LEARN

Every feature, in plain steps.

Short and practical — each guide is the exact click path. The editor has three tabs: Auto (one click, done), Manual (every tool, by hand) and Colours (separate into inks). Inside the editor, hover any button for a full explanation, and press ? for keyboard shortcuts.

The free browser tools need no account to USE and work at up to 8 megapixels — a free account (Google, Apple or an emailed code — no card) is what lets you download the finished file lifts that to your picture’s full resolution. The server-AI tools (marked with a credit cost) unlock the moment you sign in — you get 14 free credits, no card required. Run out and a one-off pack of 50 more is a single purchase, no subscription — see pricing for the current price in your currency, and the monthly plans if you're here every week. HD generative fill is the one exception: it runs on a paid engine, so it needs credits you've bought.

Auto: one click, doneYour first layered PSD by handSeparate a subject from its backgroundColours: separate artwork into printable inksTurn flat text into editable text layersMake a small image big enough to useFix a selection by hand

Auto: one click, done

The tab you land on. Pick a job, choose a photo, and it runs as it opens.

  1. Remove background (free). Cuts the main subject out to transparency, entirely in your browser. No account, no upload, no credits — the image never leaves your device.
  2. Remove people · 3 cr+. Finds every person in the shot, masks each one with our strongest model, erases them and fills the hole behind. You can review who was found and untick anyone you want to keep, so it is never all-or-nothing.
  3. Erase one thing. Click the object or person you want gone. The grab is free; you only spend credits if the hole needs a server fill.
  4. Split into layers · 1 cr+. The differentiator: every object lifted onto its own named layer ("person", "chair", "dog"), stacked exactly where it already was. On a photo with no distinct objects — a landscape, say — it falls back to splitting by region instead, so you always get layers back.
  5. It looks like nothing happened — check the Layers panel. A successful split is a lift in place, so the picture is pixel-identical before and after. The Layers panel on the right is the evidence, and it opens itself when a run succeeds.
  6. "This picture hasn't been split yet". If the export panel says this and the layered formats look dimmed, the file you would download has exactly one layer — you have opened a picture but not run anything on it yet. The two buttons in that message run the jobs that make layers. The exports still work if a flat file is genuinely what you want.
  7. The first server job of a visit takes longer. Our AI engines start on demand rather than idling at your expense, so the first job after a quiet spell spends about 15 extra seconds waking one up. The status line counts the seconds so you can see it is working; everything after that is at full speed.

Your first layered PSD by hand (2 minutes)

Switch to the Manual tab when you want to choose exactly what gets cut.

  1. Open your image. Drag any JPG, PNG, or WebP onto the editor, or click the drop zone. Free tools run entirely in your browser — the image never uploads.
  2. Pick AI grab, then click the object. Manual opens on "Select — mark it by hand". Press A (or click AI grab), then click once anywhere on an object — the model finds its outline. The pink tint shows your selection.
  3. Lift it to a layer. Under "Cut — lift it out & fill behind", press "Lift selection (cut)". The object becomes its own layer; leave "Clean edges on lift" ticked to remove colour fringing at the cut.
  4. Export. Under "Finish", or from the right panel: "Layered .psd". It opens in Photoshop, Photopea, Affinity, GIMP and Krita with named layers, groups, and a real layer mask on every lift.

Separate a subject from its background

One button instead of clicking — best for portraits, products and pets.

  1. Pick your engine. "Separate subject" (free) runs in your browser — first use downloads the model, ~40 MB. "Separate subject · 1 cr" (server) uses our strongest model: noticeably better hair, fur and fine edges.
  2. Press it. You get a Subject layer on top and the background underneath — nothing else to do.
  3. Complete the background. The background now has a subject-shaped hole. Over sky, gradients or a plain wall it is filled free in your browser, with no ghost. Only holes sitting on real texture need "Fill behind · 2 cr" — and we tell you which ones rather than charging you quietly.

Colours: separate artwork into printable inks

The third tab — for screen printers, and for anyone who wants flat colour as flat layers.

  1. Separate (free). Choose Spot (logos, flat art, text — one screen per colour) or Index (photos, fewest screens, dots with no angles to line up), set the ink count, and press Separate. It runs in your browser and applies the plates to your document as named layers — "Cyan", "Warm white", "Light vermilion", not hex codes.
  2. Tune it. Rename any ink, adjust choke/spread, set halftone frequency and angles, or let the angles pick themselves. Re-separating replaces the plates rather than stacking a second set on top.
  3. Print-ready render · 10 cr. Charged per IMAGE, not per render — once an image is unlocked, re-rendering it with different halftone or choke settings is free for 24 hours. Slide the sliders as much as you like.
  4. Take it to the shop. Export a multichannel .psd with a real named spot channel per ink, or "Plate PNGs + film · ZIP" — one film per screen, with a printer-readable README in the zip.

Turn flat text into editable text layers

Posters, memes, screenshots — recover the text as text.

  1. Detect text (free). Press "Detect text" — OCR runs in your browser and boxes every line it finds. No account, no upload.
  2. Make it editable + remove the original. Press "Make N lines editable + remove original". It creates real PSD text layers AND erases the original pixels in place. Text over sky, gradients or solid colour is erased free in your browser (no ghost); only lines sitting on a textured photo use a server fill — up to 2 credits.
  3. Edit in your tool. The exported PSD carries them as real text layers — change the words in Photoshop or Photopea. (Already lifted a text layer by hand? "Lift → editable text" turns that layer into type.)

Make a small image big enough to use

The free upscaler at /editor?tool=upscale, and why it does two completely different things depending on what you drop in.

  1. Flat artwork — free, unlimited, never uploaded. Logos, lettering, vector-style graphics. These are enlarged right in your browser: instant, no account, no daily limit, and the file never leaves your machine. We do NOT send these to the AI, because a photographic model invents texture — measured on a clean logo, a flat red went from almost perfectly even to visibly uneven. A plain enlargement keeps the colour honest, and for flat art that is simply the better result.
  2. Photographs — the AI, 2 free a day signed out. Grass, fabric, skin, brickwork. Here the model earns its keep: it rebuilds detail rather than stretching pixels, roughly doubling edge definition with no halos. It takes 10–30 seconds because it runs on a server, not your device — which is also why it works on a phone, unlike the desktop upscalers that need a dedicated graphics card. Sign in (no card) and that rises to 3 a day at double the resolution.
  3. Past the free runs · from 1 credit. 1 credit up to 8 megapixels, 2 up to 16, 4 up to 32 — priced on the size of what comes out. Paying also raises the ceiling: free runs cap at 8 megapixels, paid go to 32, which is a 16×20 inch print at 300 dpi.
  4. Printing it?. Print wants about 300 dpi at final size, so a 12-inch print needs roughly 3,600 pixels across — the result tells you what you actually got. If you are separating art for screen printing, do it in the Colours tab instead: it upscales under-resolution art as part of the job, and the colour drift above cannot reach the plates, because every pixel ends up assigned to one ink anyway.

Fix a selection by hand

When AI gets 95% of the way there.

  1. Add with the Brush (B). Turn on "Add to selection" and paint the missed areas.
  2. Remove with Erase (E). Paint over anything that should not be selected.
  3. Wand for flat areas (W). Selects connected same-color pixels. Tolerance controls how loose the color match is; "Contiguous" off selects that color everywhere.
  4. Invert when it is easier backwards. Selecting a plain background then pressing Invert is often faster than selecting the subject.

All guides

Task-specific walkthroughs — pick the one that matches what you're making.

Colour separation 11

Layered files 12

Text and type 3

Objects and subjects 3

Upscaling 3

Comparisons 5

Workflows 6

Fastest way to learn: try it.

Open the editor