For Unreal Engine 5.8

Your AI copilot,
inside the editor.

MIA wires local and cloud AI models directly into UE5 — generate meshes, textures, sprites, terrain, Gaussian Splats, animation, and logic, or drive a live Blender session, without ever leaving the editor. Prefer working outside the engine? MIA Studio is the same generation pipeline as a standalone desktop app — and it now carries a full digital sculpting suite of its own.

Fab creator account accepted — launching on Fab Marketplace soon

48 plugins 1,000+ AI-callable functions Local & cloud model support

Capabilities

Built for real production work

Every toolset runs natively inside the Unreal Editor — no context switching, no external round-trips.

Generative Toolsets

Produce meshes, textures, and landscapes from text or reference images — powered by ComfyUI, Tripo3D, Meshy, and more. Procedural terrain noise and real heightmap import bring landscapes into the same one-click pipeline.

Pixel Art Sprites

A dedicated Pixel Lab surface — text-to-sprite, style-referenced generation, multi-frame animation, and masked inpainting, with a crisp nearest-neighbor resize control for whatever comes back.

Gaussian Splats

Generate a splat from one photo, or train one locally from your own footage. Import from Postshot, Scaniverse, Polycam, and more.

Not found on any other Fab plugin

AI Video Generation

Text, a starting image, or reference frames into a real video clip — MiniMax H3 (Hailuo 3.0): up to 2K, 4–15 seconds, with native audio in a single pass. In the editor, in MIA Studio, and metered through MIA Cloud.

AI-Driven Animation

Generate baseline motion from a text prompt, import real BVH mocap, and train a neural locomotion controller — all in one tab.

Real-Time Fluid VFX

Niagara-driven fluid and particle effects, tunable straight from the panel — no manual system graph wiring required.

In-Editor Chat

Talk to Claude, OpenAI, Gemini, Grok, MiniMax, DeepSeek, local Ollama/LM Studio models, or your own OpenAI-compatible endpoint from a dockable chat panel with live streaming. OpenAI is a first-class provider row with a live-fetched model list, and both Claude Code CLI and OpenAI Codex CLI logins work as backends if you would rather use a subscription than a per-token key. Ask it to read a real webpage, PDF, or video transcript and it will — not just a search snippet.

AI Cursor Guidance

Turn it on and MIA can move your real mouse, click, type, and take a screenshot to show you exactly what to do — off by default, and every action still needs your approval.

Live Blender Integration

Let MIA's chat script a running Blender session directly — create, rig, and material objects in real time, not just import/export.

Not found on any other Fab plugin

Visual Workflow Graphs

Chain generation and automation nodes on a graph canvas, pick from 17 ready-to-run starter templates, or let an orchestrator agent build the workflow for you.

Native C++ Ports

Core toolsets ship as first-class native plugins — full API parity, no Python bridge, no runtime dependency overhead.

Local-First, Privacy Friendly

Run entirely offline with local models where it counts — cloud providers are opt-in, never required.

Unified Asset Library

Every generated texture, mesh, sprite, and sound in one searchable, filterable browser — no more hunting across toolsets for what you already made.

Survives a Rebuild

Some C++ changes need a real editor restart, which used to end the conversation. MIA can now save what she was doing, close the editor, rebuild, relaunch it, and pick the thread back up on the other side — every step still behind the same per-call approval dialog. Full history replay on the Claude CLI backend; other backends come back with a note about where they left off.

New

Build → Test → Fix, In One Loop

A native Agent Skill that chains Live Coding compiles, automation tests, a real Play-In-Editor session, and log reading into a single verification pass — so MIA can check her own work against the running editor instead of handing you code she has never compiled.

New

Generate It, Then Rig It

Take a Rodin- or Tripo-generated mesh straight through a headless Blender cleanup pass (apply scale, fix normals, optional decimate) and hand it off to Reallusion AccuRig for rigging, then import the rigged FBX back — all from the Animation tab. Gaussian Splats get an Open in SuperSplat button that launches the real hosted editor instead of telling you to go find it.

New

Bring a MIA Studio Material In

Build a shader on MIA Studio's node canvas, then land it in Unreal as a real Material — the graph compiles to self-contained HLSL, each output channel becomes a Custom expression, and MIA wires them onto Base Color, Metallic, Roughness, Normal, Emissive, and Opacity for you. Graphs that can't survive the translation are rejected with the specific reason, not silently half-imported.

New

Drop-In Fab Install

One install from the Fab Marketplace once launched. No manual builds, no missing dependencies — it just works.

In The Editor

See it running, not just described

Live captures from a real Unreal Engine 5 project running MIA — no mockups, no staged renders.

MIA Welcome tab showing its live capability count
Welcome tab MIA's own live capability count, read straight from the running Editor — 2,771 AI-callable functions across 127 toolsets and 43 plugins at the time of this capture. The family has since grown to 48 plugins.
MIA Cloud tab showing previously generated 3D models
Cloud tab — Previously Generated Models A real gallery of already-generated meshes (Tripo3D output) sitting in a real project.
MIA Splats tab for Gaussian Splat generation
Splats tab The Gaussian Splat prop-generation UI — GaussianSplattingForUnrealEngine and NanoGS, both installed and ready to go.
MIA Workflows tab showing a multi-node generation graph
Workflows tab — a real multi-node graph The "Cloud Mesh Post-Process Showcase" preset: Generate Cloud Mesh feeding Check Riggable / Segment / Rig / Retarget in parallel — a real 5-node pipeline.
MIA Kanban tab organizing generated assets into columns
Kanban tab Real cards promoted onto real columns (Images, Mesh Gen) from real generated assets — Name/Value attribute fields and Move/Browse/Delete actions.
MIA Graph tab comparing two generated assets side by side
Graph tab Two asset nodes wired into a Compare node's Left/Right inputs — a genuine, working connection drawn live in the Editor.
MIA Sound tab with Chatterbox Turbo voice cloning
Sound tab — Chatterbox Turbo voice cloning The "Clone Voice From Clip…" picker, live and reachable from the panel.
MIA Logic tab with chat, mic input, and a live status bar
Logic tab — chat, mic, and a live status bar Claude connected with a real API key, the Mic / Attach / Send row, and a live CPU / GPU / VRAM / token status bar.

Desktop Companion

MIA Studio — the same pipeline, outside the engine

A standalone desktop app for pre-production and asset generation before anything touches a project file — generate, iterate, and organize, then send finished assets straight into Unreal, Unity, Godot, or a live Blender session. It is now a digital sculpting application in its own right, not just a generator with a mesh viewer bolted on. MIA Studio Lite brings the same pipeline to an Android tablet and MIA Pocket to a phone — on-device chat, vision, and image generation that need no server at all, fully offline-capable with no desktop app required.

Full Generation Suite

Textures, PBR materials, meshes, Gaussian Splats, and pixel-art sprites — the same local ComfyUI and cloud providers as the Editor plugin, plus a fully local mesh backend (Pixal3D) that needs no cloud key at all, and a live check for whether a workflow’s required ComfyUI node pack is actually installed on your server. Side-by-side multi-model comparison, and a node graph with its own starter-template gallery including real multi-step chains.

Layered Image Editor

A Clip Studio / Krita-class 2D editor built in: raster and re-editable vector layers, groups, clipping masks, non-destructive filter/adjustment layers, layer styles (shadow / stroke / glow), paint-editable transparency masks, and 16 blend modes. Brush engines — round, textured, spray, smudge, blend, sketch, hatching, bristle, clone, and custom stamps that import .abr/.gbr/.gih/Krita tips — with a pressure curve, stabiliser, QuickShape, and mirror + radial symmetry. Painting assistants (ruler, parallel, vanishing-point, 2-point perspective, ellipse) that snap your strokes. Fill, gradient, shapes, Levels & Curves, a full filter gallery, magic-wand selections, perspective transform, a pen tool, and a frame-by-frame animation timeline with onion skin and GIF / webm / spritesheet export. Wide-gamut Display P3, linear-light compositing with soft-proofing, CMYK soft-proof, autosave, palette management (.gpl/.ase), and PSD / OpenRaster / TIFF / SVG / 16-bit PNG import & export.

Sound Lab

Text-to-speech with real voice cloning (F5-TTS, Qwen3-TTS, VibeVoice), sound effects, and music generation — local and free by default, with ElevenLabs available when you want cloud quality, transcription, or dubbing into another language.

Engine Bridges

Connect to Unreal, Unity, Godot, or Blender's own MCP server and push a finished asset straight into a running session — plus one-click BlenderMCP addon setup, no manual install.

Multi-Backend Chat

Claude, OpenAI, Gemini, Grok, MiniMax, DeepSeek, local Ollama/LM Studio, your own OpenAI-compatible endpoint, or a Claude Code / OpenAI Codex CLI subscription login instead of a per-token key — with MCP tool-calling into every capability the app has, real webpage/PDF/video-transcript reading, and voice in/out.

Persistent Memory & a Guided Browser

MIA remembers facts across sessions and brings them into every new chat automatically. Turn on Web Control and she can drive a real embedded browser tab — navigate, click, and read pages — right alongside your project.

AI Cursor Guidance

Turn it on in Logic or Learn and MIA can move your real mouse, click, type, and take a screenshot to show you exactly what to do — off by default, and every action still needs your approval.

Project Notes

Freeform notes per project, with one click to reorganize rough thoughts into an ordered checklist via AI — no detail dropped, just restructured.

A Real Sculpting Suite

Start from a preset primitive on the new 3D page and sculpt — thirteen brushes including Clay, Flatten, Inflate, Smooth, Polish, Move/Snake-Hook, Stamp, and an Insert brush that stamps instanced primitives onto the surface. Dynamic Topology adds resolution under the brush as you go, multiresolution levels let you rough in a shape and add detail without losing either, and a non-destructive layer stack plus paintable masking keep the whole thing reversible. Face Groups, Curve tools (tube and lathe), symmetry, voxel remesh, quad retopo, boolean ops, and a Bwrap-style scan-wrap retopology cage round it out. Matcap and HDRI preview lighting and an opt-in Beauty render show the form properly while you work.

New

Projected Texture Displacement

Wrap a model in real surface texture — pores, weave, knurling, scales — as actual geometry, so it survives export and 3D printing where a normal map is not an option. Brush it on a patch at a time, or run a whole-mesh pass; both land on the same surface and blend with no seam between them. Displacement is bounded rather than additive, so scrubbing over the same spot converges instead of melting. Any alpha in your brush library (.abr, .gbr, .gih, Krita tips, or one you drew) can drive it, with spacing, scatter, and an edge fade so a stamped line doesn’t read as machine-made.

New

Node Material & Brush Graphs

A real node-wire shader editor that compiles to GLSL and splices into the live material — Voronoi, Brick, Wave, Gradient, Color Ramp, Bump, and HSV procedural nodes included — plus triplanar projection that textures a model from all three axes with no UV unwrap. A second graph drives the brush itself: per-dab radius, strength, rotation, and spacing wired from pressure, speed, stroke time, and randomness. Finished materials export to Unreal as native HLSL.

New

Photo to PBR

Drop in a photograph and get a full PBR material back — base color, roughness, normal, and height — with an optional neural depth pass (Depth Anything V2) when the shading heuristic alone isn’t enough. The same endpoint is reachable from Lite and Pocket over the desktop relay, so a tablet gets the neural pass without running the model itself.

New

Pose, Rig, Groom, Drop-Test

A keyframe dopesheet with two-bone analytic IK poses a rigged character and exports one consolidated animation clip. Guide-based grooming lays hair and fur down as ribbon meshes that export to glTF. A self-contained rigid-body Drop Test answers “will this actually stand up” before it ever reaches an engine. For rigging proper, send a mesh through a headless Blender cleanup into Reallusion AccuRig and bring the rigged result back, or hand a Gaussian Splat to the hosted SuperSplat editor embedded right in the app.

New

Scene Builder

Describe a scene in a sentence and MIA breaks it into its individual objects, then generates every one of them in parallel through the mesh pipeline you already use — a populated set instead of one prop at a time.

New

Local-First, Portable

Every project is a real, Git-syncable folder on disk. Bring your own local models where it counts — cloud providers are opt-in, never required.

Android · Tablet

MIA Studio Lite — the full pipeline, on a drawing tablet

Everything that makes MIA Studio useful, running natively on an Android tablet — including the sculpting suite and the node graphs, not a cut-down version of them. On-device chat, vision, and image generation need no server and no Wi-Fi at all. Holding a phone instead? MIA Pocket is the same app, laid out for one thumb.

Coming soon to Google Play

On-Device AI, No Server

Chat, vision, and image generation run locally on the device itself — whisper.cpp voice input, Qwen3.5 on llama.cpp (0.8B to 9B, with native on-device vision on the 4B and 9B tiers), and stable-diffusion.cpp for offline image generation, with cloud providers available whenever you want them instead.

Full Drawing Suite

The same Clip Studio / Krita-class layered image editor as the desktop app, tuned for touch and stylus — pressure-sensitive brushes, layers, masks, and filters on a tablet screen.

Terrain & World Building

Procedural heightmaps, sculpt brushes, material-layer paint, sky, and water — the same terrain and world tools as MIA Studio, ported to run natively on Android.

Sculpt With Your Stylus

The desktop sculpting suite, running on the tablet: a non-destructive layer stack, paintable masking, Dynamic Topology, multiresolution levels, Face Groups, Curve tools, an Insert brush, Move/Snake-Hook and Polish, scan-wrap retopology, boolean ops, and an opt-in Beauty render. Pixel-density-aware painting snaps dabs to a texture grid for pixel-art straight onto the model, and triplanar projection textures a mesh with no UV unwrap. Two-finger twist sets brush size; a draggable orientation ball orbits the camera when every finger is busy sculpting. Cloud remesh, retexture, and rig passes are still one tap away.

New

Posable Mannequins

A new 3D tab opens on a grid of starter shapes and rigged male and female mannequins — pose one the way Clip Studio’s 3D dolls work, then save the viewport straight into the Draw tab as a reference layer and paint over it. No auto-rig step, and they sculpt and paint like any other mesh.

New

Node Graphs, Full Size

A tablet has the screen and the stylus precision for the real thing, so Lite gets the whole node-wire Material Graph and Brush Logic Graph editors, not a preset list — plus Photo-to-PBR with a neural depth pass borrowed over the desktop relay.

New

Node-Graph Workflows

The same visual Workflow tab as the desktop app and the Editor plugin — chain generation and automation nodes together, right from your phone.

One Unlock, Either Way

A 14-day trial, then a single one-time Pro unlock — buy it through Google Play, or activate an offline licence key if you installed it directly. The two are independent and grant exactly the same thing; no subscription, and nothing phones home to check.

New

Synced With Your Other Devices

Bring your own storage — S3, WebDAV, Dropbox, or Google Drive — and keep projects, brushes, and references in step between your phone and MIA Studio on desktop.

Android · Phone

MIA Pocket — the whole studio, one thumb

A phone-first build of MIA Studio Lite. The same drawing suite, sculpting tools, notes, chat, and on-device generation — rebuilt around a 6-inch screen held in one hand, with its own storage, its own keys, and no desktop app required.

Coming soon to Google Play

Built for One Hand

A fixed five-slot bottom bar — Projects, Create, Draw, Logic, More — puts every primary destination under your thumb, with a sheet for the rest. Editor panels dock as draggable bottom sheets so the 3D view keeps the top of the screen, tap targets never drop below 48px, and the layout survives the browser chrome collapsing and the notch.

Runs on the Phone Itself

whisper.cpp for voice, Qwen3.5 on llama.cpp for chat and on-device vision, stable-diffusion.cpp for image generation — all compiled native and shipped in the app. On a plane, on a trail, on no signal at all. Cloud providers are there when you want them, never required.

Sculpt on a Phone

Not a viewer — the real thing. Sculpt layers, masking, Dynamic Topology, multiresolution, Face Groups, Curve tools, an Insert brush, scan-wrap retopology, a Beauty render, and a draggable orientation ball to orbit with when every finger is already sculpting. Start from a preset primitive or a rigged mannequin on the 3D tab.

New

The Same Drawing Suite

The full layered image editor comes across intact — raster and vector layers, masks, blend modes, the brush engines, the animation timeline, PSD import and export. Pose a mannequin in 3D, snapshot the viewport, and paint straight over it in Draw.

Presets, Not a Wiring Canvas

One deliberate difference from Lite: a phone has no room to wire a node graph by hand, so Pocket drives the same material and brush engines through curated preset pickers instead. Same output, no pinch-zooming to hit a port.

Your Keys, Your Device

Projects live in on-device storage and sync through your own S3, WebDAV, Dropbox, or Google Drive — never ours. A 14-day trial, then one Pro unlock, through Google Play or an offline licence key. Read the privacy policy.

Quick Start

Up and running in a minute

Install from Fab, enable the plugin, and open the MIA tab.

01 Install the plugin

From the Fab Marketplace, add MIA to your project, then enable it in the Plugins window.

# In your project's Plugins folder, or via Fab's "Add to Project"
Edit > Plugins > Search "MIA" > Enable > Restart Editor
02 Open the MIA tab

Find it under the Window menu, then dock it wherever fits your layout.

Window > MIA > "Open Panel"
03 Connect a model

Pick a local backend like Ollama, or add your own API key for a cloud provider.

// Settings > MIA > Providers
provider: "ollama",
model: "qwen2.5-coder:7b",
endpoint: "http://localhost:11434"
04 Generate your first asset

Type a prompt in the Workflows tab and run it against a generation node.

# Workflows tab
prompt: "low-poly wooden crate, game-ready"
node: Generate Mesh → Import to Content Browser

Launching on Fab soon.

MIA's Fab creator account has been accepted — the listing itself isn't live yet. Follow along for the real launch date and a first look before it drops.