What shipped, and what changed.

MeshForge and MotionForge were rebuilt and released with every plugin around them — twenty-one plugin releases since the last entry, and a new hub. The suite is thirty-seven plugins in nine sets.

MeshForge now starts from a workflow: a prompt, your own picture, or a mesh you already have. After generation every step is yours to steer — take the mesh to Blender and back, start any step from the exact take or run you liked, reorder the chain without losing a result. MeshForge Garment, the paid layer, adds sculpting inside Unreal and the Garment Studio, where the character is posed and the garment placed, sculpted and wrapped in one window. A new open toolset lets an agent that can see do that fitting by sight.

MotionForge opens on a guide from nothing to a first motion, provider by provider, with what each one costs. Where fetching a take is free, it plays on the character it is for before you choose it — beside another if you like, exactly as it will import. Generate says what it will cost and asks before spending, and Kimodo and Uthana each bring their own settings under their own names. Kimodo can run on another machine in the studio, and says plainly whether your access to the model it needs has been granted.

Localisation keeps the acting. A line that was performed is now dubbed rather than read out again, so the translation carries the performance, and its face follows the performance captured in the original language. One costed step generates what was synthesised and dubs what was performed.

The hub shows what is new before you update — the notes for every version since yours, skipped ones included, now that every release carries its own. Its menu inside Unreal is on in every project from the moment it is installed, and that is where a project turns on the plugins it uses. Paid plugins are delivered only to the accounts that own them. Payments are still simulated: nothing has been bought for money yet.

And there is somewhere to follow it. The Discord community is open at kovati.dev/discord, and new releases and new films are posted there as they happen. Three films went up — the end-to-end walkthrough, MeshForge from pictures to a placed prop, and installing the hub — each on the page it shows. Kovati is on X as @kovatidev.

Next: card payments, so the paid plugins can be bought for money, and the next hub film — service keys, and where each tool runs: on your machine, on a rented GPU, or through a provider’s API.

The suite is thirty-six plugins in nine sets, and everything in it is released. Two sets are new since the last entry: MeshForge, which turns a prompt or a reference image into a game-ready static mesh and can fit a generated garment onto a character; and SurfaceForge, which does the same shape of work for materials. PerformanceForge — record a performance in the editor and bake it onto a playable character — landed with them.

Three plugins now carry a price, the first that do: PerformanceForge, the fitted-garment pipeline, and the DeepL translation provider. The rule behind that split has not moved. If it runs on your own machine and costs us nothing to operate, it stays free, and the agent surface stays open even where the implementation is sold — every one of those three is reachable through a toolset anyone can read.

The site moved to kovati.dev with an account beside it, and the hub learned to sign in, claim a free plugin and install a bought one. Payments are still simulated: nothing has been bought for money yet, and the account app says so where it matters.

Next: the documentation, and the first Fab listings.

The generation family is public: five sets covering animation, voice, faces, montages, quality measurement and the pipeline layer that runs them — free on your own hardware, with the cores and every toolset open source and the rest packaged on Fab.

Roughly half of them are toolset plugins holding no logic at all: a separate adapter that exposes the capability beside it as Model Context Protocol tools. You can delete any of them and the capability behaves identically. That split is the whole architectural argument, and it is easier to see in a directory listing than in a paragraph.

Next: our own voice-over pipeline re-expressed as a user-editable definition with the hardcoded one deleted. That last clause is the milestone, because it is the only version of the claim that cannot be faked.