The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Imager::Files - working with image files


  use Imager;
  my $img = ...;
  $img->write(file=>$filename, type=>$type)
    or die "Cannot write: ",$img->errstr;

  # type is optional if we can guess the format from the filename
  $img->write(file => "foo.png")
    or die "Cannot write: ",$img->errstr;

  $img = Imager->new;
  $img->read(file=>$filename, type=>$type)
    or die "Cannot read: ", $img->errstr;

  # type is optional if we can guess the type from the file data
  # and we normally can guess
  $img->read(file => $filename)
    or die "Cannot read: ", $img->errstr;

  Imager->write_multi({ file=> $filename, ... }, @images)
    or die "Cannot write: ", Imager->errstr;

  my @imgs = Imager->read_multi(file=>$filename)
    or die "Cannot read: ", Imager->errstr;

  Imager->set_file_limits(width=>$max_width, height=>$max_height)

  my @read_types = Imager->read_types;
  my @write_types = Imager->write_types;

  # we can write/write_multi to things other than filenames
  my $data;
  $img->write(data => \$data, type => $type) or die;

  open my $fh, "+>:raw", ... ;
  $img->write(fh => $fh, type => $type) or die;

  $img->write(fd => fileno($fh), type => $type) or die;

  # some file types need seek callbacks too
  $img->write(callback => \&write_callback, type => $type) or die;

  # and similarly for read/read_multi
  $img->read(data => $data) or die;
  $img->read(fh => $fh) or die;
  $img->read(fd => fileno($fh)) or die;
  $img->read(callback => \&read_callback) or die;

  use Imager 0.68;
  my $img = Imager->new(file => $filename)
    or die Imager->errstr;

  Imager->add_file_magic(name => $name, bits => $bits, mask => $mask);


You can read and write a variety of images formats, assuming you have the appropriate libraries, and images can be read or written to/from files, file handles, file descriptors, scalars, or through callbacks.

To see which image formats Imager is compiled to support the following code snippet is sufficient:

  use Imager;
  print join " ", keys %Imager::formats;

This will include some other information identifying libraries rather than file formats. For new code you might find the "read_types()" or "write_types()" methods useful.


Reading writing to and from files is simple, use the read() method to read an image:

  my $img = Imager->new;
  $img->read(file=>$filename, type=>$type)
    or die "Cannot read $filename: ", $img->errstr;

In most cases Imager can auto-detect the file type, so you can just supply the file name:

  $img->read(file => $filename)
    or die "Cannot read $filename: ", $img->errstr;

The read() method accepts the allow_incomplete parameter. If this is non-zero then read() can return true on an incomplete image and set the i_incomplete tag.

From Imager 0.68 you can supply most read() parameters to the new() method to read the image file on creation. If the read fails, check Imager->errstr() for the cause:

  use Imager 0.68;
  my $img = Imager->new(file => $filename)
    or die "Cannot read $filename: ", Imager->errstr;

and the write() method to write an image:

  $img->write(file=>$filename, type=>$type)
    or die "Cannot write $filename: ", $img->errstr;

If you're reading from a format that supports multiple images per file, use the read_multi() method:

  my @imgs = Imager->read_multi(file=>$filename, type=>$type)
    or die "Cannot read $filename: ", Imager->errstr;

As with the read() method, Imager will normally detect the type automatically.


and if you want to write multiple images to a single file use the write_multi() method:

  Imager->write_multi({ file=> $filename, type=>$type }, @images)
    or die "Cannot write $filename: ", Imager->errstr;

This is a class method that returns a list of the image file types that Imager can read.

  my @types = Imager->read_types;

These types are the possible values for the type parameter, not necessarily the extension of the files you're reading.

It is possible for extra file read handlers to be loaded when attempting to read a file, which may modify the list of available read types.


This is a class method that returns a list of the image file types that Imager can write.

  my @types = Imager->write_types;

Note that these are the possible values for the type parameter, not necessarily the extension of the files you're writing.

It is possible for extra file write handlers to be loaded when attempting to write a file, which may modify the list of available write types.

When writing, if the filename includes an extension that Imager recognizes, then you don't need the type, but you may want to provide one anyway. See "Guessing types" for information on controlling this recognition.

The type parameter is a lowercase representation of the file type, and can be any of the following:

  bmp   Windows BitMaP (BMP)
  gif   Graphics Interchange Format (GIF)
  jpeg  JPEG/JFIF
  png   Portable Network Graphics (PNG)
  pnm   Portable aNyMap (PNM)
  raw   Raw
  sgi   SGI .rgb files
  tga   TARGA
  tiff  Tagged Image File Format (TIFF)

Support for other formats can be found on CPAN, including:

  apng        Imager::File::APNG
  avif        Imager::File::AVIF
  heif/heic   Imager::File::HEIF
  qoi         Imager::File::QOI
  webp        Imager::File::WEBP

When you read an image, Imager may set some tags, possibly including information about the spatial resolution, textual information, and animation information. See "Tags" in Imager::ImageTypes for specifics.

The open() method is a historical alias for the read() method.

Input and output

When reading or writing you can specify one of a variety of sources or targets:

  • file - The file parameter is the name of the image file to be written to or read from. If Imager recognizes the extension of the file you do not need to supply a type.

      # write in tiff format
      $image->write(file => "example.tif")
        or die $image->errstr;
      $image->write(file => 'foo.tmp', type => 'tiff')
        or die $image->errstr;
      my $image = Imager->new;
      $image->read(file => 'example.tif')
        or die $image->errstr;
  • fh - fh is a file handle, typically returned from an open call. You should call binmode on the handle before passing it to Imager.

    Imager will set the handle to autoflush to make sure any buffered data is flushed , since Imager will write to the file descriptor (from fileno()) rather than writing at the perl level.

      $image->write(fh => \*STDOUT, type => 'gif')
        or die $image->errstr;
      # for example, a file uploaded via
      $image->read(fd => $cgi->param('file')) 
        or die $image->errstr;
  • fd - fd is a file descriptor. You can get this by calling the fileno() function on a file handle, or by using one of the standard file descriptor numbers.

    If you get this from a perl file handle, you may need to flush any buffered output, otherwise it may appear in the output stream after the image.

      $image->write(fd => file(STDOUT), type => 'gif')
        or die $image->errstr;
  • data - When reading data, data is a scalar containing the image file data, or a reference to such a scalar. When writing, data is a reference to the scalar to save the image file data to.

      my $data;
      $image->write(data => \$data, type => 'tiff')
        or die $image->errstr;
      my $data = $row->{someblob}; # eg. from a database
      my @images = Imager->read_multi(data => $data)
        or die Imager->errstr;
      # from Imager 0.99
      my @images = Imager->read_multi(data => \$data)
        or die Imager->errstr;
  • callback, readcb, writecb, seekcb, closecb - Imager will make calls back to your supplied coderefs to read, write and seek from/to/through the image file. See "I/O Callbacks" below for details.

  • io - an Imager::IO object.

By default Imager will use buffered I/O when reading or writing an image. You can disabled buffering for output by supplying a buffered => 0 parameter to write() or write_multi().

I/O Callbacks

When reading from a file you can use either callback or readcb to supply the read callback, and when writing callback or writecb to supply the write callback.

Whether reading or writing a TIFF image, seekcb and readcb are required.

If a file handler attempts to use readcb, writecb or seekcb and you haven't supplied one, the call will fail, failing the image read or write, returning an error message indicating that the callback is missing:

  # attempting to read a TIFF image without a seekcb
  open my $fh, "<", $filename or die;
  my $rcb = sub {
    my $val;
    read($fh, $val, $_[0]) or return "";
    return $val;
  my $im = Imager->new(callback => $rcb)
    or die Imager->errstr
  # dies with (wrapped here):
  # Error opening file: (Iolayer): Failed to read directory at offset 0:
  # (Iolayer): Seek error accessing TIFF directory: seek callback called
  # but no seekcb supplied

You can also provide a closecb parameter called when writing the file is complete. If no closecb is supplied the default will succeed silently.

  # contrived
  my $data;
  sub mywrite {
    $data .= unpack("H*", shift);
  Imager->write_multi({ callback => \&mywrite, type => 'gif'}, @images)
    or die Imager->errstr;


The read callback is called with 2 parameters:

  • size - the minimum amount of data required.

  • maxsize - previously this was the maximum amount of data returnable - currently it's always the same as size

Your read callback should return the data as a scalar:

  • on success, a string containing the bytes read.

  • on end of file, an empty string

  • on error, undef.

If your return value contains more data than size Imager will panic.

Your return value must not contain any characters over \xFF or Imager will panic.


Your write callback takes exactly one parameter, a scalar containing the data to be written.

Return true for success.


The seek callback takes 2 parameters, a POSITION, and a WHENCE, defined in the same way as perl's seek function.

Previously you always needed a seekcb callback if you called Imager's "read()" or "read_multi()" without a type parameter, but this is no longer necessary unless the file handler requires seeking, such as for TIFF files.

Returns the new position in the file, or -1 on failure.


You can also supply a closecb which is called with no parameters when there is no more data to be written. This could be used to flush buffered data.

Return true on success.

Guessing types

When writing to a file, if you don't supply a type parameter Imager will attempt to guess it from the file name. This is done by calling the code reference stored in $Imager::FORMATGUESS. This is only done when write() or write_multi() is called with a file parameter, or if read() or read_multi() can't determine the type from the file's header.

The default function value of $Imager::FORMATGUESS is \&Imager::def_guess_type.


This is the default function Imager uses to derive a file type from a file name. This is a function, not a method.

Accepts a single parameter, the file name and returns the type or undef.

You can replace function with your own implementation if you have some specialized need. The function takes a single parameter, the name of the file, and should return either a file type or under.

  # I'm writing jpegs to weird filenames
  local $Imager::FORMATGUESS = sub { 'jpeg' };

When reading a file Imager examines beginning of the file for identifying information. The current implementation attempts to detect the following image types beyond those supported by Imager:

    xpm, mng, jng, ilbm, pcx, fits, psd (Photoshop), eps, Utah RLE.

You can now add to the magic database Imager uses for detecting file types:

  Imager->add_file_magic(name => $name, bits => $bits, mask => $mask)

Adds to list of magic, the parameters are all required. The parameters are:

  • name - the file type name to return on match.

  • bits - a binary string to match.

  • mask - a mask controlling which parts of bits are significant.

While mask is mostly a bit mask, some byte values are translated, the space character is treated as all zeros ("\x00"), and the x character as all ones ("\xFF").

New magic entries take priority over old entries.

You can add more than one magic entry for a given name.

  Imager->add_file_magic(name => "heif",
                         bits => "    ftypheif"
                         mask => "    xxxxxxxx");

Limiting the sizes of images you read


In some cases you will be receiving images from an untested source, such as submissions via CGI. To prevent such images from consuming large amounts of memory, you can set limits on the dimensions of images you read from files:

  • width - limit the width in pixels of the image

  • height - limit the height in pixels of the image

  • bytes - limits the amount of storage used by the image. This depends on the width, height, channels and sample size of the image. For paletted images this is calculated as if the image was expanded to a direct color image.

To set the limits, call the class method set_file_limits:

  Imager->set_file_limits(width=>$max_width, height=>$max_height);

You can pass any or all of the limits above, any limits you do not pass are left as they were.

Any limit of zero for width or height is treated as unlimited.

A limit of zero for bytes is treated as one gigabyte, but higher bytes limits can be set explicitly.

By default, the width and height limits are zero, or unlimited. The default memory size limit is one gigabyte.

You can reset all limits to their defaults with the reset parameter:

  # no limits

This can be used with the other limits to reset all but the limit you pass:

  # only width is limited
  Imager->set_file_limits(reset=>1, width=>100);

  # only bytes is limited
  Imager->set_file_limits(reset=>1, bytes=>10_000_000);

You can get the current limits with the get_file_limits() method:

  my ($max_width, $max_height, $max_bytes) =

Intended for use by file handlers to check that the size of a file is within the limits set by set_file_limits().


  • width, height - the width and height of the image in pixels. Must be a positive integer. Required.

  • channels - the number of channels in the image, including the alpha channel if any. Must be a positive integer between 1 and 4 inclusive. Default: 3.

  • sample_size - the number of bytes stored per sample. Must be a positive integer or "float". Note that this should be the sample size of the Imager image you will be creating, not the sample size in the source, eg. if the source has 32-bit samples this should be "float" since Imager doesn't have 32-bit/sample images.


The different image formats can write different image type, and some have different options to control how the images are written.

When you call write() or write_multi() with an option that has the same name as a tag for the image format you're writing, then the value supplied to that option will be used to set the corresponding tag in the image. Depending on the image format, these values will be used when writing the image.

This replaces the previous options that were used when writing GIF images. Currently if you use an obsolete option, it will be converted to the equivalent tag and Imager will produced a warning. You can suppress these warnings by calling the Imager::init() function with the warn_obsolete option set to false:


At some point in the future these obsolete options will no longer be supported.

PNM (Portable aNy Map)

Imager can write PGM (Portable Gray Map) and PPM (Portable PixMaps) files, depending on the number of channels in the image. Currently the images are written in binary formats. Only 1 and 3 channel images can be written, including 1 and 3 channel paletted images.

  $img->write(file=>'foo.ppm') or die $img->errstr;

Imager can read both the ASCII and binary versions of each of the PBM (Portable BitMap), PGM and PPM formats.

  $img->read(file=>'foo.ppm') or die $img->errstr;

PNM does not support the spatial resolution tags.

The following tags are set when reading a PNM file:

  • pnm_maxval - the maxvals number from the PGM/PPM header. Always set to 2 for a PBM file.

  • pnm_type - the type number from the PNM header, 1 for ASCII PBM files, 2 for ASCII PGM files, 3 for ASCII c<PPM> files, 4 for binary PBM files, 5 for binary PGM files, 6 for binary PPM files.

The following tag is checked when writing an image with more than 8-bits/sample:

  • pnm_write_wide_data - if this is non-zero then write() can write PGM/PPM files with 16-bits/sample. Some applications, for example GIMP 2.2, and tools can only read 8-bit/sample binary PNM files, so Imager will only write a 16-bit image when this tag is non-zero.


You can supply a jpegquality parameter ranging from 0 (worst quality) to 100 (best quality) when writing a JPEG file, which defaults to 75.

  $img->write(file=>'foo.jpg', jpegquality=>90) or die $img->errstr;

If you write an image with an alpha channel to a JPEG file then it will be composed against the background set by the i_background parameter (or tag), or black if not supplied.

Imager will read a gray scale JPEG as a 1 channel image and a color JPEG as a 3 channel image.

  $img->read(file=>'foo.jpg') or die $img->errstr;

The following tags are set in a JPEG image when read, and can be set to control output:

  • jpeg_density_unit - The value of the density unit field in the JFIF header. This is ignored on writing if the i_aspect_only tag is non-zero.

    The i_xres and i_yres tags are expressed in pixels per inch no matter the value of this tag, they will be converted to/from the value stored in the JPEG file.

  • jpeg_density_unit_name - This is set when reading a JPEG file to the name of the unit given by jpeg_density_unit. Possible results include inch, centimeter, none (the i_aspect_only tag is also set reading these files). If the value of jpeg_density_unit is unknown then this tag isn't set.

  • jpeg_comment - Text comment.

  • jpeg_progressive - Whether the JPEG file is a progressive file. (Imager 0.84)

JPEG supports the spatial resolution tags i_xres, i_yres and i_aspect_only.

You can also set the following tags when writing to an image, they are not set in the image when reading:

  • jpeg_optimize - set to a non-zero integer to compute optimal Huffman coding tables for the image. This will increase memory usage and processing time (about 12% in my simple tests) but can significantly reduce file size without a loss of quality.

  • jpeg_compress_profile - set to either fastest, the default, or only with MozJPEG, to max. Setting this to any other value will cause an error, failing the write() call. Setting this to max without MozJPEG will cause an error.

      use Imager::File::JPEG;
      my $prof = Imager::File::JPEG->is_mozjpeg ? "max" : "fastest";
      $im->write(file => "foo.jpeg", jpeg_compress_profile => $prof)
         or die $im->errstr;

    Note that unlike MozJPEG, Imager always defaults to fastest.

  • jpeg_tune - corresponds to the MozJPEG cjpeg -tune parameters, this can be any of:

    • psnr

    • ssim

    • ms-ssim

    • hvs-psnr

    These set the same values as the jpeg_base_quant_tbl_idx, jpeg_lambda_log_scale1, jpeg_lambda_log_scale2 and jpeg_use_lambda_weight_tbl tags, which can override the values set by jpeg_tune.

    Unlike cjpeg -tune, jpeg_tune doesn't force the quality to 75.

    Requires that Imager::File::JPEG was built with MozJPEG.

  • jpeg_arithmetic - if set to non-zero, use arithmetic coding when writing the image. This requires that the libjpeg variant that Imager::File::JPEG was built with had encode support for arithmetic coding enabled when it was built.

  • jpeg_jfif - if set to zero, disable writing the JFIF header even if one is technically required, which it is for the color spaces Imager works with. This saves 18 bytes in the output file and most applications including web browsers will successfully read the file, but your experience may differ.

    This will prevent the i_xres, i_yres and jpeg_density_units tags from having an effect.

  • jpeg_smooth - if set to a value from 1 to 100 apply smoothing to the image to eliminate dithering noise (without modifying the Imager image). Default: 0 (no smoothing).

  • jpeg_restart - if set to a plain integer N, generate a JPEG restart marker every N rows, if set to an integer followed by B, generate a JPEG restart marker every N MCUs, the 8x8 pixel blocks that make up a JPEG image. Note that if this is specified in rows it will be translated to MCUs by libjpeg.

    This allows a decoder to recover from corruption, but it makes the file slightly larger.

    Default: 0, no restart markers are produced.

  • jpeg_sample - control subsampling of the YCbCr components, equivalent to the -sample parameter for cjpeg.

    Default: "2x2,1x1,1x1", default 4:2:0 JPEG subsampling.

The following options can be set on writing with MozJPEG and correspond directly to the API options described in the MozJPEG README-mozilla.txt. Attempting to set them when Imager::File::JPEG has been built with another libjpeg will result in an error.

  • jpeg_optimize_scans, jpeg_trellis_quant, jpeg_trellis_quant_dc, jpeg_tresllis_eob_opt, jpeg_use_lambda_weight_tbl, jpeg_use_scans_in_trellis, jpeg_overshoot_deringing - boolean options, set to 0 or 1.

    jpeg_lambda_log_scale1, jpeg_lambda_log_scale2, jpeg_trellis_delta_dc_weight - integer options.

    jpeg_trellis_freq_split, jpeg_trellis_num_loops, jpeg_base_quant_tbl_idx, jpeg_dc_scan_opt_mode - floating point options.

When reading a JPEG image, the following tag will be set, and ignored on writing:

  • jpeg_read_arithmetic - the image had been written with arithmetic coding. This uses a separate name from writing to avoid arithmetic coding from being accidentally propagated to a file that might need to be read by an implementation without arithmetic coding support.

If an APP1 block containing EXIF information is found, then any of the following tags can be set when reading a JPEG image:

    exif_aperture exif_artist exif_brightness exif_color_space exif_contrast exif_copyright exif_custom_rendered exif_date_time exif_date_time_digitized exif_date_time_original exif_digital_zoom_ratio exif_exposure_bias exif_exposure_index exif_exposure_mode exif_exposure_program exif_exposure_time exif_f_number exif_flash exif_flash_energy exif_flashpix_version exif_focal_length exif_focal_length_in_35mm_film exif_focal_plane_resolution_unit exif_focal_plane_x_resolution exif_focal_plane_y_resolution exif_gain_control exif_image_description exif_image_unique_id exif_iso_speed_rating exif_make exif_max_aperture exif_metering_mode exif_model exif_orientation exif_related_sound_file exif_resolution_unit exif_saturation exif_scene_capture_type exif_sensing_method exif_sharpness exif_shutter_speed exif_software exif_spectral_sensitivity exif_sub_sec_time exif_sub_sec_time_digitized exif_sub_sec_time_original exif_subject_distance exif_subject_distance_range exif_subject_location exif_tag_light_source exif_user_comment exif_version exif_white_balance exif_x_resolution exif_y_resolution

The following derived tags can also be set when reading a JPEG image:

    exif_color_space_name exif_contrast_name exif_custom_rendered_name exif_exposure_mode_name exif_exposure_program_name exif_flash_name exif_focal_plane_resolution_unit_name exif_gain_control_name exif_light_source_name exif_metering_mode_name exif_resolution_unit_name exif_saturation_name exif_scene_capture_type_name exif_sensing_method_name exif_sharpness_name exif_subject_distance_range_name exif_white_balance_name

The derived tags are for enumerated fields, when the value for the base field is valid then the text that appears in the EXIF specification for that value appears in the derived field. So for example if exf_metering_mode is 5 then exif_metering_mode_name is set to Pattern.


  my $image = Imager->new;
  $image->read(file => 'exiftest.jpg')
    or die "Cannot load image: ", $image->errstr;
  print $image->tags(name => "exif_image_description"), "\n";
  print $image->tags(name => "exif_exposure_mode"), "\n";
  print $image->tags(name => "exif_exposure_mode_name"), "\n";

  # for the exiftest.jpg in the Imager distribution the output would be:
  Imager Development Notes
  Auto exposure

Imager will not write EXIF tags to any type of image, if you need more advanced EXIF handling, consider Image::ExifTool.


Historically, Imager saves IPTC data when reading a JPEG image, the parseiptc() method returns a list of key/value pairs resulting from a simple decoding of that data.

Any future IPTC data decoding is likely to go into tags.


When writing one of more GIF images you can use the same Quantization Options as you can when converting an RGB image into a paletted image.

When reading a GIF all of the sub-images are combined using the screen size and image positions into one big image, producing an RGB image. This may change in the future to produce a paletted image where possible.

When you read a single GIF with $img->read() you can supply a reference to a scalar in the colors parameter, if the image is read the scalar will be filled with a reference to an anonymous array of Imager::Color objects, representing the palette of the image. This will be the first palette found in the image. If you want the palettes for each of the images in the file, use read_multi() and use the getcolors() method on each image.

GIF does not support the spatial resolution tags.

Imager will set the following tags in each image when reading, and can use most of them when writing to GIF:

  • gif_left - the offset of the image from the left of the "screen" ("Image Left Position")

  • gif_top - the offset of the image from the top of the "screen" ("Image Top Position")

  • gif_interlace - non-zero if the image was interlaced ("Interlace Flag")

  • gif_screen_width, gif_screen_height - the size of the logical screen. When writing this is used as the minimum. If any image being written would extend beyond this then the screen size is extended. ("Logical Screen Width", "Logical Screen Height").

  • gif_local_map - Non-zero if this image had a local color map. If set for an image when writing the image is quantized separately from the other images in the file.

  • gif_background - The index in the global color map of the logical screen's background color. This is only set if the current image uses the global color map. You can set this on write too, but for it to choose the color you want, you will need to supply only paletted images and set the gif_eliminate_unused tag to 0.

  • gif_trans_index - The index of the color in the color map used for transparency. If the image has a transparency then it is returned as a 4 channel image with the alpha set to zero in this palette entry. This value is not used when writing. ("Transparent Color Index")

  • gif_trans_color - A reference to an Imager::Color object, which is the color to use for the palette entry used to represent transparency in the palette. You need to set the transp option (see "Quantization options" in Imager::ImageTypes) for this value to be used.

  • gif_delay - The delay until the next frame is displayed, in 1/100 of a second. ("Delay Time").

  • gif_user_input - whether or not a user input is expected before continuing (view dependent) ("User Input Flag").

  • gif_disposal - how the next frame is displayed ("Disposal Method")

  • gif_loop - the number of loops from the Netscape Loop extension. This may be zero to loop forever.

  • gif_comment - the first block of the first GIF comment before each image.

  • gif_eliminate_unused - If this is true, when you write a paletted image any unused colors will be eliminated from its palette. This is set by default.

  • gif_colormap_size - the original size of the color map for the image. The color map of the image may have been expanded to include out of range color indexes.

Where applicable, the ("name") is the name of that field from the GIF89 standard.

The following GIF writing options are obsolete, you should set the corresponding tag in the image, either by using the tags functions, or by supplying the tag and value as options.

  • gif_each_palette - Each image in the GIF file has it's own palette if this is non-zero. All but the first image has a local color table (the first uses the global color table.

    Use gif_local_map in new code.

  • interlace - The images are written interlaced if this is non-zero.

    Use gif_interlace in new code.

  • gif_delays - A reference to an array containing the delays between images, in 1/100 seconds.

    Use gif_delay in new code.

  • gif_positions - A reference to an array of references to arrays which represent screen positions for each image.

    New code should use the gif_left and gif_top tags.

  • gif_loop_count - If this is non-zero the Netscape loop extension block is generated, which makes the animation of the images repeat.

    This is currently unimplemented due to some limitations in giflib.

You can supply a page parameter to the read() method to read some page other than the first. The page is 0 based:

  # read the second image in the file
  $image->read(file=>"example.gif", page=>1)
    or die "Cannot read second page: ",$image->errstr,"\n";

Before release 0.46, Imager would read multiple image GIF image files into a single image, overlaying each of the images onto the virtual GIF screen.

As of 0.46 the default is to read the first image from the file, as if called with page => 0.

You can return to the previous behavior by calling read with the gif_consolidate parameter set to a true value:

  $img->read(file=>$some_gif_file, gif_consolidate=>1);

As with the to_paletted() method, if you supply a colors parameter as a reference to an array, this will be filled with Imager::Color objects of the color table generated for the image file.

TIFF (Tagged Image File Format)

Imager can write images to either paletted or RGB TIFF images, depending on the type of the source image.

When writing direct color images to TIFF the sample size of the output file depends on the input:

  • double/sample - written as 32-bit/sample TIFF

  • 16-bit/sample - written as 16-bit/sample TIFF

  • 8-bit/sample - written as 8-bit/sample TIFF

For paletted images:

  • $img->is_bilevel is true - the image is written as bi-level

  • otherwise - image is written as paletted.

If you are creating images for faxing you can set the class parameter set to fax. By default the image is written in fine mode, but this can be overridden by setting the fax_fine parameter to zero. Since a fax image is bi-level, Imager uses a threshold to decide if a given pixel is black or white, based on a single channel. For gray scale images channel 0 is used, for color images channel 1 (green) is used. If you want more control over the conversion you can use $img->to_paletted() to product a bi-level image. This way you can use dithering:

  my $bilevel = $img->to_paletted(make_colors => 'mono',
                                  translate => 'errdiff',
                                  errdiff => 'stucki');
  • class - If set to 'fax' the image will be written as a bi-level fax image.

  • fax_fine - By default when class is set to 'fax' the image is written in fine mode, you can select normal mode by setting fax_fine to 0.

Imager should be able to read any TIFF image you supply. Paletted TIFF images are read as paletted Imager images, since paletted TIFF images have 16-bits/sample (48-bits/color) this means the bottom 8-bits are lost, but this shouldn't be a big deal.

TIFF supports the spatial resolution tags. See the tiff_resolutionunit tag for some extra options.

As of Imager 0.62 Imager reads:

  • 8-bit/sample gray, RGB or CMYK images, including a possible alpha channel as an 8-bit/sample image.

  • 16-bit gray, RGB, or CMYK image, including a possible alpha channel as a 16-bit/sample image.

  • 32-bit gray, RGB image, including a possible alpha channel as a double/sample image.

  • bi-level images as paletted images containing only black and white, which other formats will also write as bi-level.

  • tiled paletted images are now handled correctly

  • other images are read using tifflib's RGBA interface as 8-bit/sample images.

The following tags are set in a TIFF image when read, and can be set to control output:

  • tiff_compression - When reading an image this is set to the numeric value of the TIFF compression tag.

    On writing you can set this to either a numeric compression tag value, or one of the following values:

      Ident     Number  Description
      none         1    No compression
      packbits   32773  Macintosh RLE
      ccittrle     2    CCITT RLE
      fax3         3    CCITT Group 3 fax encoding (T.4)
      t4           3    As above
      fax4         4    CCITT Group 4 fax encoding (T.6)
      t6           4    As above
      lzw          5    LZW
      jpeg         7    JPEG
      zip          8    Deflate (GZIP) Non-standard
      deflate      8    As above.
      oldzip     32946  Deflate with an older code.
      ccittrlew  32771  Word aligned CCITT RLE

    In general a compression setting will be ignored where it doesn't make sense, eg. jpeg will be ignored for compression if the image is being written as bilevel.

    Imager attempts to check that your build of libtiff supports the given compression, and will fallback to packbits if it isn't enabled. eg. older distributions didn't include LZW compression, and JPEG compression is only available if libtiff is configured with libjpeg's location.

      $im->write(file => 'foo.tif', tiff_compression => 'lzw')
        or die $im->errstr;
  • tags, tiff_jpegqualitytiff_jpegquality - If tiff_compression is jpeg then this can be a number from 1 to 100 giving the JPEG compression quality. High values are better quality and larger files.

  • tiff_resolutionunit - The value of the ResolutionUnit tag. This is ignored on writing if the i_aspect_only tag is non-zero.

    The i_xres and i_yres tags are expressed in pixels per inch no matter the value of this tag, they will be converted to/from the value stored in the TIFF file.

  • tiff_resolutionunit_name - This is set when reading a TIFF file to the name of the unit given by tiff_resolutionunit. Possible results include inch, centimeter, none (the i_aspect_only tag is also set reading these files) or unknown.

  • tiff_bitspersample - Bits per sample from the image. This value is not used when writing an image, it is only set on a read image.

  • tiff_photometric - Value of the PhotometricInterpretation tag from the image. This value is not used when writing an image, it is only set on a read image.

  • tiff_documentname, tiff_imagedescription, tiff_make, tiff_model, tiff_pagename, tiff_software, tiff_datetime, tiff_artist, tiff_hostcomputer - Various strings describing the image. tiff_datetime must be formatted as "YYYY:MM:DD HH:MM:SS". These correspond directly to the mixed case names in the TIFF specification. These are set in images read from a TIFF and saved when writing a TIFF image.

You can supply a page parameter to the read() method to read some page other than the first. The page is 0 based:

  # read the second image in the file
  $image->read(file=>"example.tif", page=>1)
    or die "Cannot read second page: ",$image->errstr,"\n";

If you read an image with multiple alpha channels, then only the first alpha channel will be read.

When reading a TIFF image with callbacks, the seekcb callback parameter is also required.

When writing a TIFF image with callbacks, the seekcb and readcb parameters are also required.

TIFF is a random access file format, it cannot be read from or written to unseekable streams such as pipes or sockets.

BMP (Windows Bitmap)

Imager can write 24-bit RGB, and 8, 4 and 1-bit per pixel paletted Windows BMP files. Currently you cannot write compressed BMP files with Imager.

Imager can read 24-bit RGB, and 8, 4 and 1-bit perl pixel paletted Windows BMP files. There is some support for reading 16-bit per pixel images, but I haven't found any for testing.

BMP has no support for multiple image files.

BMP files support the spatial resolution tags, but since BMP has no support for storing only an aspect ratio, if i_aspect_only is set when you write the i_xres and i_yres values are scaled so the smaller is 72 DPI.

The following tags are set when you read an image from a BMP file:


The type of compression, if any. This can be any of the following values:

BI_RGB (0)


BI_RLE8 (1)

8-bits/pixel paletted value RLE compression.

BI_RLE4 (2)

4-bits/pixel paletted value RLE compression.


Packed RGB values.


The bmp_compression value as a BI_* string


The number of important colors as defined by the writer of the image.


Number of color used from the BMP header


The file size from the BMP header


Number of bits stored per pixel. (24, 8, 4 or 1)

TGA (Targa)

When storing Targa images RLE compression can be activated with the compress parameter, the idstring parameter can be used to set the Targa comment field and the wierdpack option can be used to use the 15 and 16 bit Targa formats for RGB and RGBA data. The 15 bit format has 5 of each red, green and blue. The 16 bit format in addition allows 1 bit of alpha. The most significant bits are used for each channel.




When reading raw images you need to supply the width and height of the image in the xsize and ysize options:

  $img->read(file=>'foo.raw', xsize=>100, ysize=>100)
    or die "Cannot read raw image\n";

If your input file has more channels than you want, or (as is common), junk in the fourth channel, you can use the raw_datachannels and raw_storechannels options to control the number of channels in your input file and the resulting channels in your image. For example, if your input image uses 32-bits per pixel with red, green, blue and junk values for each pixel you could do:

  $img->read(file=>'foo.raw', xsize => 100, ysize => 100,
             raw_datachannels => 4, raw_storechannels => 3,
             raw_interleave => 0)
    or die "Cannot read raw image\n";

In general, if you supply raw_storechannels you should also supply raw_datachannels

Read parameters:

  • raw_interleave - controls the ordering of samples within the image. Default: 1. Alternatively and historically spelled interleave. Possible values:

    • 0 - samples are pixel by pixel, so all samples for the first pixel, then all samples for the second pixel and so on. eg. for a four pixel scan line the channels would be laid out as:

    • 1 - samples are line by line, so channel 0 for the entire scan line is followed by channel 1 for the entire scan line and so on. eg. for a four pixel scan line the channels would be laid out as:


      This is the default.

    Unfortunately, historically, the default raw_interleave for read has been 1, while writing only supports the raw_interleave = 0 format.

    For future compatibility, you should always supply the raw_interleave (or interleave) parameter. As of 0.68, Imager will warn if you attempt to read a raw image without a raw_interleave parameter.

  • raw_storechannels - the number of channels to store in the image. Range: 1 to 4. Default: 3. Alternatively and historically spelled storechannels.

  • raw_datachannels - the number of channels to read from the file. Range: 1 or more. Default: 3. Alternatively and historically spelled datachannels.

  $img->read(file=>'foo.raw', xsize=100, ysize=>100, raw_interleave=>1)
    or die "Cannot read raw image\n";


PNG Image modes

PNG files can be read and written in the following modes:

  • bi-level - written as a 1-bit per sample gray scale image

  • paletted - Imager gray scale paletted images are written as RGB paletted images. PNG palettes can include alpha values for each entry and this is honored as an Imager four channel paletted image.

  • 8 and 16-bit per sample gray scale, optionally with an alpha channel.

  • 8 and 16-bit per sample RGB, optionally with an alpha channel.

Unlike GIF, there is no automatic conversion to a paletted image, since PNG supports direct color.

PNG Text tags

Text tags are retrieved from and written to PNG tEXT or zTXT chunks. The following standard tags from the PNG specification are directly supported:

  • i_comment - keyword of "Comment".

  • png_author - keyword "Author".

  • png_copyright - keyword "Copyright".

  • png_creation_time - keyword "Creation Time".

  • png_description - keyword "Description".

  • png_disclaimer - keyword "Disclaimer".

  • png_software - keyword "Software".

  • png_title - keyword "Title".

  • png_warning - keyword "Warning".

Each of these tags has a corresponding base-tag-name_compressed tag, eg. png_comment_compressed. When reading, if the PNG chunk is compressed this tag will be set to 1, but is otherwise unset. When writing, Imager will honor the compression tag if set and non-zero, otherwise the chunk text will be compressed if the value is longer than 1000 characters, as recommended by the libpng documentation.

PNG tEXT or zTXT chunks outside of those above are read into or written from Imager tags named like:

  • png_textN_key - the key for the text chunk. This can be 1 to 79 characters, may not contain any leading, trailing or consecutive spaces, and may contain only Latin-1 characters from 32-126, 161-255.

  • png_textN_text - the text for the text chunk. This may not contain any NUL characters.

  • png_textN_compressed - whether or not the text chunk is compressed. This behaves similarly to the base-tag-name_compressed tags described above.

Where N starts from 0. When writing both the ..._key and ..._text tags must be present or the write will fail. If the key or text do not satisfy the requirements above the write will fail.

Other PNG metadata tags

  • png_interlace, png_interlace_name - only set when reading, png_interlace is set to the type of interlacing used by the file, 0 for one, 1 for Adam7. png_interlace_name is set to a keyword describing the interlacing, either none or adam7.

  • png_srgb_intent - the sRGB rendering intent for the image. an integer from 0 to 3, per the PNG specification. If this chunk is found in the PNG file the gAMA and cHRM are ignored and the png_gamma and png_chroma_... tags are not set. Similarly when writing if png_srgb_intent is set the gAMA and cHRM chunks are not written.

  • png_gamma - the gamma of the image. This value is not currently used by Imager when processing the image, but this may change in the future.

  • png_chroma_white_x, png_chroma_white_y, png_chroma_red_x, png_chroma_red_y, png_chroma_green_x, png_chroma_green_y, png_chroma_blue_x, png_chroma_blue_y - the primary chromaticities of the image, defining the color model. This is currently not used by Imager when processing the image, but this may change in the future.

  • i_xres, i_yres, i_aspect_only - processed per Imager::ImageTypes/CommonTags.

  • png_bits - the number of bits per sample in the representation. Ignored when writing.

  • png_time - the creation time of the file formatted as year-month-dayThour:minute:second. This is stored as time data structure in the file, not a string. If you set png_time and it cannot be parsed as above, writing the PNG file will fail.

  • i_background - set from the sBKG when reading an image file.

You can control the level of zlib compression used when writing with the png_compression_level parameter. This can be an integer between 0 (uncompressed) and 9 (best compression).

If you're using libpng 1.6 or later, or an earlier release configured with PNG_BENIGN_ERRORS_SUPPORTED, you can choose to ignore file format errors the authors of libpng consider benign, this includes at least CRC errors and palette index overflows. Do this by supplying a true value for the png_ignore_benign_errors parameter to the read() method:

  $im->read(file => "foo.png", png_ignore_benign_errors => 1)
    or die $im->errstr;

ICO (Microsoft Windows Icon) and CUR (Microsoft Windows Cursor)

Icon and Cursor files are very similar, the only differences being a number in the header and the storage of the cursor hot spot. I've treated them separately so that you're not messing with tags to distinguish between them.

The following tags are set when reading an icon image and are used when writing it:


This is the AND mask of the icon. When used as an icon in Windows 1 bits in the mask correspond to pixels that are modified by the source image rather than simply replaced by the source image.

Rather than requiring a binary bitmap this is accepted in a specific format:

  • first line consisting of the 0 placeholder, the 1 placeholder and a newline.

  • following lines which contain 0 and 1 placeholders for each scan line of the image, starting from the top of the image.

When reading an image, '.' is used as the 0 placeholder and '*' as the 1 placeholder. An example:


The following tags are set when reading an icon:


The number of bits per pixel used to store the image.

For cursor files the following tags are set and read when reading and writing:


This is the same as the ico_mask above.


The "hot" spot of the cursor image. This is the spot on the cursor that you click with. If you set these to out of range values they are clipped to the size of the image when written to the file.

The following parameters can be supplied to read() or read_multi() to control reading of ICO/CUR files:

  • ico_masked - if true, the default, then the icon/cursors mask is applied as an alpha channel to the image, unless that image already has an alpha channel. This may result in a paletted image being returned as a direct color image. Default: 1

      # retrieve the image as stored, without using the mask as an alpha
      # channel
      $img->read(file => 'foo.ico', ico_masked => 0)
        or die $img->errstr;

    This was introduced in Imager 0.60. Previously reading ICO images acted as if ico_masked => 0.

  • ico_alpha_masked - if true, then the icon/cursor mask is applied as an alpha channel to images that already have an alpha mask. Note that this will only make pixels transparent, not opaque. Default: 0.

    Note: If you get different results between ico_alpha_masked being set to 0 and 1, your mask may break when used with the Win32 API.

cur_bits is set when reading a cursor.


  my $img = Imager->new(xsize => 32, ysize => 32, channels => 4);
  $im->box(color => 'FF0000');
  $im->write(file => 'box.ico');

  $im->settag(name => 'cur_hotspotx', value => 16);
  $im->settag(name => 'cur_hotspoty', value => 16);
  $im->write(file => 'box.cur');


SGI images, often called by the extensions, RGB or BW, can be stored either uncompressed or compressed using an RLE compression.

By default, when saving to an extension of rgb, bw, sgi, rgba the file will be saved in SGI format. The file extension is otherwise ignored, so saving a 3-channel image to a .bw file will result in a 3-channel image on disk.

The following tags are set when reading a SGI image:

  • i_comment - the IMAGENAME field from the image. Also written to the file when writing.

  • sgi_pixmin, sgi_pixmax - the PIXMIN and PIXMAX fields from the image. On reading image data is expanded from this range to the full range of samples in the image.

  • sgi_bpc - the number of bytes per sample for the image. Ignored when writing.

  • sgi_rle - whether or not the image is compressed. If this is non-zero when writing the image will be compressed.


To support a new format for reading, call the register_reader() class method:


Registers single or multiple image read functions.


  • type - the identifier of the file format, if Imager's i_test_format_probe() can identify the format then this value should match i_test_format_probe()'s result.

    This parameter is required.

  • single - a code ref to read a single image from a file. This is supplied:

    • the object that read() was called on,

    • an Imager::IO object that should be used to read the file, and

    • all the parameters supplied to the read() method.

    The single parameter is required.

  • multiple - a code ref which is called to read multiple images from a file. This is supplied:

    • an Imager::IO object that should be used to read the file, and

    • all the parameters supplied to the read_multi() method.


  # from Imager::File::ICO
     single => 
     sub { 
       my ($im, $io, %hsh) = @_;
       $im->{IMG} = i_readico_single($io, $hsh{page} || 0);

       unless ($im->{IMG}) {
       return $im;
     multiple =>
     sub {
       my ($io, %hsh) = @_;
       my @imgs = i_readico_multi($io);
       unless (@imgs) {
       return map { 
         bless { IMG => $_, DEBUG => $Imager::DEBUG, ERRSTR => undef }, 'Imager'
       } @imgs;

Registers single or multiple image write functions.


  • type - the identifier of the file format. This is typically the extension in lowercase.

    This parameter is required.

  • single - a code ref to write a single image to a file. This is supplied:

    • the object that write() was called on,

    • an Imager::IO object that should be used to write the file, and

    • all the parameters supplied to the write() method.

    The single parameter is required.

  • multiple - a code ref which is called to write multiple images to a file. This is supplied:

    • the class name write_multi() was called on, this is typically Imager.

    • an Imager::IO object that should be used to write the file, and

    • all the parameters supplied to the read_multi() method.

add_type_extensions($type, $ext, ...)

This class method can be used to add extensions to the map used by def_guess_type when working out the file type a filename extension.

  Imager->add_type_extension(mytype => "mytype", "mytypish");
  $im->write(file => "foo.mytypish") # use the mytype handler

If you name the reader module Imager::File::your-format-name where your-format-name is a fully upper case version of the type value you would pass to read(), read_multi(), write() or write_multi() then Imager will attempt to load that module if it has no other way to read or write that format.

For example, if you create a module Imager::File::GIF and the user has built Imager without it's normal GIF support then an attempt to read a GIF image will attempt to load Imager::File::GIF.

If your module can only handle reading then you can name your module Imager::File::your-format-nameReader and Imager will attempt to autoload it.

If your module can only handle writing then you can name your module Imager::File::your-format-nameWriter and Imager will attempt to autoload it.



Imager-preload> preloads the file support modules included with or that have been included with Imager in the past.

It isn't typically needed, but some cases where you might want to use it:

  • For use in forking servers such as mod_perl to allow as many pages as possible to be shared between parent and child processes.

  • to avoid runtime loading of a module from delaying output in an animation.

  • if you're loading modules from a relative $ENV{PERL5LIB} and expect to change directories.

You probably don't need it.

If the module is not available no error occurs.

Preserves $@.

  use Imager;


Producing an image from a CGI script

Once you have an image the basic mechanism is:

  1. set STDOUT to autoflush

  2. output a content-type header, and optionally a content-length header

  3. put STDOUT into binmode

  4. call write() with the fd or fh parameter. You will need to provide the type parameter since Imager can't use the extension to guess the file format you want.

  # write an image from a CGI script
  # using
  use CGI qw(:standard);
  $| = 1;
  binmode STDOUT;
  print header(-type=>'image/gif');
  $img->write(type=>'gif', fd=>fileno(STDOUT))
    or die $img->errstr;

If you want to send a content length you can send the output to a scalar to get the length:

  my $data;
  $img->write(type=>'gif', data=>\$data)
    or die $img->errstr;
  binmode STDOUT;
  print header(-type=>'image/gif', -content_length=>length($data));
  print $data;

Writing an animated GIF

The basic idea is simple, just use write_multi():

  my @imgs = ...;
  Imager->write_multi({ file=>$filename, type=>'gif' }, @imgs);

If your images are RGB images the default quantization mechanism will produce a very good result, but can take a long time to execute. You could either use the standard web color map:

  Imager->write_multi({ file=>$filename, 
                        make_colors=>'webmap' },

or use a median cut algorithm to built a fairly optimal color map:

  Imager->write_multi({ file=>$filename,
                        make_colors=>'mediancut' },

By default all of the images will use the same global color map, which will produce a smaller image. If your images have significant color differences, you may want to generate a new palette for each image:

  Imager->write_multi({ file=>$filename,
                        gif_local_map => 1 },

which will set the gif_local_map tag in each image to 1. Alternatively, if you know only some images have different colors, you can set the tag just for those images:

  $imgs[2]->settag(name=>'gif_local_map', value=>1);
  $imgs[4]->settag(name=>'gif_local_map', value=>1);

and call write_multi() without a gif_local_map parameter, or supply an arrayref of values for the tag:

  Imager->write_multi({ file=>$filename,
                        gif_local_map => [ 0, 0, 1, 0, 1 ] },

Other useful parameters include gif_delay to control the delay between frames and transp to control transparency.

Reading tags after reading an image

This is pretty simple:

  # print the author of a TIFF, if any
  my $img = Imager->new;
  $img->read(file=>$filename, type='tiff') or die $img->errstr;
  my $author = $img->tags(name=>'tiff_author');
  if (defined $author) {
    print "Author: $author\n";


When saving GIF images the program does NOT try to shave off extra colors if it is possible. If you specify 128 colors and there are only 2 colors used - it will have a 128 color table anyway.




Tony Cook <>, Arnar M. Hrafnkelsson