Video Studio

Brief, script, render, review, publish, and what the video did afterwards – all on one document. Every video keeps one copy in your library, where the calendar, the social scheduler and your reports pick it up.

Open it from Studios → Video Studio, or ask for a video in chat and it lands here. There is one video studio: an older link or bookmark to Instant Video, AI Video Studio, Instant Ads, the Shorts Generator, the UGC Ad Generator or Video to Shorts opens Video Studio at the format that does that job, and anything you asked an agent to make with one of those names still runs here.

Start with a format

The first screen asks what you are making. The format decides how the studio works; everything else you can change on the next screen.

Ad from a brief. Start from what you know about the product and the customer. The studio builds a research brief – who buys, what the competition misses, why this product works, what changes for the person – then writes five scripts, one per hook, and you pick the one that sounds like your brand. Choose a scene style and an actor, and the scripts become scenes that render, caption and publish as one video.

Prompt to video. Describe the video in your own words. The studio reads the intent and writes the generation prompt the model needs, with your brand and language carried through. Add a first frame if you have one. A short idea stays one shot; anything from twelve seconds is broken into beats.

Actor batch. One script, every face in your cast. Pick the actors and the studio makes a variant per actor, each one its own document with its own review and its own performance history, so you can compare them and publish the one that works.

Article to video. Bring a link, paste the text, or pick an article this brand already published. The studio reads it, finds the claim worth hearing out loud, and retells it as a short: a headline hook, a few beats, on-screen text and captions in your brand colours. Nothing is said that the article does not say. If the link is dead or behind a paywall, the studio tells you on the spot rather than three steps later.

A link to your own site is written as an ad; a link to an article is retold. The studio reads the page and says which it thinks it is, in one line under the source field, before anything is bought. A home page, a pricing page or a feature page is your own writing, so the scripts are allowed to make the case, take a position and ask for something, and they still may not say a single thing the page does not say: no invented number, no promise the page is careful not to make, and the ask comes at the end rather than the opening. Somebody else's article is retold exactly as before, in its own stance, with the publication named once. If the studio reads your page the wrong way, the same line lets you say so in one click, and your word wins.

The five scripts are five different things. Each one owns a different facet of what was read – the idea, the moment it matters, the proof or the caveat, how it works, and the offer – and the studio says which on each card. Five doors into the same three facts is not five scripts, so when two of them come back saying the same thing, one is rewritten before you ever see it.

Repurpose a long video. Bring a webinar, a talk or an interview. The studio transcribes it, reads the transcript for the passages that stand on their own, cuts them to the shape you publish in, and captions them. Every clip is your own footage and your own words.

The scripts

For an ad from a brief and for an article, the studio writes five scripts, one per hook. Read them aloud – the one that sounds like a person is the one that performs. You can edit the dialogue inline before you commit to one.

Each scene shows how many words fit the time it has. A script that runs long is trimmed on the hold at the end, never on your words.

Credits

The estimate sits next to the Render button before you press it, broken down line by line. Scenes are charged one at a time as they render, and a scene that fails is refunded. Cancel a render and everything unrendered comes back.

Fixed steps:

An actor is made once and used forever, so it is charged once: making a cast member in the Actor Library generates several pictures of the same person – front, angles, expressions – at one credit each, and any picture that fails to arrive is refunded. Four pictures is the default, six the most.

An actor from a photograph. If the person you want in your videos already exists – your founder, your presenter, a model you have hired – you do not have to describe them. In the Actor Library, New actor, choose From a photo and upload one picture of one person. We read the photograph and write the description for you, in two parts you can edit before you buy anything: who they are (face, skin tone, eyes, hair down to the roots and the wave, build) and how they were photographed (the wardrobe, the pose, the room and the light). Reading the photograph costs nothing; only the pictures are charged, at the same rate as any other actor. We describe what is visible and nothing else: we never guess where someone is from, how old they really are, or anything about their health or who they might be, and a photograph with more than one face in it, or with a child in it, is declined with a plain sentence instead of a description. Before the button will work you confirm one line – that you have the right to use this person's likeness in videos for this brand – and we record that confirmation, with its date, on the actor itself. The picture you uploaded stays on the actor as the strongest likeness we hold, and every film you cast them in is built from it.

Rendering is charged per second of finished video, rounded up per scene. The rate is the model and the resolution you render at, because that is what the model charges us for: a second of 1080p costs twice a second of 720p on the same model. Both are on the form, and each resolution shows its own rate.

The premium models open at 1080p. The two budget models open one rung down, at the resolution a feed actually plays, so the cheap seat is cheap: fifteen seconds on Wan is 15 credits, and the same fifteen seconds on SeeDance 2.0 is 38.

A twenty-second video in three scenes is three charges, not one.

A fast style buys one clip per beat, and the clips are longer than the beats. A brainrot cut is a beat every second and a half, and no model in the set renders a clip that short: the shortest Wan will make is four seconds, the shortest Kling five. So the studio buys one clip per beat at the model's shortest length and cuts each one back to its beat in the edit, which is where the hard cuts of that style come from. The estimate says both numbers in words, and the seconds you pay for are the seconds bought: eight beats on Wan is "8 beats · 32s bought", 16 credits at 480p, for a film that plays for fifteen seconds. Asking the model to make the cuts instead does not work – video models render one shot, whatever the prompt asks for.

Styles whose beats are already long enough (Someone talking at five seconds, Composed story at six) buy exactly what they play, and nothing is cut back.

A reference costs one image a shot. When you give a film a product or a mascot to show, the studio draws each shot's opening frame with the thing placed in it and hands the model that frame, rather than handing over the picture and hoping. That is one image per beat at 1 credit, listed on the estimate as its own line. You can turn it off in How it renders; the picture then goes to the model as it is, which is cheaper, and the thing tends to arrive as a sticker in the corner rather than an object in the room.

A brand that has reached its credit budget for the period is told so, with who can raise it – that is a different thing from an empty wallet, and the studio says which it is.

While it renders

The render runs on our side. Close the tab, come back later, open the video from the list or from your library – the page picks it back up. Each scene shows its own state: queued, with the model, rendering, done. A single scene that fails can be retried on its own, and a retry never re-buys a scene that already landed.

Captions and presenter

Captions are burned into the finished video by us, after the scenes are rendered. The studio transcribes what was actually said, lines it up with the script, and writes the words onto the picture in your brand colours. Three styles: Word highlight (one line at a time, the spoken word lit as it is said – the short-form default), Clean line (a quiet subtitle, no highlight), and Bold centre (two or three words, centred and large, for a video watched with the sound off).

Captions and reframing also work on a video you already have. Ask in chat to caption or recrop a film that is already in your library and it happens on that video, at the prices in the table above – there is no need to render a second one.

Presenter is who says it. On an article you can leave it empty and the short runs text-led – on-screen text and captions carry the piece, and nobody appears. On the other formats the same choice is called Actor: pick a face from your Actor Library and it holds across every scene. Most of the models speak the script themselves; Hailuo 3 Max does not speak at all, so it is for footage and product scenes rather than a talking one.

Reference is what every shot has to show, which is a different question from who is in it. Pick A product and the film keeps the same product through every shot, unchanged in shape, colour and label; pick A mascot and it keeps the same character, one design in every shot. The picture can come from your catalog, from an image this brand already has, or from your own upload, and the choice works under every style: a cartoon draws the mascot from it instead of inventing one, a message thread puts the product in the footage behind the card, and a presenter holds it up. Where the model you chose has only one image slot and something else is already in it – your presenter's photograph, or a first frame you picked – the studio says so on the row and keeps the product in the direction it writes. A first frame you picked holds only the opening shot, so the picture still reaches every shot after it, and the row says that too. The picture has to be this brand's: one from another brand's library is refused rather than filmed.

Review

The Review tab carries the video's status – Draft, In review, Changes requested, Approved – and the comment thread your library uses: @-mention a teammate, reply, resolve. Approval is off by default and is turned on per brand under Brand → Messaging, where you also choose who approves. With it on, a video is approved before it can be published.

Publish

The Publish tab posts through the accounts this brand has connected: YouTube, TikTok, Instagram, Facebook, LinkedIn. Accounts on your other brands are not offered here – a video publishes through the brand it belongs to.

Write one caption and the studio adapts it per platform. Publish now, or pick a time and it lands on the calendar, where it can be moved or cancelled until it goes out.

Each platform asks for its own things, and the studio asks before it sends rather than after:

• YouTube needs a title of 100 characters or fewer. A vertical video under three minutes posts as a Short.
• Instagram Reels need a cover frame. A frame from the video is offered; without one the platform picks the first frame.
• LinkedIn posts as the page or as the person, chosen at publish time.
• TikTok is the one that sends you elsewhere. Who can see the video, the preview confirmed, and your consent are the creator's own choices, and TikTok requires them to be made deliberately. The studio never guesses them: it hands you to the social composer, you make the choices there, and the post goes out from there. A TikTok publish with those settings missing is refused, not defaulted.

Videos published from here also declare themselves as AI-generated where the platform asks: YouTube's synthetic media flag and Instagram's AI label are set for you.

Performance

The Performance tab shows what the video did once it was out: views, watch time, average view, engagement, and clicks on the links in the caption. Underneath, the accounts around it over the last 28 days – labelled as account numbers, because a channel's watch time is not this video's watch time.

Platforms report on their own schedule, so a post from today may still read nothing. A number nobody reported is shown as not reported, never as zero – those are different facts. Where watch time had to be worked out from a platform's own retention curve, the tab says so and names the platform.

Limits worth knowing

• Music is not available yet. No licensed library is wired up, so the music step is skipped and refunded if you ask for it.
• Voiceover is not available yet. The models speak the script themselves, which is why the model list says which ones can. A separate narration voice is a later addition.
• An actor batch opens on the first variant. The rest are made at the same time and sit in your video list, grouped together; there is no side-by-side comparison screen yet.
• An article has to be readable. A page that hides its text behind a login, a paywall or a script-rendered shell may come back empty or wrong. Paste the text instead when that happens.
• Chat and the autonomous team quote a rough price. A video asked for in chat shows an estimate for one short scene; the real charge is per scene and can be higher. The estimate on the Render button is the accurate one.