Concepts

Concepts

Studio is an agent-driven content factory: a prompt goes in, a finished video comes out. Understanding the handful of objects it works with makes the web app — and the API — read clearly.

The objects

Project — the project is the video. This is the mental model that matters. A project isn’t a folder of unrelated attempts; it’s one piece of content that gets better over time. It holds your brief and the settings that apply across every language and render of that idea, and it always has a current version — the take that opening the project shows you.

Version (run) — one trip through the pipeline. Every new take, retry, or aspect-ratio change produces a new version. The newest good one becomes the project’s current version; the earlier ones stay available in the versions dropdown, so you can look back at, or return to, any take the project has ever had. Under the hood these are called runs — that’s the word the API uses.

Variant — one language of a project. The first language you ask for becomes the primary variant; each additional language is its own variant. Variants exist so the same project can ship in many languages while sharing everything that isn’t language-specific.

Variant vs. version

A variant is a language of the project; a version is a take — one trip through the pipeline. Asking for a second language adds a variant; asking for another take of the same language adds a version. Most projects you start from a single brief have one variant and accumulate versions as you iterate.

Scene — once a version reaches the approval stage, Studio materializes its scenes as first-class, editable rows: each has a script line, a voice, an image, a duration, a transition, and optional effects. Scenes are what you edit, whether you chat with the production agent or use the controls on the scene card.

Render — a produced MP4 (plus a thumbnail) for a version. A version can have several renders over its life — the pipeline render, plus any previews you ask for while editing.

The pipeline

Every version drives the same ordered pipeline:

1. Script — drafts the narration and scene breakdown from your prompt.
2. Voiceover — synthesizes narration audio for the variant’s language.
3. Visuals — produces an image per scene. Studio can pull licensed stock photography or generate images, governed by a per-scene or per-video image-source preference you set in chat.
4. Compose — assembles the scenes, audio, and visuals into a composition.
5. Render — turns the composition into a finished MP4 and thumbnail.

After the render, the version enters awaiting_approval. You review it, edit scenes if you want, then approve — which is the moment the video is registered in Drive.

Reaching approval means the steps really ran

If a service the pipeline depends on fails, the build fails and shows you the real error. Studio does not paper over a broken step and hand you a finished-looking video — the one narrow exception is a single scene whose voice synthesis fails, which is marked as silent so the build feed never claims a voice that isn’t there.

Visuals are language-agnostic

When you fan a project out across languages, secondary-language variants reuse the primary variant’s images and composition. Only the script and voiceover re-run per language — so a five-language project doesn’t pay the visual cost five times.

Rendering

The last step turns the composition into a real MP4. That happens on Studio’s own internal render backend: there is no separate product to sign into, no extra tile in the launcher, and nothing for you to configure or pay for on its own. Rendering is simply part of Studio.

What a finished video actually looks like

It’s worth setting expectations concretely, because the two sources produce genuinely different pieces.

A prompt project is a narrated photo film. Studio renders three still images, each given one of six fixed two-dimensional camera moves — slow zooms and pans — with cross-fades between scenes, one narration line per scene, background music ducked under the speech, and burned-in captions.

A repository project is a motion-graphics piece, usually around four scenes: an animated background, a headline that assembles itself, a real screenshot of your app drifting inside a rounded viewport, an optional flow diagram of how the system fits together, and a closing headline. It’s narrated, and it carries no burned-in captions. The screenshot is a still frame, not a screen recording.

Set the register with a template

The overall look — the style preset (explainer, cinematic, kinetic-bold, cozy, plain, reporeel) along with captions, quality, music volume, and branding — lives on the template you pick at creation. See Templates and brand below.

Voices

Studio narrates with the internal Destesi voice engine, and you pick the voice when you create a project — filtered by gender if you want, with Aria as the default when you don’t choose.

The picker and the language pills constrain each other, which prevents the most common mistake: asking for a voice that can’t speak the language you asked for.

Every voice in the picker speaks each supported language, so choosing more languages never narrows the list.

Templates, brand, and music

Templates hold the reusable part of a look, so every video you make from one starts on-brand. A template carries the video format, whether captions are burned in, the render quality, the music volume, a logo (with its corner placement and scale), a background, intro and outro cards, and the style preset. You can create, edit, duplicate, and delete them, and attach rules that decide when parts of the template apply rather than applying everything to every video.

The assets library is your workspace’s own uploaded media — images and audio you can reuse across projects.

Background music comes from that library. Upload a track once, attach it to a video, and set its volume against the narration. There is no stock music catalog; the library is your own audio.

The blocks catalog lets you browse the text effects, transitions, and overlays Studio can apply, so you can see what’s available before asking for it in chat. It’s browse-only — you apply blocks through the editor, not from the catalog.

Publishing

Publishing is real, not a stub. Studio uploads finished videos to YouTube, TikTok, Instagram, Facebook, Threads, and LinkedIn using the accounts you connected in Connect — immediately or on a schedule. A video must be approved before it can be published. Several behaviors are worth knowing before you rely on it:

• It fails loudly. If publishing isn’t configured or the channel isn’t connected, the publish is marked failed with the reason. Studio never records a success for a post that didn’t happen.

• It retries carefully, and never twice on the platform. Up to four attempts, backing off roughly 2, then 10, then 30 minutes — but only for failures that happened before anything reached the platform, like fetching a credential or downloading the video. Once bytes have gone to YouTube or Meta, the attempt is final. A duplicate public post on your channel is worse than a post that needs a human to look at it.

• Visibility is offered only where it is real. Each destination honors a different set, and Studio will not let you pick one that would be overridden:

  Destination	Visibilities it can honor
  YouTube	public, unlisted, private
  TikTok	public, private
  LinkedIn	public, private (private means connections-only)
  Instagram, Facebook, Threads	public only

  One visibility applies to a whole batch, so selecting destinations that disagree is refused up front with the reason — publish those accounts separately instead. The rule is enforced when the request is accepted and again before dispatch, so a combination cannot slip through. This exists because a visibility you picked and the platform silently overrode is the worst outcome available: it once meant a “public” TikTok post landed private and invisible while Studio recorded it as published.

Studio shows each publish’s state — queued, scheduled, published, or failed with its reason — including publishes you scheduled for later, so a post waiting to go out is visible rather than silently pending.

Scheduling offers suggested posting times per platform. Treat them as a sensible general default — they’re not derived from your audience’s behavior.

Editing by conversation

Studio’s editor is a chat. Instead of dragging timeline clips, you tell the production agent what you want — “swap the voice on every scene to something warmer”, “make scene 3 a clean fade”, “shorten the intro” — and it edits the underlying scenes for you. Each request becomes a tracked operation you can watch complete, retry if it fails, or roll back. You stay in control of when a preview re-renders; the agent only re-renders when you ask to see the result.

This is editing at the level of intent: you describe the change, Studio makes it across the scenes, and the next render reflects it.

There are two chats. The chat home is where you talk to Studio about your projects — finding them, creating them, getting started. The project chat is the editor above. Both save their conversations to your workspace, so you can rename one, come back to it later, or delete it. Some actions pause and ask you to approve them before they run, so an agent turn can’t quietly do something expensive on your behalf.

Editing a scene directly

Not everything needs a sentence. Each scene card also exposes direct controls: change its duration, regenerate the image or edit the image prompt, upload your own image, change the voice or regenerate the narration, edit the script, edit dialogue turns for multi-speaker scenes, set the transition and a filter, and lock a scene so a regenerate-everything pass leaves it alone. You can also duplicate, delete, reorder, and insert scenes.

Repository videos are composed, not photographed

A repository-sourced video’s scenes are generated motion graphics rather than per-scene images, so they don’t offer the per-scene visual controls above. Edit those videos through the script and the template.

If the queue that builds videos stalls, a worker banner appears at the top of the app and says so — you find out that nothing is being processed instead of watching a build sit still and wondering.

Getting the video out

Download it. The download menu gives you the MP4, the subtitles as a .srt file, the transcript as .json, and a temporary link you can paste somewhere. Studio keeps a render history too, so you can go back to an earlier render of the same version.

Share it publicly. Any version with a render can be shared with a public link, optionally with an expiry. Minting one creates a share token for that version; the token is embedded in a share link that opens a stripped-down public viewer — just the video, no editor, no sign-in. Because the token is the credential, anyone with the link can watch and anyone without it gets nothing. Disabling sharing invalidates the token immediately, so the same link stops working for everyone. Sharing is per-version: each take you share has its own link, and turning it off on one doesn’t affect another.

Get told about it. Attach a webhook to a video and Studio calls your URL as it progresses, so another system can react without polling.

Formats and quality

Aspect ratio is set at creation and can be changed afterwards; changing it re-renders the video for the new shape. You can also render every aspect at once as a bundle when you need the same video vertical, square, and wide.

Render quality comes in three presets — draft, standard, and hq. They trade encoding time against output fidelity by changing the encoder preset and frame rate. They do not change the resolution: a draft render is the same size as an hq one.

How Studio fits the suite

Studio is a peer product in the Destesi suite, unified with everything else by one login and one workspace.

• One workspace. Projects, runs, and renders all belong to the workspace you’re in. Switch workspaces and you switch the whole library.
• One sign-on. You open Studio from the launcher with the same single sign-on session used across the suite — no separate account.
• Credentials live in Connect. To read private GitHub repositories or publish to social channels, you connect those once per workspace in Connect. Studio fetches the credential at call time and never stores it.
• Everything lands in Drive. Every finished video and thumbnail Studio produces is registered in Drive, so your rendered content sits in the same one library as artifacts from every other product.

Next steps

QuickstartYour first render, the fastest way.

ReferenceHTTP endpoints, config, and environment.

TroubleshootingWhen a build misbehaves.