OBSCURA Photo Editor

In brief

• Layers, masks and adjustment layers Blend modes, groups, clipping, layer styles, smart objects with smart filters, and artboards
• Retouching Clone, healing, spot healing and patch on their own empty layer, sampling the layers below, content-aware move and fill, liquify, warp and perspective
• Selections Quick selection, magnetic lasso, colour range and Select and Mask, and Select Subject and Remove Background by a small model that runs in the tab
• Camera raw CR3, NEF, ARW, RAF, DNG and 22 more raw formats, decoded in the tab at sixteen bits a channel and developed with a raw converter’s sliders, colour mixer, colour grading, curves and local masks
• Photoshop PSD and PSB in and out Layers, groups, masks, clipping, guides, the common adjustment layers, and text, shapes and layer styles that stay editable, read from 8 or 16-bit files in RGB, greyscale or CMYK and written back; Photoshop’s brushes, patterns, gradients, styles and actions imported
• A library A folder of photos with ratings, colour labels, keywords and each photo’s develop settings kept in XMP sidecars that Lightroom, Bridge and exiftool read, settings copied and synced across many, and contact sheets and print layouts as PDF
• Colour management Embedded ICC profiles read, a working space of sRGB, Display P3, Adobe RGB or ProPhoto RGB, proofing on screen with a gamut warning, and 16-bit TIFF, PNG and PSD or CMYK TIFF with the profile embedded
• Kept on this device Autosaved as you work and offered back after a crash; it works offline once opened

What survives a trip through Photoshop

A PSD or PSB is read and written here from the published format, byte by byte. Text, shapes and layer styles are kept as Photoshop keeps them, as descriptors, which ag-psd, an open-source library this site serves from its own address, reads and writes. The pixels in the file and the flattened picture at its end are this editor’s own rendering, so a program that does not draw text and effects itself shows what was on screen; Photoshop sets the words again with its own fonts when they are edited, so the text can move a little if a font differs.

What it cannot do yet

• Sixteen bits only for a photo and its Develop layers. The sixteen bits of a raw file, a 16-bit TIFF or PNG, or a flat 16-bit PSD go through to a 16-bit export while the picture is that photo with Develop layers over it. Any other layer, painting on the photo, a crop or a resize, or a layered 16-bit PSD puts the picture back to eight bits a channel, and the export dialog says which it is writing and why. The screen shows eight bits; CMYK is for proofing and TIFF export, not a mode to edit in.
• Large pictures have a ceiling. A picture, a raw file, a PSD, a TIFF or a project larger than 40 megapixels opens scaled down to fit (a 61-megapixel photo at about 81 per cent of its size), up to 150 megapixels, and the status line says its own size and the size it opened at; the file is not changed, and nothing is saved at its full size. A PDF page is drawn at a lower resolution to stay under it. The graphics card, through WebGL2, does the blend modes the browser’s canvas lacks, hue and saturation, the adjustments that mix colours, and the Gaussian blur, each held to the editor’s own answer within a level or two and used only where it is the quicker on your computer. In our tests on a 6-megapixel picture, with WebGL drawn in software rather than on a graphics card, the ten extra blend modes ran between 1.7 and 2.9 times as fast as in JavaScript over several runs, hue and saturation between about as fast and twice as fast, and the blur slower, so there the blur stays in JavaScript; on a real graphics card we have not measured it. The rest is the canvas and JavaScript, so a very large picture with many layers is still slower than in a desktop program.
• Select Sky is found by rule, not learnt. A sky the same colour as what meets it, such as white sky over snow, runs on into it, and a distant hazy hill touching a small part of the sky can be taken with it; check the edge with Select and Mask.
• No pictures fetched from other sites. A picture dragged from another OBSCURA tab opens; one dragged from another website has to be saved first, or copied and pasted, since the editor makes no requests to other sites.
• The library needs Chrome or Edge on a computer for a folder. Firefox and Safari have no File System Access API, so there you choose files instead, and their sidecars are downloaded as a zip to put beside them. The library reads a folder each time it is opened; it is not a catalogue of a hundred thousand photos. Virtual copies are kept in the photo's sidecar, so they go where it goes; picks, collections and smart collections have no place in a sidecar, so they are kept in this browser for each folder, and another browser does not see them.
• Develop settings in Lightroom. The settings are written under Camera Raw’s names, so Lightroom and Camera Raw read the sliders, the mixer, grading and the curves; the arithmetic behind each slider is each program’s own, so the picture is similar, not identical. A white balance Lightroom wrote in kelvin is measured from the raw file’s own as-shot reading through its colour matrix, which can differ a little from Lightroom’s, whose camera profiles are Adobe’s own; on a picture that is not a raw file it is left out, and the report says so. The masks are kept where only this editor reads them; a subject or sky mask is kept by its kind and found again on the photo when it is opened from the library.
• Lens profiles are lensfun’s, not Adobe’s. Develop’s Lens section corrects distortion, colour fringes and vignetting from the lensfun database (CC BY-SA 3.0), about 1,500 lenses measured by its contributors and served from this site. A lens it does not have is corrected by hand in Lens Correction, and a fisheye is not corrected.
• Sidecars and JPEGs. Lightroom Classic reads a sidecar for a raw file; for a JPEG it reads the XMP inside the file, which the library reads but does not write. Bridge, darktable, digiKam and exiftool read the sidecars; the library writes photo.xmp, as Adobe’s programs name them, or photo.jpg.xmp for darktable and digiKam.
• The Filter Gallery’s effects are models. All forty-seven of Photoshop’s are here, but Adobe does not publish how they work: each is a model that does the same job, so the result is like Photoshop’s, not the same.
• Photoshop actions in part. An .atn file brings in the steps this editor has a counterpart for, fifty-seven kinds of them, such as Levels and Curves channel by channel, Colour Balance, Gaussian Blur, Motion Blur, Unsharp Mask, Hue/Saturation, Image Size and Canvas Size, to a size or by so much, anchored where it was; the rest, such as a selection made by hand, are named when it is imported and left out.
• Puppet Warp pins cannot be turned or stacked. A pin moves; it cannot be rotated by hand or brought in front of another part of the mesh, as Photoshop’s can.
• Vanishing Point has one plane. Clone along it and paste onto it; planes joined round a corner, such as two walls of a building, are done one at a time.
• Lens Blur’s depth is not measured. It is made from the subject: the subject kept near and sharp, the rest receding up the picture. There is no model here that estimates depth from one photo.
• Photomerge does not correct the lens. Frames are matched by their corners and laid in perspective or on a cylinder, with the exposure evened out, but a lens’s barrel distortion is not taken out first, so with a wide lens the seams can show; correct each frame in Lens Correction before merging. A person who moved between frames can appear twice near a seam.
• Develop fills the window, but is not a module. F gives Develop the whole window, with the picture, its histogram and every section, but it is still one adjustment on the picture, with the editor’s own history; a shoot is gone through with the filmstrip under the picture, not in a module of its own as in Lightroom.
• PDF and Illustrator files are opened as a picture. A page is drawn at the resolution you choose, up to 40 megapixels, onto one layer; its text and paths are not kept as text and shapes. An Illustrator file saved without PDF compatibility is refused, with the reason.
• Text in a PSD. Mixed styles within one text layer come in as the first style; a warp the editor does not have is left off; gradient strokes are drawn into the pixels.
• No Photoshop .8bf plugins. They are compiled Windows and Mac programs that no browser can run; plugins here are written in JavaScript.
• Nothing generated from a prompt. Nothing here sends your picture to a model on a server; the one model it uses finds the subject and runs in the tab.

Compared with what you may use now

Against Photoshop, every tool and panel and what is missing; against Photopea, which also runs in the browser and is free with adverts. If you have a photo from an iPhone, how to open a HEIC file.