Low poly generator from an image
Your photo cut into coloured triangles: small ones on its edges, big calm ones in the background, each filled with the photo's own colour under it. Pick 200 to 5,000 triangles, then save a PNG, an SVG or a true-size poster PDF. Not AI, free, and the photo never leaves this device.
Edge focus 0 % spreads the points evenly; 100 % packs them on the photo's edges (eyes, outlines) and leaves the background in a few large triangles. The same photo, settings and seed always give the same picture.
The PDF is one page at true size, vector triangles inside a 10 mm white edge; print it at 100 % (Actual size), not Fit to page.
Privacy: your photo stays on this device. It is read and cut into triangles inside this page; nothing is uploaded or stored, and closing the tab forgets it. The page sends an anonymous page-view count and fixed event names only (lowpoly-make, lowpoly-download, lowpoly-svg, and lowpoly-download-size- followed by 1200, 2400, 4000, a4, a3, letter or tabloid). See the privacy notice.
How to use it
- Open a photo. Until you do, the preview shows a sample sunset drawn by the page. A face, a pet or a landscape with a clear subject works best.
- Set the detail (how many triangles, 200 to 5,000). The count shown under the picture is exact.
- Set the edge focus. Higher puts small triangles on outlines and features and large ones on plain areas; lower makes them all about the same size.
- Shuffle if a triangle cuts an eye badly: it draws a new seed and moves every point. Type an old seed back in to return to a version you liked.
- Pick Flat or Gradient, add edge lines if you want a wireframe look, then save a PNG, an SVG or a poster PDF. The file name says the size, for example
low-poly-a3.pdforlow-poly-2400px.png.
How it works, with numbers
The photo is redrawn 800 px on its long side. A Sobel filter measures how fast the brightness changes at every pixel; that edge map decides where points go. At edge focus 100 % a pixel on a strong edge is about 400 times as likely to get a point as a pixel in a flat area, and points keep a minimum distance from each other so they never bunch into slivers. The four corners and points along the four sides are always included, so the triangles fill the frame to its edge.
The points are joined by a Delaunay triangulation: of all ways to join them, the one in which no point lies inside any triangle's circumscribed circle, which avoids long, thin triangles. For n points with h of them on the frame, it has exactly 2n − h − 2 triangles, so the page places as many points as your triangle count needs.
Each triangle takes the average colour of the pixels whose centres fall inside it. If a triangle covers 120 pixels, 90 of them red (200, 30, 40) and 30 blue (20, 60, 200), its colour is (155, 38, 80): 90 × 200 + 30 × 20 = 18,600, divided by 120. In the gradient look, each corner takes the average of the triangles that meet there and the triangle shades from its darkest corner to its lightest.
What it cannot do
- It is not AI. It does not redraw or stylise a face; it only averages the colours your photo already has, so likeness depends on the triangle count.
- It does not cut the subject out of the background. Edge focus decides where detail goes, so a busy background gets busy triangles; a plain wall or sky works best.
- It is not a 3D model, an STL or a Blender mesh. The output is a flat picture; a photo has no depth to build a 3D shape from.
- Screen colours are not print colours. Very saturated colours print duller on most printers.
- Detail smaller than a triangle disappears. At 300 triangles an eye is one or two colours; raise the detail or the edge focus to keep it.
- Hairlines between triangles can show in some PDF viewers on screen (anti-aliasing); they do not print. The PNG has none.
- Nothing is saved. Close the tab and the settings are gone, so download first (or note the seed).
Questions
Can it make a 3D low poly model, an STL or a Blender mesh?
No. This makes flat, 2D low poly art: a picture made of coloured triangles. A photo has no depth information, so there is nothing to build a 3D shape from. The SVG holds 2D polygons only. For a 3D low poly model you need a modelling program such as Blender (its Decimate modifier reduces a mesh to fewer faces) or a 3D scan; a vinyl or laser cutter, though, can use the SVG directly.
How many triangles should I use?
About 300 to 600 for a bold, abstract look in which a face is only hinted at; 1,500 (the default) for a portrait or a pet that is clearly recognisable; 3,000 to 5,000 when small details such as eyes, whiskers or text must survive. On an A3 poster of a 4:3 photo, 1,500 triangles average about 70 square millimetres each, roughly 13 mm a side; with edge focus up, the triangles on edges are several times smaller and the background ones larger.
Why does the picture change when I press Shuffle?
Where the points fall is random, drawn from a seed number shown next to the button. Shuffle draws a new seed from your browser's crypto random generator, so the points move and the triangles change. The same photo, the same sliders and the same seed always give exactly the same triangles, so note the seed of a version you like, or type it back in to get it again.
How big can I print it?
As big as you like: the poster PDF and the SVG are vector shapes, not pixels, so they stay sharp at any size. The PDF is exactly A4 (210 × 297 mm), A3 (297 × 420 mm), Letter (8.5 × 11 in) or Tabloid (11 × 17 in), with a 10 mm white edge; print it at 100 % or Actual size. For a larger print shop size, send them the SVG. The PNG is 1200, 2400 or 4000 px on its long side; 4000 px is about 34 cm at 300 dpi.
Why do the printed colours look different from the screen?
A screen mixes light (RGB) and a printer mixes ink (CMYK). The triangles carry your photo's own colours, so they are usually inside what a printer can reproduce, but very saturated ones, such as a bright blue sky or a neon sign, print duller. Print one A4 first if the colours matter.