PhotoCraft: open-source Photoshop reimplementation in Rust lands with layers, masks and PSD support

Photoshop rebuilt in Rust reaches Hacker News

PhotoCraft, an open-source project that presents itself as a clean-room reimplementation of Adobe Photoshop written entirely in Rust, has reached the Hacker News front page through its GitHub repository. According to the project, the application covers the parts of Photoshop that matter most to working image editors — layers, masks, adjustment layers, layer styles, text, vector shapes, brushes and genuine PSD files — in a native desktop app that runs offline. The Hacker News title called it a cleanroom "port", but the project is more precise: this is a from-scratch rebuild with no Adobe code, published under the GitHub account storytold alongside a companion site and an associated ArtCraft artist community.

The project pitches itself as immediately familiar: menus, shortcuts and panels mirror Photoshop's layout, so an experienced Photoshop user should find their way around without retraining.

Non-destructive editing at the core

The feature list reads like a checklist of Photoshop's layer machinery. The app offers sixteen adjustment layers — including Curves with per-channel editing, Levels with a live histogram, Black & White, Channel Mixer, Gradient Map and colour lookup files in .cube, .3dl and .look formats — which can also be applied directly to pixels. Layer styles such as Drop Shadow, Inner and Outer Glow, Bevel & Emboss, Satin, Stroke and pattern overlays work on any layer, including live type, and styles can be copied between layers or loaded straight from existing PSDs.

Selection tooling spans marquees, lassos and the Magic Wand alongside automated options like Quick Selection, Object Selection and Select Subject, which the project says run locally, with no cloud service and no account. Text layers stay editable on the canvas and round-trip through PSD, and vector support includes the Pen tool plus shape layers that combine via unite, subtract, intersect and exclude operations.

PSD compatibility, measured

File fidelity is where the project makes its boldest, most quantified claims. According to the repository, re-saved documents keep the same render as Photoshop for 307 of the 309 PSD test files used by the psd-tools library, and all 116 shape layers in the project's own PSD corpus match Photoshop's output pixel for pixel. Both PSD and its large-document sibling PSB are supported.

Native and GPU-accelerated

Rather than wrapping a web interface in Electron, the app is native, with a GPU compositor built on wgpu that supports Metal, Vulkan, DirectX 12 and WebGPU backends. The engine uses copy-on-write tiles and multithreaded filters; the project claims a Gaussian blur with a radius of 180 on a 3.6-megapixel image completes in under a second by spreading running-sum box passes across all CPU cores. Filters preview live on the canvas through the active selection, and filters applied to smart objects remain editable, reorderable and maskable afterwards.

Colour support is broad: RGB, Grayscale, CMYK and Lab documents at 8, 16 and 32 bits per channel, with bit depth and colour model treated as runtime data so every tool works at every depth. ICC colour management is implemented in pure Rust, with embedded profile support, rendering intents, black point compensation, GPU-accelerated soft proofing and gamut warning.

The toolbox lists 34 tools, from the Clone Stamp and Healing Brush through Dodge, Burn and Sponge. The brush engine supports dynamics such as Shape Dynamics, Scattering, Dual Brush and Wet Edges, driven by pen pressure, tilt, rotation and direction, and strokes are deterministic and replayable.

Built to be driven by software as well as people

Unusually for a graphics application, every action is implemented as a command, so the same engine can be driven from the UI, a CLI, a JSON control channel or an MCP server, the protocol AI agents increasingly use to operate external tools. The project says the screenshots on its site were rendered offscreen through that control channel, which doubles as a demonstration of the automation surface.

Why it matters

For decades, the practical barrier for professionals considering a Photoshop alternative has not been feature count but file compatibility: layered PSDs are the lingua franca of commercial image work. A clean-room implementation in a memory-safe language, measuring itself against Photoshop's actual output on a public test corpus, is a different proposition from approximate alternatives — if the numbers hold up under independent testing. The command-first architecture also anticipates a near future where routine image editing is orchestrated by AI agents rather than performed click by click. All specifics above come from the project's own repository, surfaced via Hacker News, and await hands-on verification.