Creative Studio: Reels, Shorts, carousels and images with AI

The Studio (menu Production → Studio) writes the script with AI for the chosen persona, following the workspace's editorial lines, and produces the piece: Reels/Shorts (9:16) with voice, word-by-word captions and music, a carousel (4:5) or an image (4:5 for the feed or 9:16 for Stories).

You create it in steps: the format, the video template (full-screen animation, split screen with a presenter or full-screen presenter), the editorial line and the persona, the channels and the content, with the product photos when the piece shows a product, and you review everything before generating. The speaker can be you, recording with your phone, or an AI avatar. And you decide who drives it: one click, a fixed sequence of steps, or the agent, where Claude drives the same tools, looks at the result and redoes what did not turn out well.

The three sub-tabs and where personas and lines liveCopy Anchor Link

Personas and editorial lines belong to the business's marketing, not just the Studio: since 10/25/2026 they live in the Marketing menu.

The Studio still uses both in Create. Old links to the Studio's Personas and Editorial lines sub-tabs open the new screen.

Create a piece in 6 stepsCopy Anchor Link

In the Create sub-tab, each step asks for a choice. Continue only unlocks when the step is complete and, meanwhile, says below what is missing; Back goes back one step. On mobile you see "Step N of 6" with a bar; on desktop, the list of steps, where the ones already visited open with one click.

Product photos. When the piece shows a product, click Add photo and paste the https link to the photo, preferably the product cover on your website, or tap From catalog to use the photo of a product from Marketing → Products (up to 4). The AI only describes the scene (person, place, light) and the piece shows the product exactly as in the photo, without making up shape, color or detail.

Story 9:16. The image comes out at 1080×1920, with a short title and the text away from the edges. The Studio's Publish posts to the feed; to post it as a Story, schedule it in Posts with the Story type and choose the piece in From Studio.

Reels as a Meta ad. When you select Meta Ad on a Reels, the Channels step reminds you of the best practices: vertical 9:16 and with sound (Reels play with sound by default), with the key messages inside the safe zone (the top 14% and bottom 35% free, where Meta's profile, caption and button sit). After the render, the piece goes straight to an ad set in Advertise on Meta, in the production. For images, Meta can expand them to 9:16 and add music (Advantage+), choices made when advertising; a carousel ad, for now, is built in Meta Ads Manager.

Coming from Create piece on an editorial line, Create opens with the line, the format and the template of its first format already selected.

Personas from real dataCopy Anchor Link

In Marketing → Audience → Personas, in the Generate from real data block, choose the period, how many personas you want (1 to 3) and, if you like, a focus (product, segment, region) and other data (surveys, CRM notes, reviews). When you click Analyze and suggest personas, XBASE reads the workspace's recent WhatsApp and Instagram conversations, only what customers wrote, without phone numbers, emails or links, and the AI suggests personas with pains, desires, objections, behavior, what makes them act and the words they use, with the evidence of where each one came from.

Comments on profiles and videos. In Profiles, posts or videos (optional), paste up to 5 links, one per line: an Instagram, YouTube or TikTok profile, post, Reels, Shorts, video or channel, yours or a competitor's or a creator's in your niche. The AI reads the public comments (up to 150 per link; on a profile, from the 6 most recent pieces of content) without the commenter's name. Uncheck Include WhatsApp and Instagram conversations to analyze only the comments. After the analysis, each link shows how many comments were read or why they could not be read.

Suggestions are not saved on their own: click Review and save, adjust and Save persona. You can also write a persona from scratch in New persona.

The persona goes into the script: choose it in Create → Persona or keep the editorial line's persona. The script speaks to them, about their pains, in their words.

Video templatesCopy Anchor Link

In the templates with a presenter, they say the CTA and the end card with the button goes full screen right after.

Who speaks: you or an avatarCopy Anchor Link

• Your recording: generate the script, open the production and record while reading the teleprompter in the Your recording panel: vertical video (MP4, MOV or WebM) in the templates where you appear; in Animation with your voice, audio only too (MP3, M4A, WAV, OGG or AAC). Up to 200 MB and 5 minutes, with sound. Click Produce.
• Automatic editing: your speech is transcribed with each word's timing (ElevenLabs Scribe or OpenAI Whisper), long pauses are removed, the clips are joined and sped up and the captions show what you said. Adjust it in Tools → Recording editing: cut pauses and from how many seconds, the breathing room at each cut, remove filler words ("um", "uh") and the speed-up (up to 1.5×).
• AI avatar: in Tools → Avatar, choose HeyGen, enter the avatar id and add the HeyGen key in Provider keys. The avatar speaks the script with the generated voice; nothing to record.

What happens in each productionCopy Anchor Link

1. Script: the AI writes the hook, scenes (narration and on-screen text), CTA, caption and hashtags for the persona and checks the line's and the workspace's rules.
2. Voice and captions: the narration is generated (or comes from your recording, already without the long pauses and sped up) and each word gets its timing, so the captions light up along with the speech.
3. Presenter (avatar templates): the avatar is generated from the voice.
4. Music (if on) and visuals: each scene's background with an AI image or video clip, or the brand background.
5. Render: the video comes out at 1080×1920 with normalized audio; carousels and images come out at 1080×1350 (images for Stories, at 1080×1920).
6. Review: duration, speech ending 3 seconds before the end, banned claims, CTA, presenter and the 12-frame board for you to look at before publishing.

Tools and keysCopy Anchor Link

In Tools → Provider keys, people who manage keys add the workspace key for each provider. The key is stored encrypted and never goes back to the screen. VoiceStudio needs a public https address.

PublishCopy Anchor Link

• Instagram: Reels, images and carousels now (goes out in about 1 minute) or on a date, up to 75 days ahead, through the same queue as the post scheduler. The same piece is not published twice on the same account.
• Advertise on Meta: the Reels or single image becomes an ad in an existing Meta ad set (see below).
• Meta ad library: the video goes to the connected ad account and you build the ad in Ads Manager.
• TikTok and YouTube: download the video with Download MP4 (TikTok, YouTube) and publish it there.

Reels, images and carousels go into the post scheduler queue and are recorded as scheduled in the Publish panel. The finished carousel also shows up in Publishing → From Studio. Sends to Meta also stay in the panel history. Every publication asks for confirmation. Only a logged-in person with permission to manage integrations can publish: agents through MCP create and produce, but do not publish.

Advertise on Meta: the Reels straight into the ad setCopy Anchor Link

With the production ready, the Advertise on Meta panel creates the ad in an ad set that already exists in your Meta account, without opening Ads Manager:

1. Choose the Ad set: the ad uses its audience and budget.
2. Check the Ad name, Primary text (comes from the caption; Meta shows about 125 characters before "see more") and Headline (optional) (comes from the CTA; up to 40 characters is recommended).
3. Destination: WhatsApp, with the Prefilled message (optional), only on a WhatsApp conversations ad set; or Website, with the https address and the Button.
4. Check the Advantage+ features you want. For video: adjust the aspect ratio to each placement, automatic cropping, highlight the button and text variations. For images: expand to 9:16 with AI (for Reels and Stories) and music from Meta's library, which come on for 4:5 images, plus the same adjustments. Whatever stays unchecked goes off.
5. With Activate now off, the ad is created paused; with it on, it goes into Meta's review and, once approved, spends from the ad set's budget. Every creation asks for confirmation and is recorded in Ads → Alerts.

The Ready for Reels block shows the Reels check for the Meta Ad channel: vertical 9:16, with sound and with the text in the safe zone (the top 14% and bottom 35% free, where the ad interface covers). When you select Meta Ad in Create, the Studio already moves the captions and text up to fit this zone. Carousels are still built in Ads Manager. Only a person who manages integrations creates the ad on screen; through MCP, the agent creates a draft that someone approves in Alerts.

Rules and limitsCopy Anchor Link

• Create, edit, upload the recording, save personas and produce: people who manage integrations. Other members see everything.
• Generate personas from conversations: only a logged-in person who manages integrations (agents through MCP do not read conversations here).
• Daily execution limit per workspace (each script and each production counts; default 20, adjustable in Tools up to the platform cap) to control cost.
• One production is rendered at a time; the others wait in the queue.
• Produce again reuses voice, presenter, music and visuals and only renders; Generate everything again generates everything again.
• Editing the script discards the previous voice, visuals and render; with unsaved edits, the produce buttons stay locked until you click Save script. If you already recorded and changed the order of the lines, record again: the scenes follow the order of the narration.
• The brand logo is uploaded in Tools → Brand → Upload logo (PNG, JPEG or WebP up to 2 MB).
• The cost shown is an estimate for models with a known price; voice, image, video and avatar are charged by the provider to the account of the key used.

Frequently asked questionsCopy Anchor Link

See alsoCopy Anchor Link