Studio
Drop in an MP3 or WAV and the avatar performs it. Auto-configure the look from the audio, fine-tune in the inspector, render a high-quality MP4.
01/15 bubble — click to cycle
v3.2.0 — macOS (Apple Silicon + Intel) · Windows
Give your voice a face.
A desktop studio for audio-reactive avatars — perform any track, go live on your microphone, or let the built-in AI agent research, write, voice, and render finished videos. On your machine. On your keys.
The app
One window. A whole production floor.
Live stage, agentic recorder, and a real multi-track editor — these are actual screenshots, not mockups.
The content agent
Brief in. Finished video out.
The part a traditional editor doesn't have: scheduled jobs research your brief, write a ~30-second script, voice it, art-direct every scene, choreograph the motion, and render the MP4 — while you do something else.
Brief
Research
Script
Voice
Art direction
Motion
MP4
- An AI art director storyboards each scene from a ~6,250-piece line-art icon & graphics library — name any concept and the closest match is found.
- A motion-designer sub-agent choreographs on real spring physics: 11 named entrances, ambient accents, scene transitions — it can even author custom animations in a clamped motion DSL that can never break a render.
- A uniqueness engine fingerprints every clip — hooks, staging, choreography — so no two videos feel alike.
- Set the music bed to Generate and it composes a fresh instrumental to match each video's mood, mixed low and ducked under the voice.
- The Publisher writes cited blog and social posts from freshly fetched sources — approve one, and the agent scripts the video to tell the same story.
- Flip one switch and every run also lands as a real Editor project — pre-cut at scene boundaries, voice and music stems on their own tracks.
Writes with the LLM you already use: Anthropic OpenAI OpenRouter Straico Ollama · local
Avatar styles
Fifteen faces. All alive.
Live renders — not video. Every style is audio-reactive, deterministic, and transparent-background safe. Hover or focus one to make it speak.
Manual Recording Studio
Made by hand,
kept editable.
- An inline wizard walks source → camera bubble → microphone → keep-or-swap voice → branding → title cards → extras, saved as a reusable profile.
- Quick export bakes the live WYSIWYG composite straight to MP4 — optional mic enhance and AI voice swap included.
- Every raw layer is kept, so Open in Editor turns the same take into a fully editable multi-track project — camera bubble, branding, and title cards still separate clips.
- Teleprompter, countdown, max duration, keep-awake — the small things, built in.
The Editor
A real NLE
for the last 10%.
- Multi-track timeline with drag, trim, split, fades, cross-dissolves, titles, per-clip transform and volume.
- Chroma key: render the avatar on the Broadcast green preset, then key it over any footage.
- Save clip: mark an in/out range and file just that slice into the Asset Library — one long demo becomes many described, reusable assets.
- Autosaved projects with atomic writes, and MP4 export — streaming-ready
faststartfiles at 720p, 1080p, 4K, square, or vertical.
The modes
Eight rooms. One studio.
From a dragged-in audio file to a finished, ready-to-post MP4.
Live
Pick a microphone and go live — the avatar reacts in real time. Record takes, render them, or send them on to the Voice Lab.
Voice Lab
Send any take through the Voispark voice changer. Browse ElevenLabs and Cartesia enhanced voices, transform, A/B compare, use the result anywhere.
Record
An agentic screen-recording studio with a presenter avatar. One-click Polish — silence cuts, filler removal, voice swap — and an Auto-Director pass.
Camera
A clean, chromeless avatar window for OBS, Zoom, Meet, or Teams. Chroma-green Broadcast preset keys the avatar over anything.
Agent
A managed content factory: scheduled jobs research your brief, write a script, voice it, and render the finished clip — automatically, per brand.
Editor
A real multi-track timeline: drag, trim, split, chroma key, titles, fades, per-clip transforms, undo/redo, autosave, MP4 export.
Assets
A described, tagged library the AI casts from. Missing files are flagged, never silently broken — relink once and every project heals instantly.
Streamers
Your meeting camera, without the camera.
Click Camera and a clean, always-on-top avatar window opens. Capture it in OBS, key the chroma-green preset over your scene, or share it in Zoom, Meet, or Teams — your voice, with a face that isn't yours.
Agencies
One install, many clients.
A brand owns its entire world — API keys, asset library, knowledge, posts, editor projects, brand kit, even Demo Browser logins — and nothing leaks between brands. The setup wizard can crawl a client's website and write the brand brief for you.
Bring your own keys
Your keys. Your machine. Your videos.
Orb Studio ships with no API keys baked in and no account to create. Every key is yours, entered once, stored only on your machine, and sent nowhere but the provider you configured.
VoisparkVoice changing — ElevenLabs & Cartesia enhanced voices
VoiSpark MusicGenerated, royalty-free music beds
AnthropicScript writing & agentic demos
OpenAIScript writing
OpenRouterScript writing — any routed model
StraicoScript writing
OllamaLocal models — no key at all
A first-run Setup Wizard walks each service with a plain-English note, a link to get a key, and a Test key button — and the built-in Guide documents every mode, searchable and fully offline.
Download
Get Orb Studio 3.2
No account, no sign-up — it's a desktop app. Pick your machine:
Which Mac do I have? Apple menu → About This Mac. "Apple M1/M2/M3/M4…" → Apple Silicon; otherwise Intel.
MP4 encoding uses ffmpeg on your PATH: brew install ffmpeg (macOS) or winget install ffmpeg (Windows). Everything except the final MP4 conversion works without it.
Unsigned builds, for now. macOS: right-click the app → Open, the first time. Windows SmartScreen: More info → Run anyway. Signed, auto-updating builds are on the roadmap.
FAQ
Questions, answered.
Do I need an account?
No. There's no sign-up, no login, and no Orb Studio server. You install the app and, for the features that call AI services, enter your own API keys.
Which services need keys — and where do my keys live?
Voice changing uses Voispark, music generation uses VoiSpark Music, and script writing uses your choice of Anthropic, OpenAI, OpenRouter, or Straico — or a local model via Ollama with no key at all. Keys are stored only in the app's local data folder on your machine and are sent nowhere except the provider you configured.
Does it work offline?
The core studio does: performing audio, the live avatar, the camera window, recording, editing, and rendering are all local. Features that call a provider — voice transforms, music generation, agent script-writing — need the network (or a local Ollama model). The built-in Guide is fully offline too.
Can the avatar be my webcam in Zoom or OBS?
Yes. Camera mode opens a clean, chromeless avatar window: capture it in OBS (Window Capture → Orb Camera), or share the window in Zoom/Meet/Teams. For a keyable avatar, pick the Broadcast chroma-green preset and add a chroma key filter. Route OBS's Virtual Camera to use it as a true system camera.
Why does my OS warn me on first launch?
Current builds are unsigned — macOS Gatekeeper and Windows SmartScreen flag any app that isn't code-signed, regardless of what it does. macOS: right-click the app → Open, once. Windows: More info → Run anyway. Signed and notarized builds with auto-update are on the roadmap.
Do agent jobs run when the app is closed?
Jobs run on their schedule while the app is open — everything happens on your machine, so there's no cloud queue running without you. Renders land in your output folder (default ~/Movies/Orb Studio) with a JSON sidecar of the script and voice settings.