Free online color separation for screen printing

What "free" means here, precisely

Spot separation and index separation are free, forever, with no account needed to run them. Signed out, an image is fitted to 8 megapixels; a free account removes that account-tier cap and is required to download, while the editor can still fit unusually large art to the memory ceiling of the current device. Ink count and tolerance, merging and renaming inks, reordering the local plate stack, per-plate preview and the proof over your garment colour remain free. The generated plate layers can also be brushed in the document and taken through the editor's ordinary document exports; the Colours-panel plate and spot-channel writers continue to use their raw separation masks. After sign-in, per-ink PNG, ZIP and ORA downloads are unlimited, never watermarked and written at the working canvas size without another export-tier downscale. The two PSD formats share one free export a day: a layered PSD with one named layer per ink, or a multichannel PSD carrying one real spot channel per ink with its ink colour and name attached. A monthly plan removes that PSD cap; without one, each later PSD costs 3 credits. Quantity is capped, quality never is. Those files are written client-side and no image pixels are uploaded; the account gate and anonymous outcome beacons send metadata only. The paid half is the print-ready render, and it is print maths at output resolution: halftone screening with separated angle sets, a choked underbase, a separate highlight white, dot-gain compensation, registration marks, 1-bit film targeting 600 or 1200 dpi within the renderer's disclosed 16,800px/200MP output caps, optional SVG tracing for suitable plates, and CMYK and simulated-process separation. When a local Spot or Index result is retained on screen, its exposed choke/spread values feed a Spot, Index or Simulated-process render, and its whole-channel density values feed Simulated process. A direct paid render instead lets the server choose its palette and returns an opaque ZIP; it does not populate editable channels in the panel. Those values do not modify the local preview or local files, and the shipped workspace does not expose per-channel CMYK choke or density. There is no separate automatic-trapping control; use per-ink spread where the exposed non-CMYK plate controls need intentional overlap. The render genuinely needs a server, because a 14-inch plate targeting 1200 dpi is 16,800 pixels along that dimension. It is 10 credits per image, and re-renders of that image are free for 24 hours. The boundary is explicit: a working-canvas local separation with the base heuristic and raw masks described above, or server-rendered print maths with its own limits and manifest.

What color separation actually requires

Every ink in a screen print needs its own screen, which means your flat design has to be split into one artwork layer per color: every pixel of that color on one plate, no matter where it appears in the design. Traditionally that means Photoshop channel gymnastics, or dedicated desktop software. The established options are all installs: Separation Studio NXT is a standalone Windows/Mac application on a yearly subscription, while UltraSeps and T-Seps are Photoshop plug-ins bought outright, which means a full Photoshop licence underneath them. They are good tools and they earn their keep in a shop doing daily simulated process work. But they are also a purchase, an install, and a computer capable of running Photoshop — before you have separated anything at all.

One button: Split by color

Layersmith reads your design, finds the distinct ink colors, and lifts each one onto its own layer — with a human-readable ink name and the exact hex kept beside it, so the layer panel reads like a screen list. And it groups by color, not by connected shape: every letter of the same cream lettering lands together on one plate, exactly as it will share one screen. One behaviour to know: the free engine keeps the largest colour area as the base rather than cutting a screen for it, on the assumption that the biggest flat field is the garment. On flat art that is right; on a photograph, check what landed in the base before you trust the plate count.

A three-ink job, separated reproducibly

Below is a worked regression proof, not a mock-up. The committed scripts/make-seps-proof.mjs bundles and calls the same app/src/lib/pipeline/seps.ts shipped in the Colours workspace, feeds it the 4,200×3,000 source above, and writes the masks the engine returns. The result is three ink plates — Blue #1e3ea2, Cream #f2e6ca and Red #ca1e2a — plus Black #121212 as the base rather than a screen. The cream plate is the one to inspect: all 19 disconnected letters share one screen. Every plate and film positive below is a lossless PNG; the accompanying manifest records the source and engine hashes so npm run test:seps-proof fails when the checked-in proof drifts.

And the multichannel PSD, checked by something that is not us

The multichannel writer has a separate four-channel regression fixture named Red, Navy, Gold and Cream. The test suite parses the bytes itself and reads them back through ag-psd and psd-tools, checking multichannel mode, real spot records, names, ink colours, dimensions, DPI, compression and polarity. The garment base is deliberately absent, because the substrate is not a screen and nobody should burn one for the shirt. That writer fixture is not the plate artwork pictured above, and Photoshop has still not opened it in a hands-on test; until that happens, independently parsed is the claim we can support.

Export formats for the next tool

Three shapes of file, all written in your browser. A layered PSD: one named layer per ink, every plate at full canvas size and in register, for a conventional layer editor. A multichannel PSD: one real spot channel per ink, each carrying its ink colour and name, at the physical DPI implied by the print width you set; the garment base is deliberately left out so nobody burns a screen for the shirt. Or a ZIP of per-ink transparent PNGs and film positives, numbered in stacking order for the widest compatibility. There is no built-in RIP or direct-print integration: your RIP, image editor or driver takes the appropriate file from there. The multichannel structure is covered by independent parsers, but the outstanding Photoshop hands-on test is stated above.

Halftones, underbase and simulated process

These are no longer roadmap items. The Color separation tab handles all four separation types — spot, index, CMYK, and simulated process — and the print-ready render adds the print math on top: halftone screening with separated angle sets (your choice of frequency, dot shape, and output DPI), a choked underbase, a separate highlight white printed last, a dot-gain compensation curve, registration marks, and optional vector-traced plates. A retained local Spot or Index ink list can also send per-ink choke/spread into a Spot, Index or Simulated-process render; the shipped CMYK workspace does not expose per-channel choke or density. Spot and index start with halftoning off; selecting CMYK or simulated process turns it on. Turn it off when your RIP should receive continuous-tone grayscale plates, or leave it on for pre-halftoned 1-bit film.

Where the manual density adjustment actually appears

Fully automatic sim-process is unreliable in every tool on the market, so the paid renderer accepts a custom ink list with a whole-channel density value for each ink. In the current workspace those sliders appear only when you first make a free Spot or Index result and keep that local palette on screen before switching to Simulated process; a direct Simulated-process render lets the server choose its palette, returns an opaque ZIP, and does not populate editable channels afterwards. The retained-palette path lets you pull each ink back or restore it toward full strength, set per-ink choke/spread, set underbase choke, and turn highlight white on or off. It still does not expose a per-channel tone-curve editor or an adjustable highlight threshold, so a difficult photograph can need a dedicated separator or downstream channel work.

Priced against what separations actually cost you

The alternative most shops use for the awkward jobs is an outside separator, and those are per-design fees. Published prices, checked 28 July 2026: ContractPrinter lists $15 for one or two spot colours, $30 for three to five, $55 for six to eight, and $85 and up for nine-plus or photorealistic simulated process, with $5 for each revision; Victory Screen Factory lists $18.99 for two colours rising to $78.99 for eight. That $85 is the top of the field rather than the going rate for photographic work, and it is worth knowing the flat-fee services exist: checked 29 July 2026, Burn My Screen lists a flat $60 per design, 24 Hour Seps — Scott Fresener's own service — lists a flat $65 per set for any number of colours (a promotional rate when checked; their volume packs go lower), and Color-Separation.com lists $15 to $50 including the underbase and highlight white. So realistically, outsourcing a hard sim-process design runs somewhere in the $50 to $85 band, and a simple two-colour job runs $15 to $19. Turnaround field-wide is commonly a day, sometimes two, though neither ContractPrinter nor Victory publishes one. Here, spot and index separation cost nothing at all, and the print-ready render is 10 credits per image, with free re-renders of that image for 24 hours instead of a revision fee and a fresh wait. Credits come with a monthly plan, or as a one-off pack whose credits never expire — see the pricing page for current prices in your currency. For a genuinely difficult photograph on a dark garment, a skilled human separator is still worth the money; for the everyday spot-colour job list, paying per design is overhead you can stop.