Turn an Image Into a Line Drawing

Most guides about turning an image into a line drawing assume the image is a photo. A screenshot is not a photo, and neither is a logo, a piece of clip art, a scanned drawing or a dithered GIF. They behave differently in the same converter — sometimes dramatically. Here is what each one actually does, measured on the same pipeline, and the settings that work for it.

The one number behind every result

Every measurement on this page comes from the same two rules, both verified against the converter's own 3×3 Sobel kernels:

Two consequences you can use without doing any maths. First, almost everything in a non-photo image scores far above the slider. The Edge Detail slider runs 10–80, and a plain 1-pixel divider line that is only 30 levels darker than its background already scores 120. That is why Edge Detail feels broken on screenshots — the detail you want to remove is scoring ten times higher than the maximum setting.

Second, a smooth gradient scores almost nothing. A full black-to-white gradient spread across 800 pixels changes by 0.32 levels per pixel and scores 2.6 — invisible at every slider position. Squeeze the same gradient into 200 pixels and it scores 10.2 and crosses the bottom of the slider. That is the mechanism behind the gradient section below.

Screenshots and app windows

We built a synthetic 800 × 600 UI frame — a header bar, a light side panel, 1-pixel borders, a button, and three blocks of anti-aliased body text at roughly 10-pixel cap height — and ran it through the pipeline next to an 800 × 600 photo at the tool's default settings (Edge Detail 35, Smoothness 1, Line Width 1).

A screenshot produces nearly three times as much ink and about a hundred times as many separate shapes — 1,676 disconnected blobs where a photo gives 17. Those blobs are the letters. We re-ran the same frame with the text removed and subtracted:

About 90% of all the ink on a converted screenshot is the text, and Edge Detail cannot remove it. Dragging the slider from 10 to its maximum only takes the text contribution from 11.89% down to 8.36%, because anti-aliased glyph edges score in the hundreds while the slider tops out at 80. If you try to fix a busy screenshot with Edge Detail, you will strip the panels and buttons first and still be left with the words.

The control that works is Smoothness. On the same frame:

Smoothness 4 with Edge Detail 35 gives 0.91% — essentially the panel borders and the button outline, with the text gone. Note that raising Smoothness on its own makes things worse (at Edge Detail 10 it climbs from 7.71% to 19.27%), so the two sliders have to move together: Smoothness to kill the glyphs, Edge Detail to keep the blur from smearing the shapes that survive. Start at Smoothness 4 / Edge Detail 35 and lower both one notch if the page comes out too empty.

One more screenshot-specific quirk, measured on a single 1-pixel line on white: a divider that is only 15 levels darker than the background (score 60) still inks at Edge Detail 35 — and it comes out as two lines with a white gap between them, because the line's own pixel is a local peak where brightness stops changing. Setting Line Width to 2 closes the gap into one solid stroke. A divider only 5 levels darker (score 20) disappears entirely.

Logos and flat clip art

This is the easy case, and it behaves the opposite way from a screenshot. On a synthetic 800 × 600 frame of solid shapes with hard edges — the kind of thing an SVG export or a piece of clip art looks like:

Across the entire slider, ink coverage moves by 0.07 percentage points. On flat artwork, Edge Detail is inert. Every edge is a hard step scoring in the hundreds, so nothing you can set will drop any of them. If you want the outlines thinner or thicker, use Line Width: at Smoothness 1 / Edge Detail 35, Line Width 1 → 2 → 4 gives 3.12%, 4.87% and 8.10%.

Smoothness, meanwhile, does the same counter-intuitive thing it does to photos — it adds ink, because a blur spreads a hard edge over several pixels and each of those pixels still clears the threshold. At Edge Detail 35: Smoothness 0 → 1.84%, 1 → 3.12%, 2 → 4.39%, 4 → 6.03%. Leave Smoothness at 0 for flat artwork unless the art is actually noisy, and set Line Width 2 so the outlines print solidly.

Gradient backgrounds: an all-or-nothing trap

Logos and web graphics often sit on a gradient, and gradients are the one input where the converter has no middle gear. Because a ramp's score is 8 × levels per pixel, a full black-to-white gradient either scores below the slider everywhere or above it everywhere. There is no setting that draws a single line down the middle of it — you get nothing, or you get the whole band filled black.

Measured on a composite frame — a solid disc sitting on a gradient that completes in 204 pixels — the page carries 13.74% ink at Edge Detail 10 and 0.53% at Edge Detail 20, where only the disc outline survives. At 150 pixels it is 18.97% versus 0.53%. At 400 pixels the gradient never triggers at any setting.

So the practical rule is short: keep Edge Detail at 35 or above when your artwork has a gradient in it. At 35, a gradient has to complete its full black-to-white sweep inside 58 pixels before it misbehaves, which almost no real design does. Drop to Edge Detail 10 and the safe zone collapses to 204 pixels — a very ordinary background gradient.

Dithered, 1-bit and halftone images

Ordered-dither graphics, 1-bit GIFs and newspaper-style halftones are the worst input the converter can get. A Bayer 8×8 dithered gradient at 800 × 600, default settings:

At the tool's defaults the page is 60.49% black — more ink than paper, spread over 747 separate specks. Worse, every dither dot is a full black-to-white step scoring around 510, so Edge Detail does literally nothing at Smoothness 0 (37.61% at every position). Line Width makes it catastrophic: at Smoothness 1 / Edge Detail 35, Line Width 2 gives 91.22% and Line Width 4 gives 97.50% — a solid black rectangle.

You can force it down to 0.68% with Smoothness 4 and Edge Detail 35, but be honest about what that costs: at that blur radius the dither pattern and any genuine fine detail are gone together. Dithering is not really recoverable — find a smooth version of the source. If the only copy you have is a dithered GIF, converting it to smooth grayscale in any image editor first will do more than every slider on the page.

Scans and artwork that is already line art

This is the case where the converter's method works against you. It draws contours: the boundary where brightness changes. A black stroke on white has two such boundaries, one on each side, so a drawing that is already line art comes out outlined rather than copied.

Read the 3-pixel row: the drawing had 12.30% ink, the page came back with 17.61% — 1.8× as much, because every stroke is now two strokes. Read the 10-pixel row: a drawing with 21.82% ink comes back with 5.57%, and 85.4% of the original stroke area prints white. A thick marker line is replaced by a thin empty outline of itself.

Line Width partly repairs the thin case. On 3-pixel strokes, the hollow share falls from 30.3% at Line Width 1 to 12.7% at 2, 10.4% at 3 and 8.8% at 4 — the dilation closes the gap between the two contours back into one solid stroke. It cannot rescue thick strokes: at 10 pixels, Line Width 4 still leaves 73.1% hollow.

So: do not convert artwork that is already black and white. Print it directly, or use it as the input to Picture to Color by Number, which is built to fill flat regions rather than trace their edges.

Transparent PNGs: why you get a blank page

This one surprises people, and it has a precise cause. The converter's first step is gray = 0.299 × R + 0.587 × G + 0.114 × B. It never reads the alpha channel. A canvas starts out transparent black, so pixels your PNG leaves transparent arrive as R = G = B = 0. If your artwork is black lines on transparency, every pixel in the frame ends up at 0 — a completely flat image with no brightness change anywhere — and the edge detector returns nothing.

A blank sheet. The fix is one step, outside the tool: paste the PNG onto a white background in any editor, or export it without transparency, and it converts normally. The same trap catches any image with no edges at all — a flat mid-grey fill measures 0.00% ink even at Edge Detail 10, because there is nothing for the detector to find.

Small images, thumbnails and crops

The converter caps its working size at 1,400 pixels on the long side, but it never scales up. A 320 × 240 thumbnail is processed at 320 × 240, and two things follow from that.

First, the page gets denser. The same synthetic clip-art frame rendered at 800 × 600 and at 320 × 240, identical settings:

2.4× as much ink from the same picture, because each outline takes a bigger share of fewer pixels. Start one notch higher on Edge Detail and one lower on Line Width than you would for a full-size image.

Second, it prints softer. On the tool's US Letter page — 215.9 × 279.4 mm with 14.1 mm margins, giving a 187.7 × 251.2 mm image area — a 4:3 image is fitted across that width:

Line Width dilates by a fixed number of pixels, so one click is 4.3× heavier on a 320-pixel thumbnail than on an image at the 1,400-pixel cap. If you are thickening outlines for young children, judge it at full resolution — a test on a small crop will lie to you.

Heavily compressed JPEGs

Block artifacts are a different failure: they sit at the very bottom of the score range and flip with a single notch. On a synthetic frame with 8×8 block quantisation producing 12-level steps:

At Edge Detail 10 the block grid covers 87% of the page; at 35 it vanishes completely. Artifacts of that size score around 48, which puts them inside the bottom third of the slider and nowhere else. If a compressed image comes out as a grid of squares, one or two notches up on Edge Detail deletes the grid — and Smoothness 4 does the same job at any threshold, which is the safer move when the image also has real detail worth keeping.

A working order for any non-photo image

1. Decide whether it is already black and white. If it is, stop — print it, or turn it into numbered regions instead of outlines.
2. If it has transparency, flatten it onto white first. Nothing inside the tool can recover it.
3. Set Smoothness by noise, not by type. Clean vector art: 0. Screenshot with text: 4. Dithered or heavily compressed: 4. Everything else: 1.
4. Then set Edge Detail. 20 for flat art, 35 as the safe general value, 50 or more if the artwork has a gradient background. Below 35, gradients and compression blocks can both take over the page.
5. Then Line Width. 2 for printing, 1 if there is fine detail you want to keep, and never higher than 2 on an image smaller than about 700 pixels across.
6. Change one thing and look again. Too black → Smoothness first, then Edge Detail. Too empty → Edge Detail down. Outlines too faint → Line Width, nothing else. Still unusable after three tries → the source is the problem, not the settings.

What to do with the result

Print the page as it is, or take it further. If you would rather have numbered colour blocks than open outlines, the same image goes through Picture to Color by Number. A pile of images becomes one PDF with Coloring Book Maker, one drawing per page — which is also how you get a set of screenshots or logos into a single printable file. For a page that is not an image at all, Name Coloring Page Generator draws any name in big outlined letters.

The walkthrough for ordinary photo conversion is on Convert a Picture Into a Drawing, the settings broken down by photo subject are on Turn Photo Into Line Drawing, and the number underneath the Edge Detail slider — including why it fails on grainy photos — is on Photo to Line Art. Page sizes, margins and resolution for printing are covered on Coloring Page Generator.

Questions

Is turning an image into a line drawing different from turning a photo into one? The pipeline is identical — grayscale, blur, edge threshold, thicken, despeck. What changes is the input: photos are mostly gentle gradients that score near zero, while screenshots, logos and dithering produce hard steps that score in the hundreds. That is why the recipes differ even though the tool does not.

Does my image leave my device? No. Every stage runs inside your browser tab. Nothing is uploaded, there is no account, and there is no server copy of your file — see the Privacy Policy. It also means the conversion speed depends on your own machine rather than a queue.

Can I convert a screenshot of a website? You can, but expect the text to turn into ink blobs rather than letters, and expect the result to be far busier than a photo at the same settings. Smoothness 4 with Edge Detail 35 is the only combination we measured that reliably strips the text while keeping the layout shapes.

Why is my page completely blank? Two causes cover almost every case: the image has transparent areas (flatten it onto white), or it has no brightness change at all (a flat colour fill, or an extremely soft gradient). Both measure 0.00% ink regardless of slider position.

How reproducible are these numbers? All of them come from running the converter's own pipeline — the same grayscale weights, Gaussian blur, 3×3 Sobel kernels, dilation and despeck pass shipped on the site — over synthetic frames built to model each image class, plus one 800 × 600 test photo. The pipeline and the measurement scripts are open source on GitHub (coloringzap-algorithms). Your images will land at slightly different percentages; the structure — which slider moves which class, and by how much — is what carries over.

Is there a limit? No daily limit, no watermark, no sign-up. Convert as many images as you want.