Vector vs raster: which format to use for logos and print

Vectors store shapes as maths, rasters store a fixed grid of pixels. Logos, type and line art belong in vector; photographs can only be raster. Here is how to pick the file.

A colourful vector illustration of a jester built from flat shapes and smooth curves
Photo: LeoBlanchette, CC0, via Wikimedia Commons

Use vector for anything made of shapes: logos, type, icons, line art, diagrams and signage. Use raster for anything made of continuous tone: photographs, paintings and textures. A vector file (SVG, PDF, AI, EPS) stores paths as coordinates and curves, so it prints sharp at any size from a visiting card to a hoarding; a raster file (JPEG, PNG, WebP, AVIF) stores a fixed grid of pixels, so it has one honest maximum size, about 300 pixels per inch for print held in the hand.

Key takeaways
  • Vector (SVG, PDF, AI, EPS) for logos, type, icons and line art: one file covers every size.
  • Raster (JPEG, PNG, WebP, AVIF) for photos: they have no shapes to describe, only tones.
  • 300 ppi at final size for hand-held print; a 2 m sign viewed from 2 m needs only about 44 ppi.
  • Send printers a PDF/X with logos and type still vector and photos embedded at full resolution.
  • JPEG XL returned to Chrome in version 155 (Oct 2026), with 30 to 50% smaller files than JPEG.

Most print disasters a small business sees, such as a blurry logo on a standee or a jagged letterhead, come from a raster file being stretched past its pixels. Most web performance problems come the other way, from photos saved as huge PNGs.

This post explains:

  • what each kind of file actually stores;
  • why logos must be vector and photos cannot be;
  • which format to send for which job;
  • what changed in 2026 with JPEG XL returning to Chrome.
1vector logo file covers every size, from favicon to hoarding
300 ppiusual raster requirement at final print size
30 to 50%smaller files Google claims for JPEG XL vs JPEG

What is the difference between vector and raster?

A raster image is a grid of pixels, each storing a colour; a vector image is a list of drawing instructions, such as "move here, draw a cubic Bézier curve to there, fill it with this colour". The raster's detail is fixed when it is made. The vector's detail is computed fresh each time it is drawn, at whatever resolution the screen or printer has.

Vector curves are usually cubic Bézier segments: two end points and two control points that pull the curve, the same mathematics used in fonts and in the W3C's SVG specification. A circle needs only four such segments. The same circle as a raster needs every edge pixel stored, and at 4,000 pixels across that is millions of values, most of them repeating the same red.

The story

The idea that a page should be a set of curves, not a set of dots, became an industry in one deal. In December 1982 John Warnock and Charles Geschke founded Adobe to sell PostScript, a language that describes text and graphics as straight lines and cubic Bézier curves and turns them into dots only at the output device. Steve Jobs visited in spring 1983, and that December Apple signed: a $1.5 million advance against royalties plus $2.5 million for 20% of Adobe. On 23 January 1985 Apple announced the LaserWriter, the first printer to ship with PostScript, at $6,995. Its Canon engine printed only 300 dpi, yet the same file could run on a high-end Linotronic typesetter at far higher resolution. Desktop publishing grew from that one property: the file did not care what printed it.

Wikipedia: PostScript (history, the Apple deal and the LaserWriter)
Vector: 2 end points + 2 control pointsRaster: 64 stored pixels at this sizecontrol pointcontrol point
One cubic Bézier curve is four points no matter how large it is drawn. The raster version stores every pixel, and at this zoom its staircase edge is already visible.
FormatTypeCompressionTransparencyBest for
SVGVectorText, can be gzippedYesLogos, icons, diagrams on the web
PDF (PDF/X for print)Vector container, can hold rastersVariesYesPrint-ready artwork, logos for printers
AI, EPSVector (editable source, legacy exchange)VariesAI yes, EPS limitedMaster logo files, older print workflows
PNGRasterLosslessFull alphaScreenshots, flat graphics when no vector exists
JPEGRasterLossyNoPhotographs
WebPRasterLossy or losslessYesWeb photos and graphics
AVIFRasterLossy or losslessYesWeb photos, smallest files; no progressive display
JPEG XLRasterLossy or losslessYesLarge photos, progressive display (Chrome 155+, Safari 17+)

Why must a logo be vector?

Because a logo is used at sizes no single raster can cover. The same mark may appear as a 16-pixel favicon, on a visiting card at 2 cm, on a shop sign at 2 m and on a hoarding at 10 m.

A vector file scales to all of them without loss, and it also lets a printer work with paths, not pixels:

  • separate it cleanly into spot inks;
  • cut it in vinyl;
  • engrave or embroider it, both of which follow outlines.

The arithmetic of one logo at three sizes

A logo printed 2 m wide on a sign viewed at 2 m needs, at the eye's limit of about 1 arcminute, roughly 44 pixels per inch, so about 3,400 pixels across. On a visiting card, at 300 ppi and 2 cm, it needs 236 pixels.

UsePrinted sizeResolution neededRaster pixels across
Faviconscreen1 px per px16
Visiting card2 cm300 ppi236
Shop sign seen from 2 m2 mabout 44 ppiabout 3,400
Vector masteranynonenone: recalculated each time

A 500-pixel PNG emailed by a client works for neither the sign nor a sharp letterhead. A vector does both from one file. This is also why a brand colour should sit in the vector as a named spot or Lab colour; see is Pantone still worth it.

original, 40 pxvector, scaled 5.5×raster, scaled 5.5×
Enlarging a raster cannot invent the missing edge detail, so each source pixel becomes a visible block, or a blur if the software smooths it. The vector is recalculated at the new size.

Why can't photos be vector?

Because a photograph has no shapes to describe, only millions of slightly different tones. Turning one into paths, called vectorisation or tracing, ends one of two ways:

  • thousands of tiny shapes that are larger and slower than the original pixels; or
  • a simplified, poster-like illustration that is no longer the photograph.

The research on automatic vectorisation shows how hard the general problem still is.

Image Vectorization: a ReviewMaria Dziuba, Ivan Jarsky, Valeria Efimova and Andrey Filchenkov · arXiv 2306.06441, published in Zapiski Nauchnykh Seminarov POMI 530 · 2023

The ITMO University team tested four machine-learning vectorisers (Mang2Vec, Deep Vectorization of Technical Drawings, DiffVG and LIVE) on 66 images, half of them rasterised vector art and half ordinary bitmaps, on an NVIDIA RTX 3090 Ti for the GPU methods. DiffVG took 296 seconds to fit 32 paths to a 1920 × 1080 image and 837 seconds for 1,024 paths; LIVE took about two hours (7,150 s) for 256 paths at full HD. Mang2Vec produced up to 16,600 paths for a simple dragon drawing that originally had 25. The authors conclude that no fast, universal, automatic method exists and that every one needs human control. Limit: a small test set and no standard error metric across methods.

Read the paper (free)
Time to vectorise a 1920 × 1080 image with DiffVG
See the data as a table
Seconds (500 iterations) (s)
32296
64319
100349
256435
512583
1024837

Source: Dziuba et al., Image Vectorization: a Review, 2023 · x-axis: number of vector paths. Even 1,024 paths is far too few to describe a real photograph faithfully.

Pixel art is the interesting middle case: a raster made by hand from flat colours, where every pixel is a design decision. That is exactly where a specialised tracer can work well.

Depixelizing Pixel ArtJohannes Kopf and Dani Lischinski · ACM Transactions on Graphics 30(4), SIGGRAPH · 2011

The authors turn hand-made pixel art into smooth vector shapes. They build a similarity graph linking neighbouring pixels of similar colour, resolve ambiguous diagonal crossings with three voting heuristics (keep long curves connected, favour the sparser colour, avoid leaving single-pixel islands), reshape the pixel cells, then fit and optimise spline curves with corner detection. On 54 examples, total processing on one core of a 2.4 GHz CPU had a median of 0.62 s and a maximum of 3.06 s. Limits they state: it is designed for hand-crafted sprites, struggles with anti-aliased sprites made by downsampling photos, and sometimes smooths corners that should stay sharp.

Read the paper (free)
Pixel art of a city skyline at night, enlarged so the individual square pixels are clearly visible
Pixel art enlarged with plain nearest-neighbour scaling: each pixel becomes a square block. For a logo, this is the look you get when a small raster is stretched for print.Photo: Stvk Công Cuối (VN), CC0, via Wikimedia Commons

Which file should I send to a printer?

Send a PDF, ideally PDF/X, with logos and type left as vectors and photos embedded at 300 ppi at the final size. PDF/X is the ISO 15930 family of PDF subsets for print production, and it requires:

  • an output intent naming the printing condition;
  • trim boxes;
  • embedded fonts (mandatory in PDF/X-1a);
  • no active content.

Together these remove most of the surprises at the press.

  1. Keep the master logo as vector. An AI or SVG source, plus exported PDF and EPS for older vendors. Convert text to outlines in the delivered files so missing fonts cannot substitute.
  2. Check photo resolution at final size. Pixels divided by inches printed; 300 ppi for hand-held print, less for large formats viewed from afar. Our megapixels guide works through the distances.
  3. Convert colour on purpose. Photos to the printer's CMYK profile, brand colours as named spots or exact CMYK builds; see RGB vs CMYK.
  4. Never upsample to fake 300 ppi. Raising pixel count adds no detail; printers such as PsPrint warn this makes a fuzzy print. Ask for the original or a vector instead.
  5. Export PDF/X. PDF/X-1a for CMYK-only workflows, PDF/X-4 when transparency and colour management are kept live.
A metal offset printing plate with the exposed image areas visible
An exposed offset printing plate. Plates are imaged with a fine halftone screen, so even vector artwork becomes dots in the end; the point of sending vectors is that the screen is applied once, at full output resolution.Photo: User:Raboe001, CC BY-SA 2.5, via Wikimedia Commons

How many pixels does a raster logo need?

Use this to check what you just read. If a vector really is not available, enter the print width and resolution to see how many pixels the raster needs, and how much memory that is before compression.

Pixels a raster needs for a print size

A 20 cm square logo at 300 ppi needs 2,362 pixels each way, about 5.6 MP and 21 MB uncompressed. For a 3 m shop fascia at 50 ppi the raster needs 5,906 pixels across, while the vector file is the same few kilobytes in both cases.

A vector logo has no size limit: its file size depends on the number of points, not on the print size.

What about web formats in 2026?

On the web, use SVG for logos and icons, and AVIF or WebP with a JPEG fallback for photos; JPEG XL is now an option for large photos. MDN's guidance is that SVG should be preferred whenever a vector version of an icon exists, and that lossy formats blur the sharp edges of logos and line art.

JPEG XL comes back

The 2026 change is JPEG XL. Its path back into the browser took four years:

  1. Chrome removed it in 2022;
  2. a Rust decoder (jxl-rs) landed in Chromium in January 2026;
  3. Google's developer blog announced in October 2026 that Chrome 155 ships JPEG XL decoding, citing 30 to 50% better compression than JPEG.

MDN lists support in Chrome 155, Firefox 158 and Safari 17. For more of my notes on design files and tools, see the Kala page.

A web-fed offset press with inked plate cylinders printing newspapers
A web-fed offset press. Whatever format the artwork arrives in, it ends as dots on a plate, which is why the rule is to keep shapes as vectors until the last possible step.Photo: Tom T, CC BY 2.0, via Wikimedia Commons

When is raster the right choice for graphics?

Raster is right when the artwork has texture or tone:

  • painted illustrations and grain;
  • photographic backgrounds;
  • complex gradients with noise;
  • screenshots.

It is also the safe delivery format for social media, where platforms recompress uploads anyway. Make the raster from the vector master at the exact size needed, rather than resizing an old raster.

For logo projects, the deliverable list should therefore always include both: the vector masters and a set of rasters exported at fixed sizes. The logo design process guide covers what to hand over, and why logos look dated explains why simple vector marks age better.

Questions people ask

Can I convert a JPEG logo to vector?

Yes, by tracing, and for a simple flat logo auto-trace in Illustrator or Inkscape can get close. Expect to clean up curves and corners by hand; the 2023 review found every automatic method needs human control. For anything complex, redraw it.

Is PNG a vector format?

No. PNG is a lossless raster format. It keeps sharp edges and transparency, which is why it is often used for logos on the web, but it still has a fixed pixel size and blurs or blocks when enlarged.

What is the difference between a point and a pixel?

A point is a physical unit, 1/72 of an inch in digital typography, so it has a fixed size on paper. A pixel is one cell of a raster grid and has no physical size until you set a resolution: at 300 ppi a pixel is about 0.085 mm, at 72 ppi a pixel equals a point.

Is SVG good for print?

SVG is a web format and many print workflows do not accept it directly. Open it in a vector editor and export a PDF/X or EPS for the printer; the paths stay vector.

Sources

  1. Dziuba, M. et al.: Image Vectorization: a Review, arXiv (2023)
  2. Kopf, J. and Lischinski, D.: Depixelizing Pixel Art, ACM TOG (SIGGRAPH) (2011)
  3. W3C: Scalable Vector Graphics (SVG) 2
  4. MDN Web Docs: Image file type and format guide (2026)
  5. ISO 15930: PDF/X for prepress digital data exchange
  6. Wikipedia: PDF/X (requirements of X-1a, X-3 and X-4)
  7. Chrome for Developers: Shipping JPEG XL in Chrome (2026)
  8. The Register: Google rekindles relationship with JPEG XL (2026)
  9. PsPrint Help Center: Artwork resolution
  10. Graphic Design Stack Exchange: Differences between vector and raster graphics
  11. Wikipedia: PostScript (source for Adobe's founding in December 1982, the Apple licence and the LaserWriter's launch, price and 300 dpi engine)
vector vs rasterSVGPDF/XPNGJPEGAVIFJPEG XLBezier curveslogo files
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Artblog.

Keep reading in Artblog

From the other blogs