▶ Watch the full 45-second demo with sound

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

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.

Projects remain ordinary local folders. Their revisions, job state, source copies, usage records, and content-addressed media stay with the project rather than a hosted product account.

• Guided planning with real review points. Creative, Grounded, and Strict modes carry claims, citations, objectives, and unresolved support into the plan instead of hiding them in a prompt.
• A practical editor. The Studio workspace combines scene preview, timeline, waveform, captions, titles, media, presenters, and an icon-switched inspector. Side panels can be resized so transcript and media work get the space they need.
• Changes stay focused. Scene edits, pronunciation fixes, and accepted media candidates create revisions and invalidate dependent work without discarding unrelated completed assets.
• Durable background work. Generation and downloads report progress, survive navigation, and expose retry, resume, cancellation, or a clear blocker where that operation is supported.
• Useful exports. Video, captions, transcript, bibliography, chapters, metadata, thumbnail, provenance, and portable .alytutorial project archives share one reviewed project state.

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. Successful short samples do 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.

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.

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.

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.

Several providers offer free or evaluation allowances. Local inference uses your hardware and has no per-generation API charge.

Download the Windows installer from the latest release.

Run the installer and open AI Video Tutorial Generator from the Start menu. The core Python, Node, Chromium, and FFmpeg tools are bundled. Onboarding offers optional model downloads and cloud provider setup; no terminal or developer tools are required.

The 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.

The app checks for updates automatically on launch and every six hours. When one is available, choose Update and restart; pending edits are saved before the signed installer runs. You can also check manually in Settings & diagnostics. Project files and downloaded models live separately from the installed application and remain in place during updates.

The installer is not Authenticode-signed, so Windows may show an unknown-publisher prompt. Update packages and the bundled runtime have separate cryptographic signatures.

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

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.

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 pnpm@10.15.1 --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.

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.

The Release Windows installer GitHub Actions workflow builds and publishes each stable vMAJOR.MINOR.PATCH tag. It packages the Python worker, pinned rendering tools, and desktop app, then uploads the installer, signed updater feed, checksums, and dependency inventory. The existing signing keys are stored as repository secrets so installed copies can trust future updates.

To prepare the next version, run node scripts/set-release-version.mjs 1.8.2, commit the changes, and push the corresponding v1.8.2 tag. A draft release can supply new notes; otherwise the workflow reuses the preceding release notes with updated version links. Failed builds can be retried from Actions using the existing tag. Published versions are never overwritten.

For local packaging:

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

The commands above build the desktop shell. For a standalone installer, first build the renderer and run scripts/stage-windows-release.mjs with the hash-pinned runtime root, staging directory, private signing-key directory, and built worker. Set TAURI_SIGNING_PRIVATE_KEY to the existing updater key, then build with tauri build --bundles nsis --config <staging>/tauri.release.json. The runtime and updater public keys are tracked; private keys must be retained securely outside the repository for future updates. Optional model weights remain separate downloads.

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.