faster for individual images. But if you find yourself applying the same sequence of operations to dozens or hundreds of files, chaiNNer's node graph pays for its learning curve within the first serious batch run.
A basic comfort with concepts like resolution, colour spaces, and what an AI upscaling model does will help — chaiNNer does not hide its technical seams the way consumer apps do.
How does chaiNNer compare to Topaz Gigapixel?
Topaz Gigapixel AI is the commercial benchmark: polished UI, curated proprietary models, and a one-click experience optimised for photographers. chaiNNer is the power-user alternative. You trade Gigapixel's polish for the freedom to load any community-trained model, compose multi-step pipelines (upscale, then sharpen, then convert colour space, then export at a specific DPI — all in one graph), and run unlimited batches without a subscription.
For raw upscaling quality the gap depends entirely on which models you choose; community ESRGAN models trained on specific domains (anime, film grain, architectural scans) often outperform Gigapixel's general-purpose models on their target content. chaiNNer also integrates well into GPU-heavy Mac setups and supports both ONNX and NCNN backends for Apple Silicon acceleration where available.
What are the best chaiNNer alternatives?
The closest open-source peer is Upscayl, which is simpler and more approachable but lacks the node graph and multi-step pipeline support. Topaz Gigapixel AI and Topaz Photo AI are the premium commercial options with a friendlier learning curve. For pure node-based image work without the AI focus, Darktable covers non-destructive raw editing in a similar philosophy. If you're already living in the terminal, Python scripts with the spandrel or basicsr libraries replicate chaiNNer's core capability — but without the visual graph or the batch iterator convenience.
