Custom emoticons and stickers have become commonplace in modern messaging apps. They allow users to express themselves in ways that words, nor the standard Unicode set cannot. And they can be functional, allowing bots to convey meaning with custom symbols in messages or reactions.
If Matrix is to remain both a fun and useful messaging protocol, as well as a flexible platform for bridging into other protocols, support for custom emoticons and stickers are a must.
This proposal outlines how custom emoticons and stickers can both be sent into rooms, as well as organised into "packs" and shared between users.
Since there is a lot of confusion of how this relates to m.emote
, why this isn't called "custom emoji"
etc,:
m.emote
is for emotion - and it has been incorrectly named this way. m.action
would have been more
appropriate, as you use it to describe actions, not emotions. E.g. "/me is walking to the gym", or
"/me is happy" and not "/me happy".
That, however, is not what this MSC is about. Instead, it is about emoticons, also known in short as emotes.
Emoticons are just little images or text describing emotions or other things. Emoji are a subset of emoticons, namely those found within unicode. Custom emoji here would actually refer to a custom emoji font, that is your own rendering of 🦊, 🐱, etc., not new images. New images is what custom emoticons are for.
Now, a client may choose to name these however they like. In the spec's history, it has had naming disparity with clients, i.e. "groups" vs "communities". It is, however, imperative to name things in the spec accurately after what they are.
Stickers already exist in Matrix. They are reusable images one can send, usually as a reaction to something sent in the timeline. This MSC adds a way to distribute and define a source for a client to send them.
Emoticons have at least a shortcode and a mxc uri. They are sent as <img>
tags, which are currently in
the spec. As such, many existing clients are able to render them.
To allow clients to distinguish emoticons from other inline images, a new
property, data-mx-emoticon
, is introduced. A client can choose to ignore the size attributes of emoticons
when rendering, and instead pick the size based on other circumstances. This could e.g. be used to
display messages with only emoticons and emoji as larger than usual, which is commonly found in
messengers. Such an <img>
tag of a shortcode emote
and a mxc uri mxc://example.org/emote
could look as follows:
<img data-mx-emoticon src="mxc://example.org/emote" alt=":emote:" title=":emote:" height="32" />
alt
and title
are required attributes for their roles in accessibility and
improved UX. Their meaning is inherited from the HTML specification
(alt,
title).
The height
SHOULD be set to "32px". This is a default that looks good on most
devices. In practice, receiving clients are expected to override the height when
rendering emotes based on their particular environment (the user's font size,
etc.). In order to maintain backwards-compatibility with clients that do not
support emotes, specifying the height
is required. No width is specified as to
maintain the aspect ratio of non-square emotes.
If the new data-mx-emoticon
attribute has a value, that value is ignored. Due to limitations of some libraries
the attribute may even look like data-mx-emoticon=""
.
The src
attribute must be a mxc URI. Other URI schemes, such as https
,
mailto
etc. are not allowed. Clients MUST NOT attempt to render images
accessible through other URI schemes.
To send stickers, the already spec'd
m.sticker
is
used.
Emoticons SHOULD be at least 128x128 pixels. Even though the fallback specifies a height of 32, this is to ensure that emotes still look good on higher DPI screens.
Stickers SHOULD be at least 512x512 pixels, for the same reason.
Images filetypes are not limited by this proposal. Instead they are
equivalent to the formats allowed in an
m.image
event. As
of writing, no limitations for m.image
are currently defined (see this spec
issue).
Emoticons and stickers may be animated.
The image pack event has a type of m.image_pack
. It contains a key images
, which is a map from a
short code to an image object. It may also contain a key pack
, which is a pack object, described in
the following section.
The pack object consists of the following keys:
display_name
: (String, optional) A display name for the pack. Defaults to the room name, if the image pack event is in the room. This does not have to be unique from other packs in a room.avatar_url
: (String, optional) The mxc uri of an avatar/icon to display for the pack. Defaults to the room avatar, if the pack is in the room. Otherwise, the pack does not have an avatar.usage
: (String[], optional) An array of the usages for this pack. Possible usages are"emoticon"
and"sticker"
. If the usage is absent or empty, a usage for all possible usage types is to be assumed.attribution
: (String, optional) The attribution of this pack.
Example:
{
"display_name": "Awesome Pack",
"usage": ["emoticon"]
}
The image object consists of the following keys:
url
: (String, required) The mxc URL for this image.body
: (String, optional) An optional text body for this image. Useful for the sticker body text or the emote alt text. Defaults to the shortcode.info
: (ImageInfo
, optional) The already spec'dImageInfo
object used for theinfo
block ofm.sticker
events.usage
: (String[], optional) An array of the usages for this image. The possible values match those of theusage
key of a pack object. If present and non-empty, this overrides the usage defined at pack level for this particular image. This is useful to e.g. have one pack contain mixed emotes and stickers. Additionally, as there is only a single account data level image pack, this is required to have a mixture of emotes and stickers available in account data.
Taking all of this into account, an example pack event may look like the following:
{
"images": {
"myemote": {
"url": "mxc://example.org/blah"
},
"mysticker": {
"url": "mxc://example.org/sticker",
"usage": ["sticker"]
}
},
"pack": {
"display_name": "Awesome Pack",
"usage": ["emoticon"]
}
}
There are several places where a client is expected to look for events with type m.image_pack
, mainly
in their own account data and in room state.
Each user can have their own personal image pack defined in their account data, under the
m.image_pack
key. The user is expected to be presented with these images in all rooms.
A room can have an unlimited amount of image packs, by specifying the m.image_pack
state event with
different state keys. By default, the user is expected to be presented with these images only in the room they
they are defined in. To enable them to be presented in all rooms, see the "Image pack rooms" section.
E.g. a discord bridge could set as state key
de.sorunome.mx-puppet-bridge.discord
and have all the bridged emotes in said state event, keeping
bridged emotes from matrix emotes separate.
Clients should suggest image packs of a room's canonical space, if the user is also in that space. This should be done recursively on canonical spaces. So, if a room has a canonical space and that space again has a canonical space, the clients should suggest image packs of both of those spaces.
While room image packs are specific to a room, they can be made accessible from anywhere by setting
the m.image_pack.rooms
key in a user's account data. The value is an object, with a room
key containing
a map of room ids to the state keys to an object. If a room id / state key combination is provided in this form,
clients should make the corresponding room image pack globally accessible in all rooms.
Note that while this MSC does not define any keys for the bottom-level object, defining it as an object means greater flexibility in the case of future changes.
The contents of m.image_pack.rooms
could look like the following:
{
"rooms": {
"!someroom:example.org": {
"": {},
"de.sorunome.mx-puppet-bridge.discord": {}
},
"!someotherroom:example.org": {
"": {}
}
}
}
Here three emote packs are globally accessible to the user: Two defined in !someroom:example.org
(one with blank state key and one with state key de.sorunome.mx-puppet-bridge.discord
) and one in
!someotherroom:example.org
(with a blank state key).
If a client gives image suggestions (emotes, stickers) to the user in some ordered fashion (e.g. an ordered list where you click an entry), the order of the images should be predictable between rooms. A suggestion for clients of image pack ordering is as follows:
- User image pack (defined in your own account data)
- Image pack rooms (defined in the
m.image_pack.rooms
account data object) - Space image packs (defined in the hierarchy of canonical spaces for the current room)
- Room image packs (defined in the currently open room's state)
Furthermore, clients are expected to de-duplicate images so that same images are not displayed multiple times. Such scenarios include:
- when viewing a room that has a pack defined in the
m.image_pack.rooms
account data object, and - a bot which syncs emotes over multiple rooms.
This could be done by deduplicating by the mxc URI.
For emoticons a client could add delimiters (e.g. :
) around the image shortcode, meaning
that if an image has a shortcode of emote
, the user can enter :emote:
to send it. If there are
multiple emoticons with the same shortcode in a room, the client could e.g. slugify the packs display
name and then have the user enter :slug~emote:
. As slugs typically match ^[\w-]+$
, that should
ensure complete-ability.
The alt / title text for the <img>
tag is expected to be the body
of the emote. If absent, its
shortcode should be used instead.
When sending a sticker, the body
of the m.sticker
event should be set to the body
defined for that
image, or if absent, its shortcode.
The info
object of the m.sticker
event should be set to the info
object of the image, or if absent,
an empty object.
When sending an <img>
tag in an encrypted room, the client will make a request to fetch
said image, in this case an emote. As there is no way to encrypt content behind <img>
tags yet,
this could potentially leak part of the conversation. This is not a new security consideration,
it already exists. This, however, isn't much different from posting someone a link in an e2ee chat and
the recipient opens the link. Additionally, images, and thus emotes, are often cached by the client,
not even necessarily leading to a http query.
Related issue: https://github.com/matrix-org/matrix-doc/issues/2418
The m.image_pack
in the account data is replaced with im.ponies.user_emotes
. The m.image_pack
in
the room state is replaced with im.ponies.room_emotes
. The m.image_pack.rooms
is replaced with
im.ponies.emote_rooms
.
Some existing implementations using im.ponies.user_emotes
and im.ponies.room_emotes
currently use
a dict called short
which is just a map of the shortcode to the mxc url.
Some other implementations currently also call the images
key emoticons
.
One can easily think of near infinite ways to implement emotes. The aspect of sending an <img>
tag
and marking it as an emote with data-mx-emoticon
seems to be pretty universal though, so the question
is mainly on different emote sources. For that, a separate MSC, MSC1951
already exists, so it is reasonable to compare this one with that one:
MSC1951 defines a dedicated room as the only image pack source. This MSC, however, also allows you to bind image packs to your own account, offering greater flexibility. In MSC1951 there can also only be a single image pack in a room. This could be problematic in e.g. bridged rooms: You set some emotes or stickers from the matrix side and a discord bridge would plop all the discord emotes and stickers in another pack in the same room.
The original sharing-focused idea of MSC1951 is still preserved: Once room types are a thing, you could still easily have an image pack-only room.
MSC1951 defines a way to recommend using a pack of a different room - this MSC does not have an equivalent to that. Instead, this MSC allows multiple image packs for a room, typically one where you already chat in anyways. Furthermore, it allows you to enable an image pack to be globally available for yourself across all rooms you are in.
The core difference is how MSC1951 and this MSC define the image packs themselves: In MSC1951 you have to set one state event per image. While this might seem like a nice idea on the surface, it doesn't scale well. There are people who easily use and want hundreds or even thousands of image packs accessible. A simple dict of shortcode to mxc URI seems more appropriate for this.
In general, MSC1951 feels like a heavier approach to image pack sources, while this MSC is more lightweight.
A couple of interesting points have been raised in discussions of this MSC that tangentially touch
custom emoticons. Each warrant an MSC for themselves however, as they touch more on how <img>
is works.
- Figuring out how
<img>
should work with encrypted media. - Allow SVGs in the
<img>
tag. Current problem: Clients typically try to thumbnail the mxc URL, and most media repositories can't thumbnail SVGs. Possible solution: Somehow embed the mimetype. - For stickers: Recommend rendering sizes / resolutions