How to use

The default, and the only format that times itself to a song rather than to a target duration. Pick it when you have audio and want the cut to follow it; the Director plans scenes against the track's length. It renders 16:9.

Renders at 16:9.

A vertical ad built for a feed, at 9:16. Choose this over Short video when the point is to sell something specific rather than to be watched — the Director weights the brief toward a hook and a call to action.

Renders at 9:16.

A vertical clip at 9:16 with no sales framing. This is the one for organic content, cutdowns, and anything that has to earn attention before it asks for something.

Renders at 9:16.

1:1, for feeds that crop a vertical video badly. Reach for it when one asset has to work across several placements, since a square survives more crops than a tall one.

Renders at 1:1.

16:9 and the most produced-looking of the formats. This is also the format the UGC studio opens when you film a script with lip-sync, so it is where a spokesperson read belongs.

Renders at 16:9.

Stills, not video, at 1:1. Charged as images at two credits each rather than as a video scene, which makes it the cheapest way to explore a look before committing to a render.

Renders at 1:1.

Swaps a face into a still image. Needs two images: the picture you are changing and the face you are putting into it. Use it for a photo you already have; if the person needs to appear across several new scenes, make them a Character instead so they stay consistent.

Swap a face into an image.

Generates several camera views of one subject from a single image. The fastest way to get a product or character sheet out of one photo, and a good input to Create Character.

Generate multiple camera views of a subject.

Re-renders one image from a new camera angle. Narrower than Multi View and better when you know the exact shot you want rather than wanting a spread to choose from.

Re-render an image from a new camera angle.

Adds skin detail and realism to a face. Worth running on an AI-generated portrait before it goes into lip-sync, because the flatness that reads as fine in a still becomes obvious once the face moves.

Enhance skin detail and realism.

Outpaints an image beyond its original edges. The fix for a photo that is the wrong shape for a placement — expand it rather than crop, and you keep the whole subject.

Outpaint / expand an image's canvas.

Replaces an image's background from a prompt. Use it to put a product into a setting you do not have; keep the prompt about the environment, not the product, since the product is already in the frame.

Replace an image's background from a prompt.

A prompt-based edit of a video you already have. This is the general-purpose video tool — reach for a more specific one below when it matches, because a narrower tool gives a more predictable result.

Prompt-based edit of an existing video.

Applies a new visual style across a video while keeping its motion. Good for making footage from different sources look like one piece; describe the look, not the content.

Apply a new visual style to a video.

Relights a scene from a prompt. Use it when footage is correct but the lighting is wrong — a daylight shot that needs to be evening — instead of re-rendering the whole thing.

Relight a video scene with a prompt.

Adds AI visual effects to a video. The broadest of the video tools, so be specific about the effect and where in frame it belongs.

Apply AI visual effects to a video.

Drives a still subject with the motion from a reference video. This is how a character you have generated performs a movement you have footage of. You supply both: the image to animate and the video to copy.

Drive a subject image with a reference video's motion.

Swaps a face into a video rather than a still. Slower and more expensive than the image version, so get the face right on a still first.

Swap a face into a video.

Text to speech narration for a script you already have. For a spokesperson who appears on camera, use lip-sync in the Creation Studio instead — this produces audio only.

Text-to-speech narration.

Generates a music track from a prompt. Describe the mood and instrumentation rather than naming an artist. Useful when a stock bed is not quite right and licensing one is not worth it.

Generate a music track from a prompt.

Generates individual sound effects. Short, specific prompts work best — one sound per run, not a soundscape.

Generate sound effects.

Speaks your text in a voice cloned from a sample you provide. Use it to keep one narrator across a series. Only clone a voice you have permission to use.

Speak text in a cloned voice from a sample.

Transforms an existing recording into another voice, keeping the original performance. Choose this over Voice Clone when the delivery is already right and only the voice is wrong.

Transform a voice recording into another voice.

Generates a reusable environment you can set later scenes in. Doing this once, up front, is what stops a series of renders looking like unrelated locations.

Generate a reusable environment.

Creates a reusable, consistent character. The single highest-leverage tool in the bay for multi-scene work: a saved Character is what Character Lock and lip-sync hold steady across scenes.

Create a reusable, consistent character.

The cheapest format, and the only one with no narration — the cut and the music carry it. Pick it for a launch or a restock where the product is the whole message and nothing needs explaining.

Hype reveal for a launch. No narration, let the cut carry it.

Matched framing so the change reads in about a second. It only works when you have a genuine visual delta; if the improvement is a number rather than a picture, use Listicle Hook instead.

Matched framing so the change reads in one second.

Three reasons, escalating, built to survive the scroll. The safest performer when you are not sure what to make, because its structure does the retention work.

Three reasons, escalating. Built to survive the scroll.

Phone-camera realism that deliberately does not look produced. Choose it over Brand Anthem when trust matters more than polish — which, on a feed, is most of the time.

Phone-camera UGC that doesn't look produced.

A documentary-warm origin story in five beats. The format for a personal brand or a founder-led launch, where who you are is part of the offer.

Documentary-warm origin story in five beats.

The most expensive format, and the only widescreen one — sixty seconds across six beats, meant for your site and your deck rather than a feed. Budget for it deliberately.

Sixty seconds, six beats, widescreen. For the site and the deck.

A big cinematic swing: golden hour and an orchestra. High production feel, so it rewards a brand people already recognise and punishes one they do not.

Big cinematic swing. Golden hour and an orchestra.

Eight beats of vlog realism, the longest structure in the set. Good for recruiting, culture, and showing a service that is hard to photograph.

Eight beats of vlog realism. The longest format, same price.

Three to five nurture emails, each with a subject line, preview text and body. Brief it with the goal and the deadline — 'upgrade to paid by day fourteen' — because the sequence's pacing is built backwards from that.

A 3–5 email nurture sequence, each with a subject line, preview text and body.

A first-touch email plus two short follow-ups. Name the exact segment you are reaching, not your product: a cold email's quality comes almost entirely from how specific the recipient is.

A first-touch cold email plus two short follow-ups.

Headline, subhead, dateline body and boilerplate. Quotes come back as bracketed placeholders and you fill them in — it will not invent a spokesperson saying something.

Headline, subhead, dateline body and company boilerplate — quotes left as placeholders.

Title options, a hook, a sectioned script, a call to action and the video description. Give it the runtime you are targeting; an eight-minute script and a two-minute one are different shapes, not the same shape trimmed.

Title options, a hook, sectioned script, call to action and video description.

Ten opening lines, each labelled with the hook pattern it uses. The labels are the point — they let you test patterns rather than lines, which is what actually tells you something.

Ten scroll-stopping opening lines, each labelled with the hook pattern it uses.

The objections your buyers raise, each paired with a usable response. Feed it your price and your audience, since most real objections are about one or the other.

The objections your buyers raise, each paired with a response you can use.

An email asking a happy customer for a testimonial, plus a short text-message version. Say when in the relationship you are asking; a request two weeks after delivery reads differently from one two years later.

An email asking a happy customer for a testimonial, plus a short SMS/DM version.

Title, subtitle, promo text, description and keywords, already trimmed to the store's character limits — thirty, thirty, a hundred and seventy, and a hundred. Nothing to count by hand.

Title, subtitle, promo text, description and keywords — trimmed to the store's character limits.

A satisfied customer talking about a result. The default for a product with proof behind it, and the format that most needs its bracketed placeholders filled with something real.

First contact with the physical product. Only works for something that ships; for software, use Demo.

The state change, narrated. Pick it when the outcome is visible and the process is not worth explaining.

Showing the thing working. The right choice for software and anything where the value is in the using rather than the owning.

A balanced take that names a drawback. Counter-intuitively the highest-trust format, because an ad that concedes nothing is read as an ad.

A narrative the product sits inside rather than leads. The slowest to get to the point, and the best at holding someone who would have scrolled past a pitch.

Problem, Agitate, Solution. The hardest-hitting of the three and the easiest to overdo — good for a real, felt pain, grating for a nice-to-have.

Attention, Interest, Desire, Action. The most balanced structure, and the safest default when you do not know which will land.

Straight problem to solution, with no agitation step. Choose it for a professional audience or an expensive product, where pressure reads as a reason to distrust you.

One idea and one call to action. The right length for a single sharp hook, and too short for anything that needs a reason to believe.

The workhorse. Room for a hook, a benefit and a close, which is the shape most ads actually need.

Enough for a second benefit or an objection handled on camera. Use it when thirty seconds keeps forcing you to cut the part that persuades.

The longest read, for storytelling and demos. Only worth it when the extra time is doing work — a padded sixty performs worse than a tight fifteen.

Five scripts, each attacking a different angle — pain point, social proof, curiosity, cost of inaction. One credit per variant, and anything the model does not deliver is refunded automatically. The cheap way to stop guessing which angle works.

Ten variants for a wider spread, at the same per-variant price. Worth it when you are launching something new and have no prior on what resonates; overkill for an offer you already have data about.

Vertical, for TikTok, Reels and Shorts. Setting it here frames every shot in the ad, so you never restate the ratio in the brief.

Square, the safest single asset when one ad has to run in several placements without being cropped badly.

Widescreen, for YouTube, a site hero, or a deck. The wrong choice for a feed, where it arrives as a letterboxed strip.

Ava, a warm US female read. The default, and the most forgiving across categories — it suits a benefit-led script better than a hard-sell one.

Bella, a brighter US female read with more energy than Ava. Good for launches and younger audiences; too much lift for anything sober.

Adam, a deep US male read. Carries authority, which helps a technical or high-price claim and works against a playful one.

George, a calm UK male read. The most understated of the four, and the one to pick when you want the script trusted rather than sold.

Bright and human. The most broadly usable bed, and the least likely to fight a spoken narration.

Low energy and unobtrusive. Choose it when the narration is doing the work and the music only needs to stop the silence.

Big and rising. Powerful behind a visual montage and overwhelming behind a talking head — it will make a modest claim sound oversold.

Neutral and professional. The safe choice for B2B, and deliberately forgettable, which is the point.

Eighties-leaning and stylised. Strong personality, so it either matches your brand or actively works against it — there is no middle here.

The engine-unlock tier. This is the step that widens which engines you can reach on platform credits, and the entry point for Viral Shorts. Pick it if the limit you are hitting is what you can make rather than how much.

$29/month · 800 credits

Volume, plus the Editing Bay, plus bring-your-own-key access to the flagship engines. The tier most people should be on, and the only one that unlocks the nineteen post-production tools.

$79/month · 2,400 credits

Volume and the best per-credit rate. Ultra unlocks no engine that Pro does not already have, so the only reason to take it is the rate — which means it is worth it strictly above a certain monthly spend, and not before.

$199/month · 7,000 credits

Text and image posts, and part of the weekly autopilot. The strongest channel here for B2B, and the one whose posts the autopilot writes most naturally.

Text and image posts, on the autopilot. If your account manages more than one Page, use the Posting to dropdown to choose which — the selection is remembered per connection.

Text and image posts, on the autopilot. Also the channel the engagement agent works best on, because its mention data is the richest.

Text and image posts, on the autopilot, with the same Page picker as Facebook when your account manages several.

Video only, and deliberately not on the weekly autopilot. It is for pushing a finished Director film on demand, so it stays idle until you have rendered something.

Video only and on-demand, like YouTube. Note that posts publish privately until the app clears TikTok's content-posting review — that is TikTok's gate, not a setting you can change here.

Text and image posts, on the autopilot. The one connection that does not use an OAuth flow — it takes an app password instead.