Chromaform started with a simple plan: download darktable on an iPad and start editing. Then we discovered it wasn’t on iOS. Then we discovered it couldn’t be ported — darktable’s architecture is deeply tied to desktop Linux and its CPU-bound processing model (and its GPLv3 license is incompatible with App Store distribution). There was no path to getting it onto an iPad.
But we didn’t want to give up the workflow. If you’ve used darktable, you know — it’s a superior editing experience. The non-destructive pipeline, the module-based processing, the level of control it gives you over every stage of the image. Nothing on the App Store comes close. Other editors either dumb things down to a stack of sliders or route your photos through someone else’s cloud before you can touch them.
So we set out to build it. Not a port, but a ground-up effort to bring that darktable philosophy to iPad — while embracing what makes the platform unique. The M-series GPU. The Apple Pencil, which makes mask painting and local adjustments feel as intuitive as drawing on paper. The color-calibrated Retina display that shows you exactly what your file looks like. We wanted to take everything that makes darktable great and combine it with everything that makes iPad great.
Standing on the Shoulders of Open Source
Chromaform is built on vkdt, a Vulkan-native photography pipeline created by Johannes Hanika — who also created darktable, one of the most respected open-source RAW editors in the photography world. If you’ve used darktable, you know the philosophy: non-destructive editing, real processing pipelines, and respect for the photographer’s control over his image.
vkdt takes that philosophy further. Instead of the CPU-bound processing model of traditional editors, vkdt builds the entire pipeline as a directed graph of Vulkan compute shaders. Every operation — demosaicing, color transforms, denoise, tone mapping — runs as a GPU kernel. The result is a fundamentally different architecture: one where the GPU does the work it was designed for, and the processing graph can be rewired by the user.
vkdt is released under the 2-Clause BSD License, with LGPL dependencies for rawspeed and FFmpeg. We’re grateful to Johannes and the open-source community for building such a solid foundation.
Why iPad, and Why Now
Apple’s M-series chips changed everything. An M2 iPad Pro has GPU compute performance that rivals a desktop workstation from a few years ago — and Apple’s Neural Engine adds 15+ TOPS of dedicated ML inference hardware. This is not a phone with a big screen. It’s a legitimate processing platform.
Yet the App Store has nothing that takes this hardware seriously for RAW photography. The dominant editors are cloud-dependent or flatten your files into 8-bit RGB before you ever see them. The M-series GPU sits mostly idle.
Chromaform puts that hardware to work.
The Pipeline
The core of Chromaform is a Vulkan compute pipeline running through MoltenVK on Metal. When you load a RAW file, here’s what actually happens:
Full sensor decode. We read the actual sensor data — not an embedded JPEG preview. A 50-megapixel DNG gives you all 8192x5464 photosite values, decoded at the scale you choose.
Node-based processing graph. Your image flows through a configurable graph of processing nodes: demosaic, white balance, exposure, denoise, color transforms, tone mapping, and output. Each node is a Vulkan compute shader dispatched to the GPU.
16-bit floating point precision. The pipeline stores intermediate results as FP16 half-float textures — the same format used in professional VFX pipelines via OpenEXR — while compute shaders perform math in full FP32 precision. The floating exponent gives you more precision in the darks where your eye is most sensitive, and FP32 intermediate math means quantization error stays well below perceptual thresholds even through a long chain of processing nodes.
Non-destructive by design. Your edits are parameters on graph nodes — the source file is never modified. Change a parameter and the affected portion of the graph re-executes on the GPU in real time.
Foreground/background separation. A core architectural decision in Chromaform is treating foreground and background as independent editing targets. Want to push the saturation in an ocean sunset without shifting the color of a model’s clothing? Grade the sky warmer without affecting skin tones? Selective foreground/background color grading makes this a first-class operation, not a workaround involving masks and layers.
Local AI, Not Cloud AI
Every other app wants to upload your photos to run “AI enhancements” on a server farm. We think that’s backwards. Your iPad has a Neural Engine purpose-built for inference. Chromaform runs its ML models — parameter inference, scene analysis — entirely on-device through CoreML and the Apple Neural Engine.
Your photos never leave your iPad. Not for processing, not for “AI features,” not ever.
What’s Next
Chromaform is in active development. This blog will be a mix of development updates and technical deep-dives into the pipeline, the challenges of bringing Vulkan to iOS, and what it takes to build a professional tool for a platform that deserves one.