Timg – Terminal Image Viewer

by setra • 9 years ago

124 43 9 years ago

_pgmf• 9 years ago

The terminal web browser w3m can display actual images pixel for pixel in the terminal. Ranger, a terminal file browser, leverages this to do image previews in the term. It's quite cool!

More details, including other methods, are described here:

https://github.com/ranger/ranger/wiki/image-previews

a3n• 9 years ago

Hey, that even works in the console (ctl-alt-F[1-6]). I guess I shouldn't be surprised, but I am. Delightfully.

digi_owl8• 9 years ago

As long as you use framebuffer consoles.

digi_owl8• 9 years ago

Best i understand you need to use a xterm derived terminal for this to work, because the sub-program used by w3m for this draws over the terminal window to make things happen.

One more thing we can kiss bye bye with Wayland i guess.

mrkgnao• 9 years ago

It's curious how we're prepared to replace X (with people writing tiling WMs and the like that support it) but somehow the move away from xtermy things (and backwards-compatibility-all-the-way-back-to-way-back-when) hasn't really caught on much, at least on the Linux side of things yet. Any exciting work in this area? I'm aware of Black Screen and some other work on the Electron/rewrite-everything-in-JS front (Hyper, I think, is another).

I've been using Alacritty myself for a couple of weeks now. It's not too bad: the htop header bar won't render properly and catting random binary files crashes the terminal pretty quickly, but it's smooth and does everything a good "tmux VM" should do well.

I understand iterm2 on macOS is a significantly improved experience?

rnhmjoj• 9 years ago

The image preview of ranger has different backends: besides xterm it works on iTerm2 and urxvt using their custom methods. Not sure about Wayland though.

jdormit• 9 years ago

Does anyone know how to make this work with tmux splits? I tried setting this up but the images would be displayed across splits as if w3m was taking up the whole terminal.

knolax• 9 years ago

For anyone interested, I wrote a standalone bash script wrapper for it a while back https://github.com/knolax/dotfiles/blob/master/scripts/image....

FreeFull2• 9 years ago

I have written something similar before here: https://github.com/FreeFull/termimage . My program though is just a simple toy, since it doesn't even consider the size on the terminal, and doesn't play animated gifs. On the other hand, the source code for my image viewer almost completely fits on my screen (if you ignore its use of libraries for terminal handling and image reading).

saagarjha2• 9 years ago

Those using iTerm should try out imgcat: https://www.iterm2.com/documentation-images.html

stefco_• 9 years ago

Agreed. There is even a matplotlib backend called itermplot[0] that uses iterm's image display capabilities, which, combined with iterm's tmux integration, makes developing scientific code on remote servers a dream.

[0] https://github.com/daleroberts/itermplot

saagarjha2• 9 years ago

Looks great! Now if only there was one for Mathematica…

daleroberts5• 9 years ago

Glad you like it :)

daleroberts5• 9 years ago

If you are using iTerm2 and doing computer vision / machine learning on images or working with large images, my tool called 'bv' has even more features and supports TMUX as well:

https://github.com/daleroberts/bv

It uses GDAL as a backend.

captn3m0• 9 years ago

2 npm modules I've relied on in the past for this: https://github.com/substack/picture-tube and https://github.com/switer/imaging

michaelmior• 9 years ago

pxl is another nice option

https://github.com/ichinaski/pxl

JdeBP7• 9 years ago

It makes an interesting comparison to the others mentioned here. timg and pixterm both hard-wire the ISO/IEC 8613-6 control sequences for 256x256x256 RGB colour. Neither of them, as I have already mentioned elsewhere, optimize their output at all.

pxl employs a far more complete third party library for its terminal rendering, named termbox. This has ncurses-style double-buffering for no-op redraw suppression and redundant colour control sequence optimization. It does not optimize as far as BSD SCREEN did in the 1980s, which made use of several more tricks to reduce the need for emitting lengthy cursor motion control sequences (such as carriage return, line feed, and the various relative-motion control sequences; and not optimizing away writing a character if it is shorter to just re-output the next cell position than to move the cursor).

The downside of termbox is that whilst it supports 8-colour and 256-colour terminal types, and reads the TERM environment variable and terminfo entry to determine colour capability, it does not support 256x256x256 RGB colour and hardwires the 8-colour and 256-colour control sequences to ECMA-48 and ISO 8613-6 indexed colour control sequences rather than employing the appropriate terminfo fields, even though it reads the terminfo database.

timg, pxl, and pixterm also only display 2 pixels per character cell, because they all use the same single Unicode character for rendering (making termbox, with all of its worrying about character widths and replacement characters, overkill to an extent). As mentioned by its author on this page, tv uses a whole bunch of the box drawing and block graphic Unicode characters for increasing the number of pixels per character cell.

But whilst tv does both 256-colour and 256x256x256 RGB colour, it does not do 8-colour or greyscale like pxl does, and it completely ignores the terminfo database and the TERM environment variable.

xuhu• 9 years ago

Does this or any of the mentioned alternatives work over ssh ?

tedunangst• 9 years ago

ssh doesn't really care about your terminals capabilities. (Yes.)

JdeBP7• 9 years ago

There is no real reason that any of them would not, SSH not involving decoding and then reëncoding terminal control sequences. The more relevant concerns would be whether they worked with tmux, screen, and mosh, which do involve that; and whether the terminal emulator on the local end of one's SSH connection supports 256x256x256 RGB colour control sequences.

That said, one of the issues on timg's issues list deals in the sheer amount of superfluous data that are generated and that that thus have to be transmitted across an SSH connection or a serial device. There is no optimization of the colour change control sequences at all. pixterm similarly does zero optimization, but no-one has pointed that out yet.

* https://github.com/hzeller/timg/issues/13

pxl, mentioned elsewhere on this page, does optimize colour control sequences and indeed uses an ncurses-style double-buffer mechanism to optimize re-drawing.

executesorder66• 9 years ago

This is great.

Since we are all posting alternatives, here's another:

https://github.com/withoutboats/notty

digi_owl8• 9 years ago

https://github.com/saitoha/libsixel

FreeFull2• 9 years ago

Note that most commonly used terminals don't support sixel graphics (The readme for libsixel has a list of some terminals that do have support), while a lot more terminals support 24-bit colours (all terminals that use a new enough libvte in particular, and some others).

digi_owl8• 9 years ago

Same "problem" as the w3m solution mentioned in the top most comment right now...

digi_owl8• 9 years ago

Just reminded myself of another option:

https://github.com/cacalabs/libcaca

JdeBP7• 9 years ago

The downside of libcaca is that it makes several presumptions that aren't necessarily true outwith the world of IBM PC Colour Graphics Adapter compatibility.

For example, it has a tendency to abuse boldface black on black as a way to get a dark grey colour, making it not work so well on terminals where boldface really is a switch to a boldface font rather than brightening the foreground colour. On such terminals, black on black text is invisible, boldfaced or no.

Sadly, it actually works internally in terms of a 4x16x16x16 ARGB colour system, and could work quite well with 88-colour or 256-colour terminals, and very well with terminals that have 256x256x256 RGB, were it not for the fact that most of the information in the internal colour representation is thrown away by the ncurses and CGA-like layers that are built on top of it.

iaml• 9 years ago

What's the difference with tv[0]?

[0] https://github.com/daleroberts/tv

daleroberts5• 9 years ago

Tv pushes the rendering further by using more unicode characters. It also uses GDAL as a backend which allows subsampling of massive images and access to many scientific image formats (GeoTiff, NetCDF, JPEG2000, etc)

JdeBP7• 9 years ago

Some other differences are given at https://news.ycombinator.com/item?id=13906483 .

One problem that all of these programs share is something that you cannot really do much about absent user-supplied information, since it is something that isn't exposed in the abstraction of a POSIX General Terminal Device. It is probably more acute for tv than for the others, albeit that as I have mentioned I couldn't get tv to actually run without error to see what the effect is.

This problem is that there is a baked-in assumption about the glyphs of the characters. This is not the usual concern that glyphs are different shapes in different fonts. That's not really much of a concern for box drawing and block graphic characters.

The baked-in assumption is that glyphs are rectangular, with a 1:2 aspect ratio. This is the assumption that underpins the idea that U+2580 is two square pixels, one atop the other.

But this is not true even for 9x15 fonts, which yield a slight distortion with all of these programs, let alone for yet squarer fonts. Barack Obama looks very odd indeed when U+2580 is actually a 16x16 or 32x32 square, as it sometimes is. (http://fileformat.info/info/unicode/char/2580/fontsupport.ht... shows several square appearances, and it is square in one of the fonts that I use here.)

tv, with the way that it already works, is probably best placed to incorporate user-supplied information about the font's character cell aspect ratio. pxl could probably be extended fairly simply, too. The others including timg, I suspect would be harder.

Interestingly, M. Zeller has had an open issue about this for timg since 2016.

* https://github.com/hzeller/timg/issues/4

daleroberts5• 9 years ago

TV gives even more interesting results as it uses more unicode 9.0 characters:

http://www.github.com/daleroberts/tv

JdeBP7• 9 years ago

But unfortunately for Debian 8 users, it is not compatible with the python3-gdal provided in Debian 8 and fails to work with a runtime error.

    Traceback (most recent call last):
      File "./tv", line 82, in <module>
        SAMPLING = {'nearest': gdal.GRIORA_NearestNeighbour,
    AttributeError: 'module' object has no attribute 'GRIORA_NearestNeighbour'

* https://packages.debian.org/jessie/python3-gdal

timg built and ran as-is on Debian 8 with the libraries that it needs.

daleroberts5• 9 years ago

Thanks for telling me, it looks like Debian is using GDAL v1...

msollami• 9 years ago

gpicview anyone?

taeric• 9 years ago

For emacs users, you can get a similar thing with iimage-mode and eshell. Kind of cool to have "$ cat SomeImage.png" do what I would expect and put it in the terminal.

Klasiaster• 9 years ago

Some time ago a Rust version of these kind of programmes was presented here:

https://github.com/nabijaczleweli/termimage

A useful additional script: #!/bin/bash for i in `find $@` ; do termimage --size 150x33 $i ; done

LeicaLatte• 9 years ago

Currently using pixterm which works on any ANSI terminal. https://github.com/eliukblau/pixterm

JdeBP7• 9 years ago

No, it does not. Like timg here, it only works on terminals that have 256x256x256 RGB colour, for which it hardwires the control sequences; and this is explained up-front in its doco.

An "ANSI" terminal need only support ECMA-48, which has 8 colours; as can be seen from the terminfo entry for the "ansi" terminal type. "true colour" is ISO/IEC 8613-6.

* http://unix.stackexchange.com/a/289871/5132

frabbit• 9 years ago

FYI: "catimg" is available as a Fedora rpm.

Timg – Terminal Image Viewer

Related Stories