--- title: "PixGamma" url: https://phatdev.net/en/projects/pixgamma lang: en alternate: vi: https://phatdev.net/vi/projects/pixgamma markdown: https://phatdev.net/en/projects/pixgamma.md --- # PixGamma > macOS app for AI image upscaling and restoration that runs entirely on the Apple Silicon GPU: product shots, portraits, old family photos, wildlife, landscapes, illustration and screenshots. Text stays correct, Vietnamese included, with diffusion-based Recover and a CLI and local API for developers. - Role: AI · macOS · Image restoration - Status: Version 0.2.1, free for non-commercial use; the new interface and Photo v2 are on GitHub - Technology: Swift, SwiftUI, Core ML, PyTorch, Diffusion - Docs and changelog: https://phatdev.net/docs/pixgamma - All projects: https://phatdev.net/en/projects.md ## Project facts - Role: Product owner, built with Claude Code - Stack: Swift 6, SwiftUI, Core ML, PyTorch - Version: 0.2.1 (Oct 20, 2026) - Runs on: macOS 14+, Apple Silicon ## Metrics - 0,170: DISTS of PixGamma Photo v2 on sharp photos (v1: 0.183; lower is closer to the original) - 91,1%: banner text read correctly after a 4x upscale - 16x: maximum upscale, stacked 2x and 4x passes - 6: AI modes, including diffusion-based Recover ## Highlights - **Photos never leave the Mac**: Every model runs through Core ML on the Apple Silicon GPU: no account, no server, nothing uploaded. - **No invented text**: Apple Vision finds text, Vietnamese included; text regions are measured from the pixels and rebuilt with a faithful model for straight strokes. - **Own models, benchmarked before shipping**: PixGamma Photo v2 is Photo v1 trained further with a light GAN for crisper fur, hair and fabric; text regions keep the GAN-free Photo v1 so no strokes are invented. Any version that loses on the benchmark or real photos is dropped. - **Diffusion-based Recover**: SD 1.5 + ControlNet Tile on Core ML. Apple Vision describes the photo on the Mac (e.g. "parrot, detailed feathers") for the prompt; animals, hair and skin get Resemblance 85; photos with text, maps or products get a safety lock that keeps shapes. - **New interface, Light and Dark**: The photo fills the window, a small dock picks mode and scale, a filmstrip sits below and the tuning column collapses. Fit shows the whole photo with the preview area outlined; 100% and 200% compare split or side by side. Export one photo or all of them. - **Install once, updates itself**: DMG installer, auto-update from GitHub with SHA-256 checks and rollback. - **One app for many kinds of images**: Product shots, portraits, old family photos, wildlife, landscapes, architecture, illustration and screenshots; Auto picks the model for each image. - **For developers**: The pixgamma command handles whole folders with JSON reports; an HTTP API on the Mac only plugs into scripts, websites or apps. ## In-depth analysis ### The problem Images that need enlarging are everywhere: product shots for an online store, old photos from the family album, wildlife and landscapes headed for a large print, illustrations, screenshots for documentation. They are often small, soft or compressed several times over. Popular AI upscalers tend to fail in familiar ways: text gets redrawn as other characters, skin, fur and fabric get covered in invented texture, colors drift, and the photos have to be uploaded to someone else's server. ### The solution PixGamma is a macOS app written in Swift that runs entirely through Core ML on the Apple Silicon GPU: photos never leave the Mac. A Light and Dark interface lets the photo fill the window: a small dock for modes, a filmstrip below, before/after at real pixels at 100% and 200% (split or side by side), and export for one photo or all. It has an **Auto** mode that analyses each image to pick the model and strength, a `pixgamma` command for batch work and a local HTTP API for scripts, websites and other apps. #### What it is for | Kind of image | Suggested mode | What PixGamma does | |---|---|---| | Product and e-commerce photos | Standard | Labels, keycaps and packaging text stay correct; aluminium, fabric and leather get no invented texture | | Portraits and group shots | Standard + Realistic faces | Clearer eyes, hair and skin, original features kept | | Old family photos, scans | Low Resolution | Grain and compression noise cleaned up, original tones kept | | Wildlife and nature | Standard | Fur, whiskers and feathers come back as strands instead of smears | | Landscapes, architecture, print | Standard at 2x | Straight lines and crisp detail for large prints | | Illustration, anime, graphics | Art & CGI | Clean lines, flat color areas, no added grain | | Screenshots and UI | Auto at 2x | Clearer text, Vietnamese diacritics kept | | Badly blurred photos | Recover | Skin, hair and fur regenerated with diffusion; it reads the scene to pick the prompt and how closely to follow the photo | | Developers and automation | CLI, local API | Whole folders, JSON reports, HTTP calls on the Mac | #### Our own photo model **PixGamma Photo v2** is the default photo model: Photo v1 trained further with a light GAN for crisper fur, hair and fabric grain (DISTS on sharp photos 0.170 vs 0.183 for v1). Text regions still run on the GAN-free Photo v1, so the GAN never draws letter strokes. **PixGamma Photo v1** is a DAT2 model fine-tuned for 15,000 iterations on the Mac itself, starting from the strongest of 10 public models we benchmarked. Training data is CC0 and public domain, low-resolution inputs follow a web-like downscale + double JPEG chain, and there is no GAN so it does not invent texture. On 40 photos never used for training: PSNR 30.5 dB (base 29.0), DISTS 0.165 (0.173), almost no color drift. #### Text protection, Vietnamese included Apple Vision finds text on the source, a 2x enlargement and the upscaled result. Each text region is measured from the pixels and rebuilt with a faithful model. Text too small or too noisy to rebuild correctly keeps the plain upscale instead of being redrawn as other letters. On the Vietnamese and English banner benchmark: 91.1% OCR accuracy. #### Diffusion-based Recover Realistic Vision 5.1 + ControlNet Tile on Core ML, seamless tiling, a hand-written DPM++ 2M SDE sampler and a dedicated eye pass. Apple Vision describes the photo on the Mac (for example "parrot, detailed feathers") for the prompt; animals, plants and portraits get Resemblance 85 so fur and hair follow the photo. Photos with text, maps or products get a safety lock that caps creativity and keeps shapes. #### Shipping and updates DMG installer and auto-update from GitHub Releases with SHA-256, bundle id and version checks and a rollback-safe app swap. A model only ships when it beats the current one on both the benchmark and real photos. ### Current limits - A 4x upscale from a very small image (under ~500 px) with dense foliage or ornament can look painterly; for print, a 2x upscale from a larger source looks more natural. - Text that is too small (about 3 px tall in the input) cannot be rebuilt: PixGamma keeps a soft upscale instead of guessing letters. Text Apple Vision cannot find gets no dedicated protection. - Face restoration keeps black-and-white and sepia photos uncoloured. - Photo v2 makes fur and hair clearer than v1, but at 4x from a small image it is not yet strand-for-strand like the original: detail fully gone from the input cannot be recreated; sharpness is still being improved model by model. > PixGamma is free for non-commercial use and runs entirely on the Mac. Sample images on this page come from Wikimedia Commons (CC0 and public domain).