AI Embedded Systems
Public proof ยท Active development

AIWF Studio: a local-first creative AI workspace.

AIWF Studio brings image, inpaint, video, audio, model, data, and workflow tooling into one Windows-oriented application. The work demonstrates the same system concerns that matter in client projects: routing, runtime state, hardware limits, source control, visible configuration, and failure handling.

AIWF Studio workflow code blocks with a drag-and-drop queue and captured generation settings
Workflow blocks preserve route, model, precision, and captured settings for reviewable execution.
ReactProduction interface
FastAPIApplication and route layer
LocalWindows and NVIDIA focus
MultiImage, video, audio, and model lanes
The problem

Local AI tools become fragmented before the work becomes repeatable.

Models, runtimes, settings, sidecars, queues, receipts, and outputs often live in separate utilities with no shared operating view.

Model sprawl

Different model families need different precision, attention, memory, and loading behavior. One global setting is not a safe operating model.

Invisible runtime state

A polished button is not enough. The user needs to know which backend, model, device, precision, and resource path is active.

Broken continuity

Without saved settings, receipts, and workflow blocks, a successful output can be difficult to reproduce or hand off.

Consumer hardware limits

VRAM, RAM, storage, optional SDKs, and model size have to be treated as product constraints, not installation footnotes.

Design goals

Expose the operating state without making every user assemble the stack.

Local-first operation

Run core creative AI workflows on controlled Windows and NVIDIA hardware where models and resources permit.

Family-aware routing

Keep model-family behavior, backends, precision, sidecars, and optional engines explicit instead of applying one hidden default.

Reviewable workflows

Capture settings as blocks, preserve execution order, and keep route metadata available for later review.

Visible readiness

Startup and runtime checks distinguish available, optional, missing, idle, and failed capabilities.

Architecture

One product surface, multiple specialized lanes.

The production path uses a React interface and FastAPI application layer. A broader Gradio lab remains available for experimental and blueprint surfaces.

React frontend

Dedicated workspaces for image, inpaint, video, audio, models, data, settings, monitoring, and workflow blocks.

FastAPI backend

Bootstrap, capabilities, runtime state, settings, logs, data, and generation contracts are exposed through explicit application routes.

Model and engine layer

Multiple inference backends and model families are routed through family-specific loaders and runtime checks.

Install and receipts

The project includes Windows installation paths, dependency setup, default model bootstrap, build steps, and local verification receipts.

Interfaces

Current working surfaces.

These screenshots are from active development builds, not concept art.

AIWF Studio video workspace showing model, canvas, runtime status, and hardware resource usage
Video workspace with route controls, local runtime state, and live resource information.
AIWF Studio unified lab workflow for image, video, and audio processes
Unified labs reveal only selected stages and resolve execution order for image, video, and audio work.
What works today

Current implementation, stated without a maturity shortcut.

Production UI path

The React and FastAPI surface serves bootstrap, runtime, capability, settings, logs, data, and workflow requests in local development and clean-install checks.

Creative workspaces

Image, inpaint, video, audio, model, data, and settings lanes are represented in the current application.

Workflow capture

Generation settings can be captured into movable blocks with route and model context for repeatable plans.

Windows installation

A documented clean-install pass created the Python environment, installed runtime dependencies, built the frontend, created shortcuts, and reached the local API and UI.

Current limitations

Active software, not a finished universal appliance.

Hardware varies

Model support and performance depend on GPU, VRAM, RAM, storage, driver, model family, and requested workload.

Optional SDKs

Some acceleration and media features require separate vendor SDKs. Missing optional components should reduce capability, not break the core install.

Route maturity differs

Production React routes and broader experimental lab surfaces do not all carry the same release confidence.

Stable download pending

The public download remains intentionally unlinked until a versioned stable branch or release artifact clears installation, license, and smoke checks.