RetouchPi File Formats: What It Opens and What It Exports

• Bitmaps the browser decodes: PNG, JPEG, GIF, WebP, BMP, SVG and AVIF. This family goes through the browser's own decoder, so which of them open depends on the browser you are in.
• PSD: read with its layers - layers, masks, blend modes and layer styles all come in - and written back as a layered PSD.
• TIFF, including 16-bit channels and CMYK scans. Sixteen-bit narrows to eight on read, and CMYK converts to sRGB on read - see "Colour and bit depth" below.
• HEIC / HEIF, through a decoder of its own rather than the browser's. The files an iPhone writes open directly, with no conversion on the phone first.
• Camera RAW: extensions and magic numbers for twenty-odd makers. DNG is genuinely developed - lossless JPEG (SOF3) sensor data, demosaiced, using the colour recipe the file itself carries - and that includes phone-written DNG such as ProRAW. For other makers, what opens is the JPEG preview the camera embedded, and the status line says so out loud.
• EPS: what opens is the TIFF preview the file carries, which is how the mainstream layout and vector programs have written them for thirty years. There is no PostScript interpreter here, so a bare PostScript EPS, or one whose only preview is a Windows metafile, is refused by name rather than opening as a blank canvas.
• Design files (Sketch, XD, Figma, Affinity): what opens is the flattened preview those files keep for software that cannot run the format, not the document - and the status line says the copy came in flattened. Reading them properly means writing a second implementation of somebody's application; half-reading one means showing you a document that is not the file.
• Project files (.wcproj.json): the editor's own save, carrying layers, masks and styles, so saving and reopening gives you the same document back.

The export half of the File menu is grouped by what the file is rather than laid out flat: four raster formats, two exits that are not "the same pixels in another wrapper", one that moves, and three that are specifications in their own right. Add saving a layered PSD and saving the project, and there are twelve ways to get something out. Exports carry no watermark.

• Images: PNG, JPEG, WebP, TIFF.
• Page and vector: a one-page PDF (one pixel to one point, laid out at 72 dpi, so a 1200x800 picture is 1200x800 points on the page); and SVG, which keeps whatever is still a vector and lists every loss in the dialog before you commit rather than flattening quietly.
• Video: MP4, recorded by the browser from the canvas in motion (a camera move and the like), with the long side capped at 1280.
• Specifications: ID photo sheets (six named sizes, all at 300 dpi - 25x35mm one-inch, 35x49mm two-inch, 22x32mm small one-inch, 35x45mm small two-inch, 33x48mm passport, 51x51mm square); slices; and artboards.

PDF is write-only: this version can produce a one-page PDF but cannot read one. SVG is close to the same - a browser will open an SVG as a picture, and the SVG export is the one path that hands vectors back to you.

There are two working colour modes, sRGB and Lab - those are the only two rows under Image ▸ Mode. That is not an omission, it is the real boundary of this version, and it decides three things:

• CMYK and grayscale files open, but they are converted to sRGB at the moment of reading. What you see is the sRGB rendering of that picture, not the CMYK channels themselves.
• A file with 16-bit channels narrows to 8 bits on read. The file opens; a high-bit-depth pipeline is not in this version.
• Exports are sRGB. That is right for a photo lab, for the web and for social; it is not a press workflow - if the job goes to a four-colour printer and needs plates and a proofing step, this version is not that tool.

A raster is four bytes a pixel and lives in the wasm heap; undo keeps a copy per step, and compositing keeps document-sized scratch on top. An editor that opens anything it is handed is an editor that loses the tab - and the browser kills the page without a message, taking the unsaved document with it. The policy here is one hard ceiling, and below it nothing is refused:

• Hard ceiling: 64 megapixels, roughly 8000x8000, plus a longest side of 30,000 pixels - a panorama is small in area but its shape breaks resampling kernels that plan a row at a time, so the side gets its own limit.
• Below the ceiling nothing is refused: a picture bigger than this device can comfortably hold is offered a scale-down rather than rejected. A twelve-megapixel phone photograph is not turned away for being twelve megapixels.
• Undo keeps up to 256 steps and shrinks with available memory, down to a floor of two. The bigger the document, the fewer steps back you get - that is a memory policy, not an arbitrary number.

Decoding, editing and export all happen in the browser on this machine: the file comes off disk, the pixels change in wasm, and the export is produced locally. Only two things touch a server - AI generation and edits, and the account and credit ledger.

• An AI edit uploads the source image, fitted to 512..2048 per side with a 10MB ceiling, and the result is deleted from the server after 14 days. Prices and the retention window are on the pricing page.
• Editing needs no account and nothing installed: the main bundle is 845KB (272KB gzipped) and the landing page prefetches it while the browser is idle.
• There is no offline mode in this version: no service worker and no app manifest, so a reload without a network will not bring the editor up. A document already open is unaffected until it needs the network again.

• Fifty-four tools in the toolbar: selections (marquee, elliptical, lasso, polygonal, magnetic, object selection, quick selection, magic wand), crop and perspective crop, brush and eraser, healing and cloning, liquify, puppet warp, paths and vectors, type, measuring, slices, and artboards.
• Ten layer styles, in the order the mainstream editors list them: drop shadow, inner shadow, outer glow, inner glow, bevel and emboss, colour overlay, gradient overlay, pattern overlay, satin, and stroke - plus the blending options row, eleven rows in all, and all ten styles paint.
• On the layer side: groups, masks and quick mask, clipping masks, smart objects you can go back into, layer comps, adjustment layers, and the channels and paths panels.
• ID photos are a pipeline rather than a crop preset: six named sizes or your own millimetres, one shared composition judgement (12% headroom, eyes at 45%, shoulders at 85%), a cutout that lands as a mask, and then a sheet laid out at 300 dpi.

• No CMYK working mode: you cannot edit in four colours, cannot produce plates, and there is no press-proofing step. CMYK files open - as sRGB.
• No 16-bit pipeline: high-bit-depth files narrow to 8 bits on read.
• No PostScript interpreter: a bare PostScript EPS is refused by name rather than opened on a best-effort basis.
• No interpretation of Sketch, XD, Figma or Affinity document structure: what you get is the flattened preview those files carry.
• No PDF reading (it writes one page); no development for non-DNG camera RAW (what opens is the camera's embedded JPEG, and the status line says so).
• No offline mode, no desktop client, and no batch entry point - there is a scripting dialog inside the editor, but nothing that takes a command line.