DVISVGM(1)

📦
version 3.6.1 · 11 versions
, in arch rolling, ubuntu stonking

🌈 DVISVGM(1) 🌈

okay bestie, welcome to the dvisvgm Manual — the group chat where Unix actually makes sense 💖

✨ NAME ✨

dvisvgm — converts DVI, EPS, and PDF files to the XML-based SVG format

🦋 SYNOPSIS 🦋

dvisvgm [options] file[.dvi]

dvisvgm --eps [options] file1[.eps] file2[.eps] ...

dvisvgm --pdf [options] file[.pdf]

💖 DESCRIPTION 💖

The command-line utility dvisvgm converts DVI files, as generated by TeX/LaTeX, to the XML-based scalable vector graphics format SVG. It supports the classic DVI version 2 as well as version 3 (created by pTeX in vertical mode), and the XeTeX versions 5 to 7 which are also known as XDV. Besides the basic DVI commands, dvisvgm also evaluates many so-called specials which heavily extend the capabilities of the plain DVI format. For a more detailed overview, see section Supported Specials below.

Since the current SVG standard 1.1 doesn’t specify multi-page graphics, dvisvgm creates separate SVG files for each DVI page. Because of compatibility reasons, only the first page is converted by default. In order to select a different page or arbitrary page sequences, use option -p which is described below.

SVG is a vector-based graphics format and therefore dvisvgm tries to convert the glyph outlines of all fonts referenced in a DVI page section to scalable path descriptions. The fastest way to achieve this is to extract the path information from vector-based font files available in PFB, TTF, or OTF format. If dvisvgm is able to find such a file, it extracts all necessary outline information about the glyphs from it.

However, TeX’s former main source for font descriptions is Metafont, which produces bitmap output in terms of GF files. That’s why not all obtainable TeX fonts are available in a scalable format. In these cases, dvisvgm tries to vectorize Metafont’s output by tracing the glyph bitmaps. The results are not as perfect as most (manually optimized) PFB or OTF counterparts, but are nonetheless really nice in most cases.

When running dvisvgm without option --no-fonts, it creates font elements (<font>...</font>) to embed the font data into the SVG files. Unfortunately, only few SVG renderers support these elements. Most web browsers and vector graphics applications don’t evaluate them properly so that the text components of the resulting graphics might look strange. In order to create more compatible SVG files, command-line option --no-fonts can be given to replace the font elements by plain graphics paths. Most web browsers (but only few external SVG renderers) also support WOFF and WOFF2 fonts that can be used instead of the default SVG fonts. Option --font-format offers the functionality to change the format applied to the fonts being embedded. This, however, only works for DVI files or PDF files converted via the new mutool-based PDF handler (see below). Text present in PostScript files is always converted to path elements.

🎀 OPTIONS 🎀

dvisvgm provides a POSIX-compliant command-line interface with short and long option names. They may be given before and/or after the name of the file to be converted. Also, the order of specifying the options is not significant, i.e. you can add them in any order without changing dvisvgm’s behavior. Certain options accept or require additional parameters which are directly appended to or separated by whitespace from a short option (e.g. -v0 or -v 0). Long options require an additional equals sign (=) between option name and argument but without any surrounding whitespace (e.g. --verbosity=0). Multiple short options that don’t expect further parameters can be combined after a single dash (e.g. -ejs rather than -e -j -s).

Long option names may also be shortened by omitting trailing characters as long as the shortened name is still unambiguous. For example, option --exact-bbox can be shortened to --exact, --exa, or --ex. In case of ambiguous abbreviations, dvisvgm prints an error message together with all matching option names.

  • -b, --bbox=fmt — sets the bounding box of the generated SVG graphic to the specified format. This option only affects the conversion of DVI files. SVG documents generated from PDF and PostScript always inherit the bounding boxes of the input files.

Parameter fmt takes either one of the format specifiers listed below, or a sequence of four comma- or whitespace-separated length values x1, y1, x2 and y2. The latter define the absolute coordinates of two diagonal corners of the bounding box. Each length value consists of a floating point number and an optional length unit (pt, bp, cm, mm, in, pc, dd, cc, or sp). If the unit is omitted, TeX points (pt) are assumed.

It’s also possible to give only one length value l. In this case, the minimal bounding box is computed and enlarged by adding (-l,-l) to the upper left and (l,l) to the lower right corner.

Additionally, dvisvgm also supports the following format specifiers:

  • International DIN/ISO paper sizes — An, Bn, Cn, Dn, where n is a non-negative integer, e.g. A4 or a4 for DIN/ISO A4 format (210mm × 297mm).
  • North American paper sizes — invoice, executive, legal, letter, ledger
  • Special bounding box sizes
    • dvi — page size stored in the DVI file
    • min — computes the minimal/tightest bounding box
    • none — no bounding box is assigned
    • papersize — box sizes specified by papersize specials present in the DVI file
    • preview — bounding box data computed by the preview package (if present in the DVI file)
  • Page orientation — the default page orientation for DIN/ISO and American paper sizes is portrait, i.e. width < height. Appending -landscape or simply -l to the format string switches to landscape mode (width > height). For symmetry reasons you can also explicitly add -portrait or -p to indicate the default portrait format. Note that these suffixes are part of the size string and not separate options. Thus, they must directly follow the size specifier without additional blanks. Furthermore, the orientation suffixes can’t be used with dvi, min, and none.
Note: Option -b, --bbox only affects the bounding box and does not transform the page content. Hence, if you choose a landscape format, the page won’t be rotated.
  • -B, --bitmap-format=fmt — this option sets the image format used to embed bitmaps that are extracted from PostScript or PDF data. By default, dvisvgm embeds all bitmaps as JPEG images because it’s the most compact of the two formats supported by SVG. To select the alternative lossless PNG format, --bitmap-format=png can be used. There are some more format variants dvisvgm currently supports even though jpeg and png should be sufficient in most cases. The following list gives an overview of the known format names which correspond to names of Ghostscript output devices.
    • none — disable processing of bitmap images
    • jpeg — color JPEG format
    • jpeggray — grayscale JPEG format
    • png — grayscale or 24-bit color PNG format depending on current color space
    • pnggray — grayscale PNG format
    • pngmono — black-and-white PNG format
    • pngmonod — dithered black-and-white PNG format
    • png16 — 4-bit color PNG format
    • png256 — 8-bit color PNG format
    • png16m — 24-bit color PNG format

Since the collection of supported output devices can vary among local Ghostscript installations, not all formats may be available in some environments. dvisvgm quits with a PostScript error message if the selected output format requires a locally unsupported output device.

The two JPEG format specifiers accept an optional parameter to set the IJG quality level which must directly follow the format specifier separated by a colon, e.g. --bitmap-format=jpeg:50. The quality value is an integer between 0 and 100. Higher values result in better image quality but lower compression rates and therefore larger files. The default quality level is 75 which is applied if no quality parameter is given or if it’s set to 0.

  • -C, --cache[=dir] — to speed up the conversion process of bitmap fonts, dvisvgm saves intermediate conversion information in cache files. By default, these files are stored in $XDG_CACHE_HOME/dvisvgm/ or $HOME/.cache/dvisvgm if XDG_CACHE_HOME is not set. If you prefer a different location, use option --cache to overwrite the default. Furthermore, it is also possible to disable the font caching mechanism completely with option --cache=none. If argument dir is omitted, dvisvgm prints the path of the default cache directory together with further information about the stored fonts. Additionally, outdated and corrupted cache files are removed.
  • -j, --clipjoin — this option tells dvisvgm to compute all intersections of clipping paths itself rather than delegating this task to the SVG renderer. The resulting SVG files are more portable because some SVG viewers don’t support intersecting clipping paths which are defined by clipPath elements containing a clip-path attribute.
  • --color — enables colorization of messages printed during the conversion process. The colors can be customized via environment variable DVISVGM_COLORS. See the Environment section below for further information.
  • --colornames — by default, dvisvgm exclusively uses RGB values of the form #RRGGBB or #RGB to represent colors in the SVG file. The latter is a short form for colors whose RGB components each consist of two identical hex digits, e.g. #123 equals #112233. According to the SVG standard, it’s also possible to use color names (like black and darkblue) for a limited number of predefined colors (https://www.w3.org/TR/SVG11/types.html#ColorKeywords). In order to apply these color names rather than their RGB values, call dvisvgm with option --colornames. All colors without an SVG color name will still be represented by RGB values.
  • --comments — adds comments with further information about selected data to the SVG file. Currently, only font elements and font CSS rules related to native fonts are annotated.
  • --currentcolor[=color] — this option tells dvisvgm to replace the specified color with CSS variable currentColor which holds the current foreground color set by CSS property color. This way an application or website can change the color via CSS without the need to modify the SVG code.

okay so about that optional argument for --currentcolor — it's the RGB color value you wanna replace, and it has to be either a six-digit or three-digit hex color value (like, per the CSS spec at https://www.w3.org/TR/css-color/#typedef-hex-color) with a # in front, e.g. #0000ff or #00f for blue. if you don't give it an argument, black is assumed. slay. 💅

  • --embed-bitmaps — this one embeds the actual contents of bitmap files into the generated SVG files instead of just adding file references. because of the base64-encoded data, the SVG files are usually way bigger, but at the same time more portable because they don’t rely on external bitmap files. no cap. this option only affects bitmaps referenced in DVI/XDV files, e.g. by \includegraphics or the special command dvisvgm:img (see below). bitmaps present in PostScript or PDF files are always embedded. also see option bitmap-format. 🖼️
  • -E, --eps — if you give this option, dvisvgm doesn’t expect a DVI but a single or multiple EPS input files, and tries to convert them to SVG. in order to do so, a single psfile special command is created and forwarded to the PostScript special handler. this option is only available if dvisvgm was built with PostScript support enabled, and requires Ghostscript to be available. see option --libgs for further information. ✨
  • -e, --exact-bbox — this option tells dvisvgm to compute the precise bounding box of each character. by default, the values stored in a font’s TFM file are used to determine a glyph’s extent. as these values are intended to implement optimal character placements and are not designed to represent the exact dimensions, they don’t necessarily correspond with the bounds of the visual glyphs. thus, width and/or height of some glyphs may be larger (or smaller) than the respective TFM values. as a result, this can lead to clipped characters at the bounds of the SVG graphics. with option --exact-bbox given, dvisvgm analyzes the actual shape of each character and derives a usually tight bounding box. 📐 this option only affects the processing of DVI files. when converting EPS or PDF files, the bounding box information stored in these files are used to derive the SVG bounding box.
  • -f, --font-format=format — selects the file format used to embed font data into the generated SVG output when converting DVI or PDF files. the latter require the new mutool-based PDF handler introduced with dvisvgm 3.0 (also see option --pdf). option --font-format has no effect when converting PDF or PostScript files using Ghostscript. text fragments present in these files are then always converted to path elements. 🔤 following formats are supported: SVG (that’s the default), TTF (TrueType), WOFF, and WOFF2 (Web Open Font Format version 1 and 2). by default, dvisvgm creates unhinted fonts that might look bad on low-resolution devices. in order to improve the display quality, the generated TrueType, WOFF, or WOFF2 fonts can be autohinted. the autohinter is enabled by appending ,autohint or ,ah to the font format, e.g. --font-format=woff,autohint or --fwoff,ah. this functionality requires the ttfautohint library (https://freetype.org/ttfautohint). if it’s not available or can’t be found, dvisvgm issues a corresponding warning message. option --font-format is only available if dvisvgm was built with WOFF support enabled. 💖
  • -m, --fontmap=filenames — loads and evaluates a single font map file or a sequence of font map files. these files are required to resolve font file names and encodings. dvisvgm does not provide its own map files but tries to read available ones coming with dvips or dvipdfm. if option --fontmap is omitted, dvisvgm looks for the default map files dvisvgm.map, ps2pk.map, pdftex.map, dvipdfm.map, and psfonts.map (in this order). otherwise, the files given as option arguments are evaluated in the given order. multiple filenames must be separated by commas without leading and/or trailing whitespace. 🗺️ by default, redefined mappings do not replace previous ones. however, each filename can be preceded by an optional mode specifier (+, -, or =) to change this behavior:
    • +mapfile — only those entries in the given map file that don’t redefine a font mapping are applied, i.e. fonts already mapped keep untouched. that’s also the default mode if no mode specifier is given. ✨
    • -mapfile — ensures that none of the font mappings defined in the given map file are used, i.e. previously defined mappings for the specified fonts are removed. 🗑️
    • =mapfile — all mappings defined in the map file are applied. previously defined settings for the same fonts are replaced. 🔁 if the first filename in the filename sequence is preceded by a mode specifier, dvisvgm loads the default font map (see above) and applies the other map files afterwards. otherwise, none of default map files will be loaded automatically. examples: --fontmap=myfile1.map,+myfile2.map loads myfile1.map followed by myfile2.map where all redefinitions of myfile2.map are ignored. --fontmap==myfile1.map,-myfile2.map loads the default map file followed by myfile1.map and myfile2.map where all redefinitions of myfile1.map replace previous entries. afterwards, all definitions for the fonts given in myfile2.map are removed from the font map tree. for further information about the map file formats and the mode specifiers, see the manuals of dvips (https://tug.org/texinfohtml/dvips.html) and dvipdfm (https://ctan.org/tex-archive/dviware/dvipdfm). dvisvgm supports both the dvips and the dvipdfm map file formats. it extends both variants by two include statements that allow for loading other map files inside a map file. the syntax is as follows:
#include [<modechar>] <filename|filepath>
#includefirst [<modechar>] <filename|filepath>

they must be placed on a separate line and start at the first column of that line. otherwise, they are ignored. while the first variant always tries to include the specified file, #includefirst is only executed once, i.e. once a file was successfully included by this statement, all subsequent #includefirst lines are skipped. this can be used to include one of several alternative files, whereby the first one found is loaded and all others are ignored. the optional parameter <modechar> (which can be +, -, or =) determines how to integrate the data read from the included file into the already present mapping data. they work the same way as the mode specifiers described above. if <modechar> parameter is omitted, it defaults to +. the following filename or file path specifies the file to include. in case of plain filenames (without path indicators), the files are looked up using the TeX file search functionality, i.e. files in the TeX directory tree can easily be added. relative paths are relative to the location of the file containing the #include statement. path components must be separated by forward slashes (/), also on Windows systems. examples: #include pdftex.map looks for pdftex.map in the current working directory and in the TeX directory tree. #include ./pdftex.map, on the other hand, looks for pdftex.map only in the directory where the including file is located. 🫶

  • --grad-overlap — tells dvisvgm to create overlapping grid segments when approximating color gradient fills (also see option --grad-segments below). by default, adjacent segments don’t overlap but only touch each other like separate tiles. however, this alignment can lead to visible gaps between the segments because the background color usually influences the color at the boundary of the segments if the SVG renderer uses anti-aliasing to create smooth contours. one way to avoid this and to create seamlessly touching color regions is to enlarge the segments so that they extend into the area of their right and bottom neighbors. since the latter are drawn on top of the overlapping parts, the visible size of all segments keeps unchanged. just the former gaps disappear as the background is now completely covered by the correct colors. currently, dvisvgm computes the overlapping segments separately for each patch of the mesh (a patch mesh may consist of multiple patches of the same type). therefore, there still might be visible gaps at the seam of two adjacent patches. 🌈
  • --grad-segments=number — determines the maximum number of segments per column and row used to approximate gradient color fills. since SVG 1.1 only supports a small subset of the shading algorithms available in PostScript, dvisvgm approximates some of them by subdividing the area to be filled into smaller, monochromatic segments. each of these segments gets the average color of the region it covers. thus, increasing the number of segments leads to smaller monochromatic areas and therefore a better approximation of the actual color gradient. as a drawback, more segments imply bigger SVG files because every segment is represented by a separate path element. 📊 currently, dvisvgm supports free- and lattice-form triangular patch meshes as well as Coons and tensor-product patch meshes. they are approximated by subdividing the area of each patch into a n×n grid of smaller segments. the maximum number of segments per column and row can be changed with option --grad-segments.
  • --grad-simplify=delta — if the size of the segments created to approximate gradient color fills falls below the given delta value, dvisvgm reduces their level of detail. for example, Bézier curves are replaced by straight lines, and triangular segments are combined to tetragons. for a small delta, these simplifications are usually not noticeable but reduce the size of the generated SVG files significantly. 🎀
  • -h, --help[=mode] — Prints a short summary of all available command-line options, bestie. The optional mode parameter is an integer value between 0 and 2, and it picks the display variant of the help text. Mode 0 lists all options divided into categories with section headers, and this is also the default if dvisvgm is called without parameters. Mode 1 lists all options ordered by the short option names, while mode 2 sorts the lines by the long option names.

The values in brackets after the description text indicate the default parameter of the option. They’re applied if an option with a mandatory parameter is not used or if an optional parameter is omitted. For example, option --bbox requires a size parameter which defaults to min if --bbox is not used. Option --zip, which isn’t applied by default, accepts an optional compression level parameter. If it’s omitted, the stated default value 9 is used. ✨

  • --keep — Disables the removal of temporary files as created by Metafont (usually .gf, .tfm, and .log files) or the TrueType/WOFF module. Like, keep the little mess around, okay? 🎀
  • --libgs=path — This option is only available if the Ghostscript library is not directly linked to dvisvgm and if PostScript support was not completely disabled during compilation. In this case, dvisvgm tries to load the shared GS library dynamically during runtime. By default, it expects the library’s name to be libgs.so.X (on Unix-like systems, where X is the ABI version of the library) or gsdll32.dll/gsdll64.dll (Windows). If dvisvgm doesn’t find the library, option --libgs can be used to specify the correct path and filename, e.g. --libgs=/usr/local/lib/libgs.so.9 or --libgs=\gs\gs9.25\bin\gsdll64.dll.

Alternatively, it’s also possible to assign the path to environment variable LIBGS, e.g. export LIBGS=/usr/local/lib/libgs.so.9 or set LIBGS=\gs\gs9.25\bin\gsdll63.dll. LIBGS has less precedence than the command-line option, i.e. dvisvgm ignores variable LIBGS if --libgs is given. No cap. 💅

  • -L, --linkmark=style — Selects the method how to mark hyperlinked areas. The style argument can take one of the values none, box, and line, where box is the default, i.e. a rectangle is drawn around the linked region if option --linkmark is omitted. Style argument line just draws the lower edge of the bounding rectangle, and none tells dvisvgm not to add any visible objects to hyperlinks. The lines and boxes get the current text color selected. In order to apply a different, constant color, a colon followed by a color specifier can be appended to the style string. A color specifier is either a hexadecimal RGB value of the form #RRGGBB, or a dvips color name (https://en.wikibooks.org/wiki/LaTeX/Colors#The_68_standard_colors_known_to_dvips).

Moreover, argument style can take a single color specifier to highlight the linked region by a frameless box filled with that color. An optional second color specifier separated by a colon selects the frame color.

Examples: box:red or box:#ff0000 draws red boxes around the linked areas. yellow:blue creates yellow filled rectangles with blue frames. 🔥

  • -l, --list-specials — Prints a list of registered special handlers and exits. Each handler processes a set of special statements belonging to the same category. In most cases, these categories are identified by the prefix of the special statements which is usually a leading string followed by a colon or a blank, e.g. color or ps. The listed handler names, however, don’t need to match these prefixes, e.g. if there is no common prefix or if functionality is split into separate handlers in order to allow them to be disabled separately using option --no-specials. All special statements not covered by one of the special handlers are silently ignored. 🫶
  • -M, --mag=factor — Sets the magnification factor applied in conjunction with Metafont calls prior tracing the glyphs. The larger this value, the better the tracing results. Nevertheless, large magnification values can cause Metafont arithmetic errors due to number overflows. So, use this option with care. The default setting usually produces nice results.
bestie, big factor can be iconic for tracing but also chaotic for Metafont number overflows — use with care. 💀
  • --message=text — Prints a given message to the console after an SVG file has been written. Argument text may consist of static text as well as the macros listed below in the description of special command dvisvgm:raw. For example, --message="page {?pageno} written to {?svgfile}" prints the message with the macros expanded after the conversion of each page of a DVI or PDF file or after processing an EPS file.

The output of option --message is not affected by a specified verbosity level, i.e. dvisvgm prints the text even with --verbosity=0. 📣

  • --no-merge — Puts every single character in a separate text element with corresponding x and y attributes. By default, new text or tspan elements are only created if a string starts at a location that differs from the regular position defined by the characters' advance values. 🦋
  • --no-mktexmf — Suppresses the generation of missing font files. If dvisvgm can’t find a font file through the kpathsea lookup mechanism, it calls the external tools mktextfm or mktexmf. This option disables these calls. No extra font-generation side quests. 🚫
  • -n, --no-fonts[=variant] — If this option is given, dvisvgm doesn’t create SVG font elements but uses paths instead. The resulting SVG files tend to be larger but are concurrently more compatible with most applications that don’t support SVG fonts. The optional argument variant selects the method how to substitute fonts by paths. Variant 0 creates path and use elements in order to avoid lengthy duplicates. Variant 1 creates path elements only. Option --no-fonts implies --no-styles. 🌈
  • -c, --scale=sx[,sy] — Scales the page content horizontally by sx and vertically by sy. This option is equivalent to -TSsx,sy. 💖
  • -S, --no-specials[=names] — Disable processing of special commands embedded in the DVI file. If no further parameter is given, all specials are ignored. To disable a selected set of specials, an optional comma-separated list of names can be appended to this option. A name is the unique identifier referencing the intended special handler as listed by option --list-specials. 🎀
  • --no-styles — By default, dvisvgm creates CSS styles and class attributes to reference fonts. This variant is more compact than adding the complete font information to each text element over and over again. However, if you prefer direct font references, the default behavior can be disabled with option --no-styles. 🧚
  • -O, --optimize[=modules] — Applies several optimizations on the generated SVG tree to reduce the file size. The optimizations are performed by running separate optimizer modules specified by optional argument modules. It may consist of a single module name or a comma-separated list of several module names. The corresponding modules are executed one by one in the given order and thus transform the XML tree gradually. A leading hyphen (-) can be used to exclude a module from processing. For example -Oall,-collapse-groups performs all optimizations except collapse-groups. If the first module name in the list starts with a hyphen, dvisvgm automatically inserts all at the beginning, i.e. -O-collapse-groups is identical to -Oall,-collapse-groups.

The following list describes the currently available optimizer modules. 👑

  • list — Lists all available optimizer modules and exits.
  • none — If this argument is given, dvisvgm doesn’t apply any optimization. none can’t be combined with other module names.
  • all — Performs all optimizations listed below. This is also the default if option --optimize is used without any arguments. The modules are executed in a predefined order that usually leads to the best results. all must be the only or the first entry in the list of module names.
  • collapse-groups — Combines nested group elements (<g>...</g>) that contain only a single attribute each. If possible, the group attributes are moved to the outermost element of the processed subtree. This module also unwraps group elements that have no attributes at all.
  • group-attributes — Creates groups (<g>...</g>) for common attributes around adjacent elements. Each attribute is moved to a separate group so that multiple common attributes lead to nested groups. They can be combined by applying optimizer module collapse-groups afterwards. The algorithm only takes inheritable properties, such as fill or stroke-width, into account and only removes them from an element if none of the other attributes, like id, prevents this.
  • reassign-clippaths — Collects all clipPath elements that differ only by their IDs. Afterwards, the duplicates are removed so that only one remains. All clip-path attributes referencing one of the duplicates are updated accordingly.
  • remove-clippaths — Removes all redundant clipPath elements. This optimization was already present in former versions of dvisvgm and was always applied by default. This behavior is retained, i.e. dvisvgm executes this module even if option --optimize is not given. You can use argument none to prevent that.
  • simplify-text — If a text element only contains whitespace nodes and tspan elements, all common inheritable attributes of the latter are moved to the enclosing text element. All tspan elements without further attributes are unwrapped, i.e. the tspan tags are removed.
  • simplify-transform — Tries to shorten all transform attributes. This module combines the transformation commands of each attribute and decomposes the resulting transformation matrix into a sequence of basic transformations, i.e. translation, scaling, rotation, and skewing. If this sequence is shorter than the equivalent matrix expression, it’s assigned to the attribute. Otherwise, the matrix expression is used. 💜
  • -o, --output=pattern — this sets the pattern that picks the names of the generated SVG files, and yes bestie it can include a file path too 💖

The required parameter pattern may be an arbitrary sequence of characters making up the filenames or file paths relative to the current working directory. Except for the variables and expressions listed below, all characters are treated as static parts of the filenames/folders, so they’re identical for all pages processed during one run of dvisvgm. The strings %f, %p, %P, %n, %N, %hd, %ho, and %hc are variables you can use as part of the pattern. %f expands to the base name of the DVI file, i.e. the filename without suffix, %p is the current page number, and %P is the total number of pages in the DVI file. The variables %n and %N expand to the number of the current file being converted (starting at 1) and the total number of files given, respectively.

An optional number (0-9) specified directly after the percent sign of a variable holding a numeric value denotes the minimal number of digits to be created. If a particular value consists of fewer digits, the number is padded with leading zeros. Example: %3p enforces 3 digits for the current page number (001, 002, etc.). Without an explicit width specifier, %p gets the same number of digits as %P.

If you need more control over numbering, you can use arithmetic expressions as part of the pattern. The syntax is %(expr) where expr may contain additions, subtractions, multiplications, and integer divisions with common precedence. The variables p and P contain the current page number and the total number of pages, respectively. For example, --output="%f-%(p-1)" creates filenames where the numbering starts with 0 rather than 1.

The variables %hX contain different hash values computed from the DVI page data and the options given on the command-line. %hd and %hc are only set if option --page-hashes is present. Otherwise, they are empty. For further information, see the description of option --page-hashes below.

The default pattern is %f-%p.svg if the DVI file consists of more than one page, and %f.svg otherwise. That means a DVI file foo.dvi is converted to foo.svg if foo.dvi is a single-page document. Otherwise, multiple SVG files foo-01.svg, foo-02.svg, etc. are produced.

Windows bestie warning 🫶: in Windows environments, the percent sign indicates dereferenced environment variables, and must therefore be protected by a second percent sign, e.g. --output=%%f-%%p.
  • -p, --page=ranges — this option selects the pages to be processed, and it’s lowkey the page-picking queen 👑

Parameter ranges consists of a comma-separated list of single page numbers and/or page ranges. A page range is a pair of numbers separated by a hyphen, e.g. 5-12. Thus, a page sequence might look like this: 2-4,6,9-12,15. It doesn’t matter if a page is given more than once or if page ranges overlap. dvisvgm always extracts the page numbers in ascending order and converts them only once.

In order to stay compatible with previous versions, the default page sequence is 1. dvisvgm therefore converts only the first page and not the whole document if option --page is omitted. Usually, page ranges consist of two numbers denoting the first and last page to be converted. If the conversion should start at page 1, or if it should continue up to the last DVI page, the first or second range number can be omitted, respectively. Example: --page=-10 converts all pages up to page 10, --page=10- converts all pages starting with page 10. Please consider that the page values don’t refer to the page numbers printed on the corresponding page. Instead, the physical page count is expected, where the first page always gets number 1.

To select the last page, expression %P can be used instead of an explicit page number. It’s replaced by the number of the last page. For example, --page=1,%P converts the first and the last page of the document.

Additionally, arithmetic expressions of the form %(expr) in place of numbers are valid too. The resulting values are always rounded down to the nearest integer. See option --output for more information about the syntax and the supported arithmetic operators. Example: --page=%(P-2)- converts the last three pages.

At the end of the range sequence an optional filter specifier can be appended in order to remove certain page numbers from the sequence. Currently, the two filters :even and :odd are supported which restrict the preceding values to even and odd numbers, respectively. For example, --page=1-11,20:even is equivalent to --page=2,4,6,8,10,20.

  • -H, --page-hashes[=params] — this one is giving “skip the repeat work” main character energy 🔥

If this option is given, dvisvgm computes hash values of all pages to be processed. As long as the page contents don’t change, the hash value of that page stays the same. This property can be used to determine whether a DVI page must be converted again or can be skipped in consecutive runs of dvisvgm. This is done by propagating the hash value to variable %hd which can be accessed in the output pattern (see option --output). By default, dvisvgm changes the output pattern to %f-%hd if option --page-hashes is given. As a result, all SVG file names contain the hash value instead of the page number. When calling dvisvgm again with option --page-hashes with the same output pattern, it checks the existence of the SVG file to be created and skips the conversion if it’s already present. This also applies for consecutive calls of dvisvgm with different command-line parameters. If you want to force another conversion of a DVI file that hasn’t changed, you must remove the corresponding SVG files beforehand or add the parameter replace (see below). If you manually set the output pattern to not contain a hash value, the conversion won’t be skipped.

Alternatively, the output pattern may contain the variables %ho and %hc. %ho expands to a 32-bit hash representing the given command-line options that affect the generated SVG output, like --no-fonts and --precision. Different combinations of options and parameters lead to different hashes. Thus pattern %f-%hd-%ho creates filenames that change depending on the DVI data and the given command-line options. Variable %hc provides a combined hash computed from the DVI data and the command-line options. It has the same length as %hd.

Since the page number isn’t part of the file name by default, different DVI pages with identical contents get the same file name. Therefore, only the first one is converted while the others are skipped. To create separate files for each page, you can add the page number to the output pattern, e.g. --output="%f-%p-%hc".

By default, dvisvgm uses the fast XXH64 hash algorithm to compute the values provided through %hd and %hc. 64-bit hashes should be sufficient for most documents with an average size of pages. Alternatively, XXH32 and MD5 can be used as well. The desired algorithm is specified by argument params of option --page-hashes. It takes one of the strings MD5, XXH32, and XXH64, where the names can be given in lower case too, like --page-hashes=md5. Since version 0.7.1, xxHash provides an experimental 128-bit hash function whose algorithm has been stabilized as of version 0.8. When using a version prior to 0.8, the 128-bit hash values can vary depending on the used xxHash version. If the corresponding API is available, dvisvgm supports the new hash function, and option --page-hashes additionally accepts the algorithm specifier XXH128.

Finally, option --page-hashes can take a second argument that must be separated by a comma. Currently, only the two parameters list and replace are evaluated, e.g. --page-hashes=md5,list or --page-hashes=replace. When list is present, dvisvgm doesn’t perform any conversion but just lists the hash values %hd and %hc of the pages specified by option --page. Parameter replace forces dvisvgm to convert a DVI page even if a file with the target name already exists.

  • -P, --pdf — not me switching the input from DVI to PDF, but make it iconic 💅

If this option is given, dvisvgm does not expect a DVI but a PDF input file, and tries to convert it to SVG. Similar to the conversion of DVI files, only the first page is processed by default. Option --page can be used to select different pages, page ranges, and/or page sequences.

If a Ghostscript version < 10.01.0 is found, dvisvgm uses Ghostscript to process the PDF file. In this case, the conversion is realized by creating a single pdffile special command which is forwarded to dvisvgm’s PostScript special handler. Therefore, this option is only available if dvisvgm was built with PostScript support enabled, and requires Ghostscript to be accessible. See option --libgs for further information.

As of Ghostscript 10.01.0, this will no longer work due to the introduction of a new PDF interpreter. Therefore, an alternative conversion module based on mutool, a utility which is part of the MuPDF (https://mupdf.com) package, has been introduced. It’s automatically invoked if Ghostscript can’t be used and if a working mutool executable is present in a directory which is part of the system’s search path.

Alternatively, environment variable DVISVGM_PDF_PROC can be used to select the PDF processor. The currently supported values are gs and mutool.

  • -d, --precision=digits — rounding so clean it’s giving ✨polished✨

Specifies the maximum number of decimal places applied to floating-point attribute values. All attribute values written to the generated SVG file(s) are rounded accordingly. The parameter digits accepts integer values from 0 to 6, where 0 enables the automatic selection of significant decimal places. This is also the default value if dvisvgm is called without option --precision.

  • --progress[=delay] — a little “bestie, I’m working on it” indicator 🌸

Enables a simple progress indicator shown when time-consuming operations like PostScript specials are processed. The indicator doesn’t appear before the given delay (in seconds) has elapsed. The default delay value is 0.5 seconds.

  • -r, --rotate=angle — spins the page clockwise around the page center, periodt 🌀

Rotates the page content clockwise by angle degrees around the page center. This option is equivalent to -TRangle.

  • -R, --relative** — okay so SVG paths can be written with absolute coords or coords relative to wherever the pen currently is. By default dvisvgm builds everything out of absolute commands. Pass --relative and it flips to relative commands instead. Lowkey makes your SVG files smaller in most cases. 💅
  • --stdin — tells dvisvgm to read the DVI or EPS input data from stdin instead of from a file. You can also just drop a single dash (-) instead of typing --stdin, iykyk. The generated SVG is named stdin.svg by default, and you can change that with --output. 🫶
  • -s, --stdout — no file output, bestie. The SVG gets redirected to stdout instead. ✨
  • --tmpdir[=path] — sometimes dvisvgm genuinely needs to make temp files to do its thing. Those go to the system temp folder by default, e.g. /tmp on Linux. Use --tmpdir to point it somewhere else. But — and this is important — dvisvgm does not create that folder if it doesn't exist, so you gotta make sure it's actually there before you run dvisvgm. 😭

So that parallel calls of dvisvgm don't collide, the program doesn't dump files straight into the directory you gave it — it creates a uniquely named subfolder in there and puts the temp files in that. Want to skip the subfolder? Append // or \\ (on Windows) to the directory name. For example, --tmpdir=.// puts the temp files directly in the current working directory, while --tmpdir=. puts them in a dedicated subfolder of the current working directory. 🎀

If you leave out the optional path parameter entirely, dvisvgm just prints the location of the system's temp folder and exits. Cute, informative, done.

  • -a, --trace-all=[retrace] — this forces dvisvgm to vectorize all the glyphs of all fonts referenced in the DVI file, not just the ones actually needed to render the SVG correctly (which is the default). Since dvisvgm stashes tracing results in a font cache, every later conversion of those fonts gets way faster. The boolean option retrace decides what happens to glyphs already sitting in the cache: by default they're skipped, but setting retrace to yes or true forces dvisvgm to retrace the corresponding bitmaps again. 🔥
Note: this option only takes effect if font caching is active. So --trace-all cannot be combined with --cache=none. Periodt.
  • -T, --transform=commands — applies a whole sequence of transformations to the SVG content. Each transformation is a command starting with a capital letter, followed by a list of comma-separated parameters. Here's what's supported: 🦋
    • T tx[,ty] — translates (moves/shifts) the page in the direction of vector (tx,ty). If ty is omitted, ty=0 is assumed. tx and ty are expected in TeX points (1pt = 1/72.27in). There are also several constants defined to make unit conversion easier (see below).
    • S sx[,sy] — scales the page horizontally by sx and vertically by sy. If sy is omitted, sy=sx is assumed.
    • R angle[,x,y] — rotates the page clockwise by angle degrees around point (x,y). If the optional x and y are omitted, the page rotates around its center, depending on the chosen page format. When -bnone is given, the rotation center is origin (0,0).
    • KX angle — skews the page along the x-axis by angle degrees. angle can be any value except 90+180k, where k is an integer. 💀
    • KY angle — skews the page along the y-axis by angle degrees. Same deal: any value except 90+180k, where k is an integer.
    • FH [y] — mirrors (flips) the page at the horizontal line through point (0,y). Omit the optional argument and you get y=h/2, where h is the page height (see the pre-defined constants below).
    • FV [x] — mirrors (flips) the page at the vertical line through point (x,0). Omit the optional argument and you get x=w/2, where w is the page width (see the pre-defined constants below).
    • M m1,...,m6 — applies a transformation described by the 3×3 matrix ((m1,m2,m3),(m4,m5,m6),(0,0,1)), where the inner triples are the rows.
Note: all transformation commands of -T, --transform are applied in the order they appear. Multiple commands can optionally be separated by spaces — in that case you have to wrap the whole transformation string in double quotes to keep them together. All parameters are floating point expressions: plain numbers, or arithmetic terms combined with + (addition), - (subtraction), * (multiplication), / (division) or % (modulo), with the usual associativity and precedence rules. Parentheses are allowed too. 🧮 There are also some pre-defined constants: - ux — horizontal position of upper left page corner in TeX point units - uy — vertical position of upper left page corner in TeX point units - h — page height in TeX point units (0 in case of -bnone) - w — page width in TeX point units (0 in case of -bnone) On top of that you get the 9 length constants pt, bp, cm, mm, in, pc, dd, cc, and sp, e.g. 2cm or 1.6in. So -TT1in,0R45 moves the page content 1 inch to the right and then rotates it 45 degrees around the page center. Obsessed. 💖 For single transformations there are also the short-hand options -c, -t and -r. But heads up: unlike the --transform commands, the order of these options doesn't matter, so you can't describe transformation sequences with them.
  • -t, --translate=tx[,ty] — translates (moves) the page content in the direction of vector (tx,ty). Literally equivalent to -TTtx,ty. 🌈
  • -v, --verbosity=level — controls what kind of messages get printed during a dvisvgm run:
    • 0 — no message output at all
    • 1 — error messages only
    • 2 — warning messages only
    • 4 — informational messages only
    • 8 — user messages only (e.g. created by special dvisvgm:message)
Note: you can combine the categories by adding these values together. The default level is 15, i.e. all messages are printed. 💬
  • -V, --version[=extended] — prints the version of dvisvgm and exits. Set the optional argument to yes and the version numbers of the linked libraries get printed too. 👑
  • -z, --zip[=level] — creates a compressed SVG file with suffix .svgz. The optional argument is the compression level: valid values are in the range of 1 to 9, default is 9. Bigger values mean better compression but may cost slightly more computation time. 🍓
  • -Z, --zoom=factor — multiplies the values of the width and height attributes of the SVG root element by factor, while the coordinate system of the graphic content stays put. Result? Most SVG viewers zoom the graphics accordingly. If you give a negative zoom factor, the width and height attributes are omitted entirely. 🤯

🌈 SUPPORTED SPECIALS 🌈

dvisvgm supports a bunch of special commands that beef up the DVI command set with extra instructions for features like color, graphics, and hyperlinks. The term special command, or just special, comes from the TeX command \special{...} which basically does nothing on its own. TeX only forwards the argument of this command to the DVI file using the corresponding xxx opcode, delegating its execution to the DVI driver. The driver then decides whether to process it or ignore it, depending on what statements it supports. The parameter of TeX's \special command goes in curly braces and can be any arbitrary character sequence representing the actual instruction, for example color Red. 🎨

Worth mentioning: specials are only evaluated when processing DVI files. PS/EPS and PDF know nothing about the whole concept of specials, so they have no means to represent them. 🙈

To evaluate special commands, dvisvgm implements several dedicated handlers. Each handler is responsible for all special statements belonging to the same command set — i.e. commands that usually, but not necessarily, begin with a common prefix. For example, all PDF specials start with the prefix pdf:, while the TPIC special set is 13 different commands with no common identifier. PDF specials get processed by dvisvgm's pdf handler, TPIC ones by the tpic handler. Want a list of the special handlers actually provided? Use option --list-specials (see above). 👆

Here's an overview of the special sets currently supported. The handler names double as the way to disable processing of individual sets of specials — for more on that, see option --no-specials. 🫶

🌸 bgcolor

Special statements for changing the background/page color. Right now dvisvgm only supports the background statement as specified in the dvips manual (https://tug.org/texinfohtml/dvips.html#Color-specifications). Since SVG 1.1 doesn't support background colors, dvisvgm inserts a rectangle of the chosen color into the generated SVG document instead. This rectangle always gets the same size as the selected or computed bounding box. This background color command is part of the color special set but is handled separately, so the user can turn it off. 💅

🎀 color

Statements of this command set give instructions to change the text/paint color. For an overview of the exact syntax, see the documentation of dvips, for instance. ✨

Okay so real talk: dvisvgm takes the dvips color-special syntax and levels it up with two optional modifiers so you can tell fill and stroke apart — like, the color you use to fill in enclosed areas vs. the color you use to draw lines. If one of the color specifiers (a color name, or a color model followed by a sequence of color components) gets preceded by fill or stroke, then only that one color changes. No modifier? Both colors get hit. Example: color push fill rgb 1 0 1 pushes a whole new color pair onto the color stack, but only the fill color is set to magenta while the stroke color keeps its current value. Whereas color push rgb 1 0 1 pushes a color pair where both colors are magenta. 💅

Also bestie, there's a brand-new special in town: color set. Its syntax is literally the same as color push, optional fill and stroke modifiers included. The difference? Instead of pushing a fresh color pair, it modifies the topmost one already on the stack. And if the color stack is empty, it changes the default (black) fill/stroke color. 🎀

🦋 dvisvgm 🦋

Here's the rundown of dvisvgm's own specials that are currently supported. ✨

  • dvisvgm:raw text — Adds an arbitrary sequence of XML nodes to the page section of the SVG document. dvisvgm checks syntax and proper nesting of the inserted elements but does not perform any validation — so it's on you, bestie, to make sure the resulting SVG is still valid. Opening and closing tags may be distributed among different raw specials. The tags themselves can also be split, but a split tag must be continued with the immediately following raw special. Both syntactically incorrect and wrongly nested tags lead to error messages. 😭

The parameter text may also contain the expressions {?x}, {?y}, {?color}, {?fillcolor}, {?strokecolor}, {?matrix}, {?pageno}, {?svgfile}, and {?svgpath} — which respectively expand to the current x and y coordinate, an RGB hex string of the current active fill and stroke color, the current transformation matrix, the current page number, and the relative and absolute path of the SVG file being created. {?color} expands to the same string as {?fillcolor} and is still available for compatibility with older dvisvgm versions that didn't distinguish between fill and stroke colors (see the info about color specials below 👇).

Since SVG doesn't support CMYK colors — say, in fill and stroke attributes — dvisvgm provides the macro {?cmyk(c,m,y,k)} to create an RGB hex string from a CMYK color. You can use it instead of a combo of color/PostScript specials and color macros to directly insert a CMYK color value. 💖

Character sequence {?nl} expands to a newline character. And finally, constructions of the form {?(expr)} let you evaluate mathematical expressions, which may consist of basic arithmetic operations including modulo. As above, the variables x and y represent the current coordinates. Invalid operations, like divisions by zero, lead to an error message on the console. Example: {?(-10*(x+2y)-5)} is a valid expression and expands to the corresponding numeric string. 🔥

  • dvisvgm:rawdef text — Similar to dvisvgm:raw, but it puts the XML nodes into the <defs> section of the SVG document currently being generated. 🫶
  • dvisvgm:rawset name ... dvisvgm:endrawset — This pair of specials marks the begin and end of a definition of a named raw SVG fragment. All the dvisvgm:raw and dvisvgm:rawdef specials enclosed by dvisvgm:rawset and dvisvgm:endrawset are not evaluated immediately — they get stored together under the given name for later use. Once defined, the named fragment can be referenced throughout the DVI file by dvisvgm:rawput (see below). The two commands dvisvgm:rawset and dvisvgm:endrawset must not be nested, i.e. each call of dvisvgm:rawset has to be followed by a corresponding call of dvisvgm:endrawset before another dvisvgm:rawset may occur. Also, the identifier name must be unique throughout the DVI file. Using dvisvgm:rawset multiple times together with the same name leads to warning messages. ⚠️
  • dvisvgm:rawput name — Inserts raw SVG fragments previously stored under the given name. dvisvgm distinguishes between fragments that were specified with dvisvgm:raw or dvisvgm:rawdef, and handles them differently: it inserts all the dvisvgm:raw parts every time dvisvgm:rawput is called, whereas the dvisvgm:rawdef portions go to the <defs> section of the current SVG document only once. 🧚
  • dvisvgm:img width height file — Creates an image element using the current graphic position, transformation matrix, the specified extents, and the given filename. This special command also updates the bounding box of the resulting SVG so it entirely encloses the transformed image. The image itself is positioned such that the lower left corner of the untransformed image lies at the current graphic position. The lengths width and height specify the extents of the untransformed image. For a list of supported length units see option --bbox. Plain numbers without an appended unit specifier imply TeX points (1in = 72.27pt). 🖼️

Parameter file can be a plain filename or a relative or absolute path of a file. By default, dvisvgm just creates links referencing the image file so that the SVG renderer will evaluate the filename or path relative to the location of the generated SVG file. If option --embed-bitmaps is given, JPEG and PNG images are embedded into the SVG document. As this requires base64-encoding of the image data, the resulting SVG files can become quite big — no cap. 💀

  • dvisvgm:bbox lock — Locks the bounding box of the current page and prevents it from further updating, i.e. graphics elements added after calling this special are not taken into account when determining the extent of the bounding box. 🔒
  • dvisvgm:bbox unlock — Unlocks the previously locked bounding box of the current page so that it gets updated again when adding graphics elements to the page. 🔓
  • dvisvgm:bbox n[ew] name — Defines or resets a local bounding box called name. The name may consist of letters and digits. While processing a DVI page, dvisvgm continuously updates the (global) bounding box of the current page in order to determine the minimal rectangle containing all visible page components (characters, images, drawing elements etc.). In addition to the global bounding box, you can request an arbitrary number of named local bounding boxes. Once defined, these boxes get updated together with the global bounding box, starting with the first character that follows the definition. So the local boxes can be used to compute the extent of parts of the page — iconic for scenarios where the generated SVG file is post-processed. 🎯

When used together with the special dvisvgm:raw, the macro {?bbox name} expands to the four values x, y, w, and h (separated by spaces) specifying the coordinates of the upper left corner, width, and height of the local box name. If box name wasn't previously defined, all four values equal zero.

  • dvisvgm:bbox width height [depth] [transform] — Updates the bounding box of the current page by embedding a virtual rectangle (x, y, width, height) where the lower left corner is located at the current DVI drawing position (x,y). If the optional parameter depth is specified, dvisvgm embeds a second rectangle (x, y, width, -depth). The lengths width, height, and depth can be given together with a unit specifier (see option --bbox) or as plain floating point numbers. In the latter case, TeX point units are assumed (1in = 72.27pt). Depending on the size and position of the virtual rectangle, this command either enlarges the overall bounding box or leaves it as is. So it's not possible to reduce its extent. 🙅‍♀️

This special is supposed to be used together with dvisvgm:raw in order to update the viewport of the page properly. By default, the box extents are assigned unchanged and, in particular, are not altered by transformation commands. In order to apply the current transformation matrix, the optional modifier transform can be added at the end of the special statement. ✨

  • dvisvgm:bbox a[bs] x1 y1 x2 y2 [transform] — This variant of the bbox special updates the bounding box by embedding a virtual rectangle (x1,y1,x2,y2). The points (x1,y1) and (x2,y2) denote the absolute coordinates of two diagonal corners of the rectangle. As with the relative special variant described above, the optional modifier transform allows for applying the current transformation matrix to the bounding box. 🦋
  • dvisvgm:bbox f[ix] x1 y1 x2 y2 [transform] — This variant of the bbox special assigns an absolute (final) bounding box to the resulting SVG. After executing this command, dvisvgm doesn't further alter the bounding box coordinates — except if this special is called again later. The points (x1,y1) and (x2,y2) denote the absolute coordinates of two diagonal corners of the rectangle. As with the relative special variant described above, the optional modifier transform allows for applying the current transformation matrix to the bounding box. 👑

The following TeX fragment adds two raw SVG elements to the output and updates the bounding box accordingly:

\special{dvisvgm:raw <circle cx='{?x}' cy='{?y}' r='10' stroke='black' fill='red'/>}%
\special{dvisvgm:bbox 10bp 10bp 10bp transform}%
\special{dvisvgm:bbox -10bp 10bp 10bp transform}

\special{dvisvgm:raw <path d='M50 200 L10 250 H100 Z' stroke='black' fill='blue'/>}%
\special{dvisvgm:bbox abs 10bp 200bp 100bp 250bp transform}

💖 dvisvgm:currentcolor [on|off]

Okay bestie, this one is lowkey iconic: it works kinda like option --currentcolor, but you don’t have to pass an explicit color argument. Instead, it grabs the currently active color and swaps it for currentColor. So the result depends on both where you put the special and any color changes that came before it. The argument on — which activates the currentColor functionality — can be omitted. Call it with off and the functionality gets deactivated again. Slay. ✨

🎀 dvisvgm:message msg

This one prints the given text msg to the console. Besides static text, it may also contain the macros {?...} mentioned in the description of dvisvgm:raw. The messages created by this special can be suppressed with option --verbosity (see above). Cute and useful, ngl. 💅

🦋 em

These specials were introduced with the emTeX distribution by Eberhard Mattes. They provide line drawing statements, instructions for embedding MSP, PCX, and BMP image files, as well as two PCL commands. dvisvgm supports only the line drawing statements and ignores all other em specials silently. A description of the command syntax can be found in the DVI driver documentation coming with emTeX (https://ctan.org/pkg/emtex). So yeah — line drawing gets the love, everything else gets ghosted. 👻

🌈 html

The hyperref specification defines several variants on how to mark hyperlinked areas in a DVI file. dvisvgm supports the plain HyperTeX special constructs as created with hyperref package option hypertex. By default, all linked areas of the document are marked by a rectangle. Option --linkmark allows you to change this behavior. See above for further details. Information on syntax and semantics of the HyperTeX specials can be found in the hyperref manual (https://ctan.org/pkg/hyperref). Hyperlink vibes, fully explained. 🔗✨

🎀 papersize

The papersize special, which is an extension introduced by dvips, can be used to specify the widths and heights of the pages in the DVI file. It affects the page it appears on as well as all following pages until another papersize special is found. If there is more than one papersize special present on a page, dvisvgm applies the last one. However, in order to stay compatible with previous versions of dvisvgm that did not evaluate these specials, their processing must be explicitly enabled by adding option --bbox=papersize on the command-line. Otherwise, dvisvgm ignores them and computes tight bounding boxes. So don’t just assume it’ll read your paper-size mind — you gotta opt in, bestie. 📄💅

🔥 pdf

pdfTeX and dvipdfmx introduced several special commands related to the generation of PDF files. Currently, only pdf:mapfile, pdf:mapline, pdf:pagesize, and PDF hyperlink specials are supported by dvisvgm. The latter are the PDF pendants to the HTML HyperTeX specials. They are generated by the hyperref package in PDF mode, for example. If a DVI page contains unsupported PDF specials, dvisvgm prints a warning message showing the number of ignored special commands. Not me getting warning receipts. 😭

pdf:pagesize is similar to the papersize special (see above) which specifies the size of the current and all following pages. In order to actually apply the extents to the generated SVG files, option --bbox=papersize must be given.

pdf:mapfile and pdf:mapline allow for modifying the font map tree while processing the DVI file. They are used by CTeX, for example. dvisvgm supports both, the dvips and dvipdfm font map format. For further information on the command syntax and semantics, see the documentation of \pdfmapfile in the pdfTeX user manual (https://ctan.org/pkg/pdftex). Font map lore, unlocked. 🗺️✨

💜 ps

The famous DVI driver dvips (https://www.tug.org/texinfohtml/dvips.html) introduced its own set of specials in order to embed PostScript code into DVI files, which greatly improves the capabilities of DVI documents. One aim of dvisvgm is to completely evaluate all PostScript fragments and to convert as many of them as possible to SVG. In contrast to dvips, dvisvgm uses floating point arithmetic to compute the precise position of each graphic element, i.e. it doesn’t round the coordinates. Therefore, the relative locations of the graphic elements may slightly differ from those computed by dvips. Precision era, no cap. 🎯

Since PostScript is a rather complex language, dvisvgm does not implement its own PostScript interpreter but relies on Ghostscript (https://ghostscript.com) instead. If the Ghostscript library was not linked to the dvisvgm binary, it is looked up and loaded dynamically during runtime. In this case, dvisvgm looks for libgs.so.X on Unix-like systems (supported ABI versions: 7,8,9), for libgs.X.dylib on macOS, and for gsdll32.dll or gsdll64.dll on Windows. You can override the default file names with environment variable LIBGS or the command-line option --libgs. The library must be reachable through the ld search path (*nix) or the PATH environment variable (Windows). Alternatively, the absolute file path can be specified. If the library cannot be found, dvisvgm disables the processing of PostScript specials and prints a warning message. Use option --list-specials to check whether PostScript support is available, i.e. entry ps is present. Ghostscript hunt: activated. 🔎💀

The PostScript handler also recognizes and evaluates bounding box data generated by the preview package (https://ctan.org/pkg/preview) with option tightpage. If such data is present in the DVI file and if dvisvgm is called with option --bbox=preview, dvisvgm sets the width and total height of the SVG file to the values derived from the preview data. Additionally, it prints a message showing the width, height, and depth of the box in TeX point units to the console. Especially, the depth value can be read by a post-processor to vertically align the SVG graphics with the baseline of surrounding text in HTML or XSL-FO documents, for example. Please note that SVG bounding boxes are defined by a width and (total) height. In contrast to TeX, SVG provides no means to differentiate between height and depth, i.e. the vertical extents above and below the baseline, respectively. Therefore, it is generally not possible to retrieve the depth value from the SVG file itself. Big caveat, bestie — read it twice. 😭📐

If you call dvisvgm with option --bbox=min (the default) and preview data is present in the DVI file, dvisvgm doesn’t apply the preview extents but computes a bounding box that tightly encloses the page contents. The height, depth and width values written to console are adapted accordingly. Tight box energy. 🎁

🍓 tpic

The TPIC special set defines instructions for drawing simple geometric objects. Some LaTeX packages, like eepic and tplot, use these specials to describe graphics. Simple shapes, cute little drawing instructions — we love to see it. 🦋

✨ EXAMPLES ✨

Okay so here’s the group chat of example commands, and every single one is doing its thing:

  • dvisvgm file — Converts the first page of file.dvi to file.svg.
  • dvisvgm -p1-5 file — Converts the first five pages of file.dvi to file-1.svg,...,file-5.svg.
  • dvisvgm -p1- file — Converts all pages of file.dvi to separate SVG files.
  • dvisvgm -p1,3 -O file — Converts the first and third page of file.dvi to optimized SVG files.
  • dvisvgm - < file.dvi — Converts the first page of file.dvi to stdin.svg where the contents of file.dvi is read from stdin.
  • dvisvgm -z file — Converts the first page of file.dvi to file.svgz with default compression level 9.
  • dvisvgm -p5 -z3 -ba4-l -o newfile file — Converts the fifth page of file.dvi to newfile.svgz with compression level 3. The bounding box is set to DIN/ISO A4 in landscape format.
  • dvisvgm --transform="R20,w/3,2h/5 T1cm,1cm S2,3" file — Converts the first page of file.dvi to file.svg where three transformations are applied.

Periodt. Those examples ate. 💅🔥

🌸 ENVIRONMENT 🌸

dvisvgm uses the kpathsea library for locating the files that it opens. Hence, the environment variables described in the library’s documentation influence the converter. File-finding powers: inherited. ✨

If dvisvgm was linked without the Ghostscript library, and if PostScript support has not been disabled, the shared Ghostscript library is looked up during runtime via dlopen(). The environment variable LIBGS can be used to specify path and file name of the library. Ghostscript, but make it runtime detective work. 🔍

The pre-compiled Windows versions of dvisvgm require a working installation of MiKTeX 2.9 or above. dvisvgm does not work together with the portable edition of MiKTeX because it relies on MiKTeX’s COM interface that is only accessible in a local installation. To enable the evaluation of PostScript specials, the original Ghostscript DLL gsdll32.dll must be present and reachable through the search path. 64-bit Windows builds require the 64-bit Ghostscript DLL gsdll64.dll. Both DLLs come with the corresponding Ghostscript installers available from https://ghostscript.com. Windows besties, this is your lore. 🪟💀

The environment variable DVISVGM_COLORS specifies the colors used to highlight various parts of dvisvgm’s message output. It is only evaluated if option --color is given. The value of DVISVGM_COLORS is a list of colon-separated entries of the form gg=BF, where gg denotes one of the color group indicators listed below, and BF are two hexadecimal digits specifying the background (first digit) and foreground/text color (second digit). The color values are defined as follows: 0=black, 1=red, 2=green, 3=yellow, 4=blue, 5=magenta, 6=cyan, 7=gray, 8=bright red, 9=bright green, A=bright yellow, B=bright blue, C=bright magenta, D=bright cyan, E=bright gray, F=white. Depending on the terminal, the colors may differ. Rather than changing both the text and background color, it’s also possible to change only one of them: An asterisk (*) in place of a hexadecimal digit indicates the default text or background color of the terminal. Color customisation era. 🌈✨

All malformed entries in the list are silently ignored. No drama, no error message — just silently ignored. 🙈

  • er — error messages
  • wn — warning messages
  • pn — messages about page numbers
  • ps — page size messages
  • fw — information about the files written
  • sm — state messages
  • tr — messages of the glyph tracer
  • pi — progress indicator

Example: er=01:pi=*5 sets the colors of error messages (er) to red (1) on black (0), and those of progress indicators (pi) to cyan (5) on default background (*). Obsessed with a well-coloured terminal, honestly. 🎀

Variable DVISVGM_PDF_PROC allows you to select the desired PDF processor. dvisvgm requires either Ghostscript (version < 10.01.0) or mutool to process PDF files. If DVISVGM_PDF_PROC is not set, dvisvgm tries to detect the available processors. See option --pdf for further information. To select a specific PDF processor, you can set DVISVGM_PDF_PROC to gs or mutool which forces the use of Ghostscript and mutool respectively. Processor choice: handled. 💅

🗂️ FILES 🗂️

okay so where do these files actually live? the kpathsea library is the one deciding that, bestie. if you wanna check the actual kpathsea configuration, just use the kpsewhich utility.

here’s the lil file-pattern squad it’s gonna be looking for:

  • *.enc — font encoding files
  • *.fgd — font glyph data files (cache files created by dvisvgm)
  • *.map — font map files
  • *.mf — Metafont input files
  • *.pfb — PostScript Type 1 font files
  • *.pro — PostScript header/prologue files
  • *.tfm — TeX font metric files
  • *.ttf — TrueType font files
  • *.vf — Virtual font files

👀 SEE ALSO 👀

go read tex(1), mf(1), mktexmf(1), grodvi(1), potrace(1), and the kpathsea library info documentation. iykyk 💅

🌈 RESOURCES 🌈

💀 BUGS 💀

found a bug? pls don’t just scream into the void, bestie — report it using the bug tracker at GitHub: https://github.com/mgieseki/dvisvgm/issues. the maintainers will love you for it 🫶

👑 AUTHOR 👑

Written by Martin Gieseking <martin.gieseking@uos.de> — an actual icon, no cap ✨

🎀 COPYING 🎀

Copyright © 2005-2026 Martin Gieseking. Free use of this software is granted under the terms of the GNU General Public License (GPL) version 3 or, (at your option) any later version.

okay that’s the whole manual, bestie!! source/version: dvisvgm 3.6.1; date: 2026-08-03 — now go make some iconic SVGs 💅✨🦋

📦 where this page comes from