There are two genuinely different ways to turn a photo into a line drawing: an edge-detection filter that traces what is there, and a generative model that redraws the subject as an artist would. They produce different things and people are often disappointed because they picked the wrong one. This page shows both, with our own runs.
Enter a prompt on the left to start generating your AI image.





Not better and worse. Different outputs, and the choice should follow what you need the drawing for.
Canva, Photoshop's sketch filters and most free converters detect edges and threshold them. Fast, free, perfectly reproducible, and it works on anything including buildings and products. It also traces noise as though it were line work, so busy backgrounds turn to scribble and soft edges disappear entirely.
The model understands that it is looking at a face and decides which lines carry meaning. You get varied line weight, deliberate omission and clean whites. The trade is that it invents: a strand of hair may not be your strand of hair, and no two runs are identical.
If you want a faithful trace for a colouring page, cutting file or technical use, take the filter. If you want something that looks hand-drawn and would survive being framed, take the generative route.
The generative method, start to finish. Under a minute.
Side light or window light gives the model form to describe. Flat frontal flash gives it almost nothing, and the result comes back as an outline with an empty middle.
"Pen and ink", "graphite", "charcoal", "engraving" and "one continuous line" produce genuinely different drawings. The single word you pick here changes more than every other setting combined.
"No grey wash, no shading gradients" is what keeps it a line drawing. Without it, models drift toward soft pencil rendering that reads as a greyscale photo with texture on it.
"Preserve the subject's facial structure and expression exactly." Left out, the drawing is skilful and of somebody slightly different, which is the failure people notice last and mind most.
Line density varies between runs, as the four portraits above show. Generate a few and pick the density you want rather than trying to specify it numerically.
Pre-filled in the generator above and printed here so you can edit it.
"Redraw this photograph as a fine pen-and-ink line drawing: confident cross-hatched linework, pure black ink on white paper, no grey wash and no shading gradients, preserve the subject's facial structure and expression exactly, vary line weight to describe form."
"Vary line weight to describe form." It is the difference between a drawing and a trace. Uniform line weight is exactly what makes filter output look mechanical, and asking for variation is how you avoid it.
"Fine pen-and-ink" — replace it with "soft graphite pencil", "charcoal on toned paper", "single continuous contour line" or "engraved etching". Change only this and you can see exactly what the medium word does.
Measured across our runs, best to worst.
Faces are what these models understand most deeply. Structure, expression and hair direction all carry through, and this is the subject where the generative route beats a filter most clearly.
The family example above holds four faces with individual expressions. Expect quality to fall away past five or six people as faces at the back simplify.
Fur direction comes out well. Say the breed, or you get a generic animal of approximately the right shape drawn very competently.
Straight lines and repeated structure are where generative models invent windows. An edge filter is both more accurate and free for this case, and we would rather say so than sell you the wrong tool.
per run, generative method
~20s
edge filter, and free
Instant
free credits on signup
50
to land the line density you want
3–4 runs
The same redraw idea in a different tradition — flat colour instead of ink.
Where the engraved family portrait above came from.
Repair the scan first — damage gets traced as line work otherwise.
Clean line art from a description rather than from a photo.
Including the questions where the honest answer is "use something else".
Yes, and it works differently from a filter. A filter traces the edges it detects; a generative model recognises the subject and decides which lines matter, giving varied line weight and deliberate blank space. For portraits the difference is obvious side by side.
It can, using its sketch and edge effects, and for a quick free result that is a perfectly reasonable choice. What it will not do is vary line weight or decide which details to leave out, so the output reads as traced rather than drawn.
Several, and most are edge filters. Free is genuinely the right answer for colouring pages, cutting files and anything where faithfulness to the original beats artistry. Signing up here gives 50 free credits if you want to compare the generative route against them.
The Photos app has no true line-art conversion, so it needs a third-party app; most of those are edge filters running on the device. The web-based generative route works fine in a phone browser if you want the drawn look instead.
Because nothing forbade shading. Add "no grey wash, no shading gradients, pure black ink on white paper" and the same photo comes back as actual line work. This is the most common single fix.
With the likeness clause in the prompt, yes — facial structure and expression carry through reliably. Without it, models produce a competent drawing of somebody slightly different, and that is easy to miss until a family member spots it.
Line art scales badly when enlarged from a small render, so upscale the result rather than re-running at a larger size. Re-running changes the composition; upscaling keeps it.
Less well. Straight lines, repeated windows and mechanical detail are where generative models start inventing, and an edge filter will beat them on accuracy for free. Use the filter for those subjects.
Yes — ask for "a single continuous contour line, never lifting the pen". It is a distinct style with a strong look, and it needs to be requested explicitly because no model defaults to it.
The output is yours to use commercially. The constraint is the input: use your own photograph, or one you have the right to use, and be careful with photos of people who have not agreed to be drawn and sold.
Upload the picture, name the medium, forbid the grey, and run it a few times. Free credits on signup, prompt already in the box.