Roadmap¶
What comes next, ordered by how far along each one is. Most of it is something the documentation already says is coming. None of it has a date, the order can change at any time, and what gets built next depends on what people ask for.
Pending Release is finished and on its way out: a FaceCue update waiting on the Asset Store's approval, or a separate tool waiting to be published. In Progress is being built now. Researching means the shape is still being worked out. Queued is decided and next in line. Planned is decided and not yet scheduled, and there is no fixed order among the planned items: what gets built next depends on what people are asking for. Long Term is wanted, and not soon. Shipped, at the foot of the page, is out: it stays there for a release or two, and the release notes keep the full record.
If one of these matters to your project, say so on the Discord or by email at support@facecue.net, and it counts. Plans change, and something listed here can move, merge with another item, or, rarely, be dropped when it turns out not to work. Nothing here is a promise of a date.
Last updated 10 October 2026.
Spreading Across Cores Pending Release
This came in two halves. FaceCue now skips work it can prove would change nothing, which cuts the cost itself rather than moving it. And every character is calculated on worker threads, in parallel with the main thread, and written to the rig just before the engine skins the meshes, which is how a character runs unless you ask for something else.
Together, on a desktop machine, a crowd of eight high-fidelity characters costs about 50% less than it did, and an active character about 20% less. Of the cost that remains, 40 to 55% runs on worker threads instead of the main thread. If your own code is already busy late in the frame, more of the calculation finishes inside that work, and we measured 73%.
Standalone VR gains the most. On a Quest 3, 55 to 75% of what FaceCue costs the main thread moves off it. A character in a scene holding 90 fps moves 64%, and with SpaceWarp on, 69%.
These are all CPU figures. If the GPU is what holds your frame rate down, none of this will raise it. What it gives you is CPU time to spend elsewhere, on animation, physics, AI or the rest of your own code.
Figures from Windows and Quest hardware, in built players. Your rig, your platform and your scene move them, some by a lot, so measure your own project before budgeting against them.
Your own scripts carry on working the way they did. See it: Performance, and Script Coexistence for a script that reads or moves what FaceCue does.
Spanish Dialects Pending Release
Spanish is spoken differently across the regions that speak it, and some of those differences reach the mouth. A speaker from Spain puts the tongue between the teeth for the c in "cielo", and in Buenos Aires the ll of "calle" pushes the lips forward in a way it does not in Mexico City.
Spanish now bakes to a dialect you choose beside the language: Spain, Latin America, River Plate or Caribbean. Latin America is the one a Spanish line takes unless you pick another, so pick River Plate to keep the ll and y of Buenos Aires. A transcript tag can name a dialect too, which lets a line change dialect partway through without leaving its language. A line you have already baked keeps its result.
See it: Language Support.
Looking Away on Purpose Pending Release
The most visible thing a thinking person does is look away, and a character now does it too. A moment or a phrase marked as thinking in its lines makes it glance away from whoever it is talking to and come back, with the head joining in a little on a bigger look, and your own code can ask for the same look. Lines tagged before this arrived get it with no change, and a single moment can now be marked on its own with <Think/>.
This is the first of the deliberate looks away, and a Timeline track comes with it. A Thinking clip puts the moment exactly where a cutscene wants it, and sets the look's direction, its size and how much the head joins in, or leaves each of those to the character.
See it: Mental State, Asking for a Look-Away, and Cognitive Events.
Laughs, Hums and Other Non-Speech Sounds In Progress
A voice makes plenty of sounds that are not words, and a mouth that treats them as words gets them wrong. A recognizer hears a laugh as little more than silence, and reads a hum as the letters it happens to be spelled with.
Laughs are finished and go out with the next update. A laugh marked in a transcript plays on the mouth as laughter, taking its rhythm from the recording so that a slow rolling laugh and a quick giggle each move the way they sound, keeping an opening of its own so a quiet laugh still reads on the face, and carrying an emotion so the rest of the face laughs along. Hums are being built now: a hum reads as one held sound across the stretch it takes, with the lips closed where the hum is closed.
The other sounds a voice makes follow the same path, and which of them matters to your project is worth telling us.
See it: Laugh Tags and Laughs.
Nodding to What Is Said Researching
A listening character nods on its own today, on a loose schedule of its own. A real listener does not nod at random, they nod in answer to what is being said, and that is what makes a nod read as a response rather than a habit. This brings that, so a nod lands where a person would put one.
What a nod looks like is already set on the character, so the depth and the runs you tune today carry forward and get the timing when it lands.
Until then: Listening Gestures.
Dialogue System Integration Researching
The World Inputs ship with example sources that show the whole shape of a source with no game attached. The first integration built on them is for dialogue systems: a conversation feeds who is speaking, who is addressed and who owes an answer, and the character looks at whoever holds the floor, without you writing the glue or steering the look line by line.
Until then: World Inputs and Look Target.
Breathing on the Face Queued
Breath moves the body today, and the face stays out of it. This carries it into the nostrils and the mouth: the nostrils working as a character starts to exert itself, the breath moving to the mouth when it is breathing hard, and the quick breath a speaker catches through the mouth between phrases. How much of it shows follows how hard the character is working, so it is barely there at rest and plain to see after a sprint. It is authored on your rig, the way the rest of the face is.
It also brings the breathing in line with the recording. The baker already marks where a speaker drew breath, how big each break in the line is, and which breaths could be heard, and breath does not read those marks yet. With this, a character breathes where its voice actor did, deeper before a long run of speech and fuller where the breath was heard.
Until then: What It Moves and While Speaking.
Baking From a Transcript Alone Planned
Today the baker needs a recording to hear. A game with no voice track, a silent interactive novel say, has to synthesize the line, bake it, and discard the audio. This makes the baker read the transcript on its own and write the cues from that, so the round trip goes away.
Until then: Lip Sync Without Audio.
Bone Lip Sync Support Planned
FaceCue moves the mouth with blendshapes alone, and a rig whose jaw is a bone has to bake it into a blendshape on the way out of its tool. Most tools can. For the ones that cannot, this drives the jaw bone for speech directly, so a bone-only mouth talks without a bake.
Until then: What a Rig Needs.
Your Own Emotions Planned
The emotion library ships 28 named emotions. Today you shape how each of their expressions looks on your rig in the Face Rig Designer, while which expressions make up an emotion, and how, ships fixed. A fuller emotion designer opens all of it up: recomposing an emotion from its expressions, adding a 29th of your own, or renaming one, so a name you tag in a transcript can be one you invented.
Until then: The Emotion Library.
Your Own Delivery Styles Planned
The Speech driver ships five delivery styles, and the levers they are made of are all exposed. This lets you save a setting of those levers as a style of your own, name it, and pick it the way you pick the shipped five.
Until then: Speaker Delivery.
Baking on macOS and Linux Long Term
Baking runs in the Editor on Windows today, and what it bakes plays everywhere Unity runs. Bringing the baker to the other desktop platforms starts with CPU baking, since the GPU backends do not carry across evenly: DirectML is a Windows technology, and CUDA exists on Linux but not on modern macOS.
Until then: Requirements & Compatibility.
Other Asset Integrations Long Term
Beyond dialogue systems, FaceCue could meet other assets where they already are: a character system that knows who is talking to whom, an animation tool that owns the head, a dialogue editor that could write transcript tags. Which ones come first is the part we are asking about. Tell us on the Discord or by email what you would want FaceCue to integrate with, and what for, and the answers set the order here.
Until then: the World Inputs are the integration surface, and the example sources show how to wire a system of your own to them.
Shipped¶
FBX Exporter for MetaHumans Shipped
A MetaHuman's expressions live in its rig rather than as blendshapes on the mesh, and Unreal has no export that writes them out for Unity. This Unreal plugin assembles the character and bakes its expressions to blendshapes, as the standard ARKit set or as the full control set, in the form the wizard's two MetaHuman families read. It is its own product, on Gumroad.
If you own FaceCue, the exporter is free while it is in beta: mail your Unity Asset Store invoice number to support@facecue.net and we will send you a code. What comes with FaceCue once the exporter reaches 1.0.0 may change, and its licence says what the code covers.
Get it: FBX Exporter for MetaHumans, and its documentation.