AI Video Tutorial Generator

Turn a question, topic, or source into a tutorial you can inspect and edit.
Research, script, storyboard, narrate, animate presenters, refine the timeline,
and export from one Windows desktop workspace.

▸ For users  ·  ▸ See it in action  ·  ▸ AI options  ·  ▸ For developers  ·  ▸ Documentation

▶ Watch the full 45-second demo with sound (MP4)

_{The 17-second loop previews a real lesson; the sound demo continues into the app.
Music: “Immersed” by Kevin MacLeod, CC BY 3.0.}

Current status: 2.0 is a local release candidate. A public installer is not available yet.

For users

From an idea to an editable lesson

AI Video Tutorial Generator is built for explanations that need more than a talking-head export. Start with a topic, learner question, script, or source material, then keep control at every stage:

1. Choose the audience, length, visual theme, research mode, and an AI profile.
2. Work without a presenter or select a cast of up to four.
3. Review objectives, evidence, outline, script, and scene assignments.
4. Approve the storyboard and source rights before media generation begins.
5. Generate or reuse narration, captions, illustrations, presenter clips, and scene renders.
6. Refine the result in the timeline, review quality findings, and export the finished tutorial.

A studio made for tutorial work

• Guided planning with real review points. Creative, Grounded, and Strict modes carry claims,

• A practical editor. The Studio workspace combines scene preview, timeline, waveform, captions,

• Changes stay focused. Scene edits, pronunciation fixes, and accepted media candidates create

• Durable background work. Generation and downloads report progress, survive navigation, and

• Useful exports. Video, captions, transcript, bibliography, chapters, metadata, thumbnail,

Presenters and multi-cast tutorials

The casual presenter collection contains fourteen original fictional tutorial hosts across realistic, anime, cartoon, robot, cat, kitten, dog, puppy, tiger, and lion styles. Choose none, one, or up to four presenters; a cast takes turns across scenes, and each scene can have its own speaker and voice.

Presenter animation is optional. Eight exact portraits have passed short-sample review with SoulX-FlashHead Pro, including realistic, anime, cartoon, and kitten hosts. Their clips were checked for mouth anatomy, full blinks, identity, and background stability on a 12 GB NVIDIA laptop GPU. The other six current portraits remain available for still-image use while their exact animation routes are pending or incompatible. See the current comparison for measured results and rejected alternatives. A successful short sample does not guarantee every new image or long recording will animate well.

You can upload a portrait or create one with a configured cloud or local image model, save it in the reusable presenter gallery, and add it to a tutorial cast. Custom animation stays blocked until you run, play, and accept a short local SoulX preview for that exact portrait and model revision.

Models and downloads

Onboarding and Models & Providers use the same native download queue. Choosing a package during onboarding can queue it immediately; on the Models page, inspect the package and choose Download. Compatible image packs may first ask to install the shared ComfyUI runtime. The progress panel can be minimized while you keep working. The queue reports bytes and phases, resumes supported transfers, reuses verified files, and keeps model installation separate from model readiness.

Local weights download separately from the desktop app. The app checks package integrity, runtime compatibility, and hardware before marking a model ready.

The managed SoulX package downloads about 10.4 GB and assembles its pinned Windows environment offline. Download stages it; Use model verifies and activates it. No separate Python setup or terminal is required.

Music that fits the lesson

Search free music by topic and mood, preview a track, and add the chosen recording to the project. The app preserves its source, license, and attribution. Timeline music can loop, use its own gain, and duck beneath narration; attribution follows the finished export.

Local and cloud options

You can begin without an image-generation key: Designed lessons use authored slide layouts, and the included teaching backgrounds and elements are ready in the Library. Illustrated lessons can use an explicitly selected image route after you review the candidates.

| Path | What it is good for | What to know | | --- | --- | --- | | Included assets | Slides, backgrounds, diagrams, and teaching elements without an API call | Works before a project exists and does not consume a provider quota | | Local image generation | Private, repeatable illustration through the managed ComfyUI/SDXL route | No API fee after download; uses local storage, power, and GPU time | | Local presenter animation | Eight exact bundled portraits through reviewed SoulX routes; custom portraits after an accepted SoulX preview | Optional large runtime; the other six current portraits remain still-only until an exact route passes review | | Cloud text and research | Writing and research with Groq, Gemini, Mistral, OpenRouter, NVIDIA NIM, or another reviewed profile | Uses your key and the exact provider/model you select; no silent fallback | | Cloud images and licensed media | Optional generation or discovery through reviewed image and stock-media routes | Every accepted asset still keeps provenance, rights, and attribution |

Several providers offer free or evaluation allowances. See the dated free and trial provider guide for options and current-source links. Local inference uses your hardware and has no per-generation API charge.

Installation and requirements

There is no public 2.0 installer yet. The current build is available only for local release-candidate evaluation; installation artifacts have not been published.

The planned Windows baseline is:

• 64-bit Windows 10 or 11 with Microsoft Edge WebView2;
• 8 GB RAM minimum for desktop/cloud work, with 16 GB or more recommended;
• about 25 GB free for the application and working files, plus any optional model packs;
• an optional NVIDIA GPU for supported local generation and presenter runtimes.

For developers

Architecture

The application keeps privileged operations behind small typed boundaries:

React + TypeScript interface inside Tauri
                 │ typed commands and events
                 ▼
Rust desktop core
  projects · credentials · downloads · diagnostics · worker supervision
                 │ authenticated local IPC
                 ▼
Python pipeline — sole project database writer
  research · providers · revisions · jobs · QA · export
        ├── TypeScript scene renderer + pinned Chromium
        ├── managed FFmpeg / ffprobe
        ├── optional local model workers
        └── isolated document, code, and media workers

| Path | Responsibility | | --- | --- | | apps/desktop | React UI, Tauri host, native commands, downloads, and app lifecycle | | packages/contracts | Cross-language JSON Schema plus generated TypeScript, Python, and Rust bindings | | packages/scenes | Typed scene families, target-aware layout, and deterministic preview/render data | | packages/themes | Theme packs, starter kits, and the bundled presenter catalog | | services/pipeline | Projects, evidence, generation workflows, providers, presenter routing, QA, and exports | | services/renderer | Deterministic Chromium frames and media render planning | | docs | Architecture decisions, setup, provider policy, research, and acceptance evidence |

The renderer does not receive unrestricted filesystem, process, network, or credential authority. The Rust host supervises private workers, and the Python pipeline is the only writer to a project's SQLite database. Binary artifacts are immutable SHA-256 objects; accepted changes create revisions.

Set up a development machine

Use Windows PowerShell with the exact versions in runtime-manifest.json: Node.js 24.20.0, pnpm 10.15.1, Python 3.12.13, uv 0.12.7, and Rust 1.96.1 with clippy and rustfmt. Install Visual Studio C++ Build Tools and WebView2 as well.

corepack enable
corepack prepare [email protected] --activate
.\scripts\bootstrap.ps1 -Check
.\scripts\bootstrap.ps1
corepack pnpm doctor
corepack pnpm dev

pnpm dev starts the native Tauri app and its authenticated pipeline worker in one process tree. Use node scripts/dev.mjs --check for a headless readiness check or corepack pnpm dev:web for browser-only interface work.

Run checks

corepack pnpm lint
corepack pnpm typecheck
corepack pnpm test
corepack pnpm test:fixtures
corepack pnpm test:e2e
corepack pnpm render-test

Default tests use deterministic fixtures and do not prove a live provider, GPU route, native WebView2 flow, clean installation, or final media quality. Changes to those surfaces need their corresponding bounded native or live acceptance run and direct visual or audio inspection.

Build a local Windows candidate

.\scripts\build-windows-sidecar.ps1
corepack pnpm package:desktop

This creates local artifacts only. It does not publish a release, bundle model weights, or authorize distribution. Runtime packs and model downloads stay separately pinned and verified.

Documentation

• Documentation map
• Architecture overview
• Project format
• Windows setup
• Rendering
• Provider and model policy
• Security and privacy
• Accessibility
• Testing and evaluation
• Contributing
• Release policy

License

The application's original code and artwork are released under the MIT License. Dependencies, model weights, datasets, fonts, provider services, stock media, and generated output retain their own licenses and terms. Review those terms before redistributing a complete build or its output.