Kovati — Automation Forge for Unreal Engine
A commercial plugin suite for Unreal Engine 5.8 — 34 plugins in 8 sets — that generates animation, voice, facial performance, meshes and materials inside the editor and files each result as an ordinary Unreal asset. Free to run on your own hardware, with a paid tier ("Kovati Pro") on a waitlist. The vendor page is the source; every number below is its own.
The argument it opens with
"The game you cannot playtest is the game you never make." The pitch is not that generated assets are shippable. It is explicitly the opposite:
"The bet is not that generated assets are good enough to ship. The bet is that a game which has been played for six months before its first real asset is authored is a better game."
The named problem is the playable gap — the distance between what a design asks for ("inject the stimpack"), what a model returns (a four-second clip), and what the game needs (target skeleton, prop alignment, montage slot, gameplay notify, timing, blend, ability binding). The site draws that third column widest and says it is the part no generator returns. Closing it is the product; the surrounding framework is called PlayableOps, with a four-step loop: make it playable, keep it traceable, graduate it deliberately, replace it safely.
That is a sharper position than the AI-tooling claims this spoke usually records. It concedes the quality argument entirely and relocates the value to iteration speed before authoring, with replacement designed in from the start.
The suite
| Set | Plugins | What it makes |
|---|---|---|
| AutomationForge | 4 | the umbrella: pipelines across the others, costed, resumable, with a ledger |
| MotionForge | 7 | animation from a description, imported as an AnimSequence |
| SpeechForge | 4 | dialogue with character-level timing, cached and priced per line |
| FaceForge | 5 | audio → performing face, locally on CPU or via NVIDIA Audio2Face-3D, retargeted |
| MeshForge | 6 | prop or image → game-ready static mesh with collision, lightmap UVs, materials |
| MontageForge | 2 | clip → montage with notifies that survive a rebuild |
| PerformanceForge | 2 | webcam and microphone capture on one take clock, driving the same pipelines |
| SurfaceForge | 4 | prompt or reference → reviewable material candidates → PBR maps |
| ArtForge | — | announced, not shipped: one versioned style source everything generates against |
Vendor performance claims, unverified: ~14 s from prompt to a finished animation clip on your own GPU; $0.35 to voice a 95-line scene; face solve at 1:2 of audio length on CPU; zero steps re-executed on an unchanged project.
Why it belongs on engine-agent-surface
The site states an architectural position it calls Trinity Architecture: every operation lives on a typed subsystem, and both the editor UI and the agent toolset are thin clients over the same surface, neither holding logic, so "neither can drift." Its stated test — "if a function needs a widget to work, the design is wrong" — is the sharpest formulation of the machine-user constraint the corpus holds.
It also claims a consequence worth recording as a claim: designing for the machine user produced a better API for the human one — idempotent, non-blocking, typed, and costed before it commits. The same page shows an agent driving the whole suite to build a quest end to end, and a human doing the identical work through the editor, as the demonstration that both clients are real.
This makes it the corpus's fifth agent-surface instance and the first from a third party rather than an engine vendor or a community MCP plugin. It agrees with immens that the surface should be architecture rather than a bolt-on, and unlike Immens it ships.
Where the caution goes
T3. Vendor marketing for a commercial product: the throughput and cost figures are first-party
with no methodology, no hardware named for the $0.35 scene, and no independent test. Several media
slots on the page still read "placeholder", so the site is mid-build. The GitHub organization
(AutomationForgeHQ) is linked and the tagline is "open where it counts," which does not say what
is open — the engine tooling is described as ordinary production code with no generation in it, and
that is as specific as the page gets. Contact is one person's email address.
No publication date anywhere, so published: is left off.
Connections
unreal-engine · unreal-engine-5-8 · engine-agent-surface · immens · procedural-generation · synthesis
- cross-spoke: the agent side (an agent as a first-class API consumer, typed tool surfaces) is
../agentic-tooling-wiki's; the voice generation is../speech-audio-wiki's market