v4.0.26: Your AI Employee Now Directs AI Videos, With Native Voices in Every Language and a Price Before Every Click

Share
v4.0.26: Your AI Employee Now Directs AI Videos, With Native Voices in Every Language and a Price Before Every Click

Ask most AI video tools for a 30-second film and you get one of two things: a slot machine that burns credits every time you pull the lever, or a pile of clips you then stitch together yourself. v4.0.26 takes a different route. The Studio now has a Videos tab, and your employee runs the whole production: brief, storyboard, pictures, voices, music, edit, export, and a page to share it from. You see the price before every click.

There are two kinds of video, and the first one is free.

Animation: motion design that costs nothing until you export

Say what the video is for, and the employee writes a storyboard. Before writing, it looks at real motion design films close to your request (on Behance, Dribbble and Motionographer), tells you which one it is following, and keeps the screenshots in the project.

The film plays live in your browser above its storyboard. Nothing is rendered while you work, so iterating costs nothing. The layout is a real storyboard: the film big at the top, a strip with every plan always in view, then each plan as a card with its picture, voice-over line and what it shows.

You can act on any plan from its card. Change it, describe what you want in a few words, ask for three alternatives and pick one, or have it rewritten. Or open it and edit every field by hand, and the player follows each change straight away. Four styles (Dynamic, Elegant, Fun, Pro) set the motion, colours and fonts for the whole film, and you can switch between them in one click to compare.

For music, bring your own MP3 and the employee fits the plans to its length and beats. Without one, it picks a track on Pixabay Music, royalty-free under the Pixabay Content License, through your browser (via the Geta.Team Chrome extension) and fits it to the film.

When it looks right, export the MP4 on the server with a live progress bar, then publish it online in one click. The video gets its own page with a player, a download button and a preview picture for social networks. Change the film later, and the same button exports it again and updates the page.

AI video: every plan generated, and a director who checks the work

Next to Animation there is now AI video, where every plan is a generated, real-looking clip. This is where most of the work in this release went, and it shows in how carefully it spends your money.

It starts with a sentence, not a form. Describe the video, and the employee asks three to five short questions, one at a time, with big buttons: where it will be posted, how long, the mood, who speaks, photos to include. The format and length come from your answers.

Then it goes one step at a time, with the price on every step. A single card at the top says what comes next in plain words, with its cost, for example: "I create the pictures of the 6 plans · $0.90", then "Yes, go". Or choose "Do it all at once". Before any generation, a confirmation lists how many pictures, clips or voices, the price of each and the total, with prices fetched live from the provider. With every model now open, the default prices are only that: defaults. The film's real cost so far stays in that card.

A picture first, then its clip. Each plan gets one still to check, costing a few cents, before anything is animated. The clip then starts from that exact picture, so there are no expensive surprises. The music is made before the clips too, so you can watch the whole film as pictures, voices and music for free before paying for any animation.

Film memory keeps faces the same. Characters, places and objects that come back from plan to plan each get a reference picture, which can be your own photo. Every plan states who, where and what is in it, so the same person looks like the same person in plan six. Change a reference and the plans made with the old one are flagged.

Your employee reviews its own footage. It looks at contact sheets of the pictures, of the clips (start, middle and end of each) and of the cuts between plans. It warns you before animating a plan it has not checked, listens back to every voice to catch a wrong word, and looks at stills of the final export (subtitles, faces, lips) before telling you it is done. After each step the Studio shows the verdict: the good plans, and each plan to fix in plain words, with a "Redo" button and its price.

Characters can actually speak. A line can be read by a narrator or said by the character on screen, with lip sync. Each character can have their own voice. The sound is mixed for you: voices in front, music underneath, sound effects in the background, and the music steps back whenever someone speaks. Subtitles are burnt in by default, in short pieces timed to each line, for the many people who watch social video with the sound off.

Nothing is ever lost. Every picture, clip, voice, sound effect and soundtrack ever made stays in a history, and going back to any of them is free.

Every model open, and native voices in every language

The Studio no longer limits which models you can use. Pictures, clips, talking characters, voices, music and sound effects can each use any model available on FAL, searchable with its live price. Sensible defaults are in place (Nano Banana for pictures, MiniMax H3 for clips), and the employee picks the best quality for the price of each film and tells you why. If a model cannot do what the film needs, it is refused with a clear reason instead of quietly producing something else.

Voices got the same treatment. There are now thousands of native voices in French, Hebrew, Spanish and many other languages from the ElevenLabs library, without needing an ElevenLabs account. Pick by language and gender and listen before choosing. If you don't choose, a French film gets a French voice, not an English one reading French. These voices are generated through FAL and paid with the employee's FAL key. If you add your own ElevenLabs key, ElevenLabs voices go straight through your ElevenLabs account instead, and your own cloned voices become available too.

Every kind of generation (pictures, clips, voices, music and sound effects) runs on your employee's own FAL key, in line with how Geta.Team works: you bring your own keys and see your own costs. If the key is missing, you can paste it right in the Studio.

Add or delete a plan anywhere

Every plan now has a "+" on its edge that inserts an empty plan right there, between two existing ones. In an AI video, the employee writes the new plan so it links the plans around it, without generating anything until you say so. A bin on each plan deletes it after a confirmation.

A whole Studio project in one file

Any Studio project (site, app, document, slides or video) now exports to a single .studio file you can open with another employee or in another workspace. Nothing in the file names the platform, so it works the same on a white-label instance. Keys and passwords kept in the project's settings never go in it, and neither does anything that is online. A key someone hard-coded into a source file is part of the project's files, so it would travel with them: keep secrets in the settings. Test data (database rows, uploaded files) is only included if you tick the box, which is off by default.

Also in this release

  • The Cube shows your whole team. A new "My team" view shows each employee's state, live: working (and on what), done (with a short summary), or idle with whatever is waiting for you. You can answer an employee's question with OK or No right from the Cube.
  • Codex starts on GPT-6.2-Sol. Employees on Codex now default to the newest Codex model, with GPT-6.2-Astra also available and a new "ultra" effort level. Accounts that do not have it yet keep the newest Sol they have.
  • A disconnected mailbox now says so. When Google or Microsoft stops giving access to a mail account (access revoked, expired, password changed), its emails used to be lost without a word. Now the account shows "Reconnect" in red in Connectors, a dashboard banner names it, and the employee that reads it tells you once in the chat.
  • Studio keeps your place. After the employee changes something, a slide deck stays on the slide you were viewing and a document or site stays where you had scrolled.
  • Studio designs start from a reference. For a site, a landing page or an app, the employee first looks at the closest award-winning sites on Awwwards, says which one it is following and takes their visual codes (layout, type, spacing, motion), never their content.
  • Fixes: clips finishing at the same moment no longer overwrite each other; "Make the voices" works when voices are made all at once; a failed batch now says why instead of claiming it started; shared video pages always show the latest version and their own icon; subtitle outlines no longer draw lines through letters; no more surprise password box over a Studio project left open for hours.

Want to test the most advanced AI employees? Try it here: https://Geta.Team

Read more