Director Mode

1. Step in.Open Director Mode from the Video Creator Canvas toolbar or the Director Mode banner in chat's video mode. The landing plays a slow loop of cinematic footage under a black mist — type or dictate your opening shot in the left column.

2. Watch Take 1 roll. Clips are 3–10 seconds at 720p with native audio, and usually land in about a minute. The stage keeps the room alive while it renders.

3. Direct the change. On set, the take fills the whole room and the composer stays right where it was on the landing — type the next note and the new take carries over the scene, characters, and physics from the take it edits, applying only your change.

Directing take by take is perfect for one shot. For something longer — a scene, a short, a five-minute piece — pick Story on the landing instead of One shot, and describe the whole thing at once. Theo asks a couple of short questions, casts the characters, and lays the film out shot by shot before anything renders.

The questions are cards, not a chat. Theo asks two to four things it genuinely can't guess — what it should look like, who it's about, how it ends — and each one is a row of tappable options with a free-text box if none of them fit. Answering the whole interview takes about as long as reading it. Then you pick a runtime from a short menu that tells you what each length actually buys ("8 shots — enough for a full arc"), sized against the clip length your engine really renders.

Pick your engine before you plan. When your workspace has the premium tiers enabled, the interview and the shot rail both carry the same Flash / Cinema / Cinema Ultra tabs the one-shot composer has, with the quality and length settings beside them. It matters more here than anywhere else, because the tier decides how long a single shot is— around eight seconds on Flash, fifteen on Cinema, and up to thirty on Cinema Ultra — and that's what the runtime menu divides your film into shots with. The same three minutes is twenty-odd short shots on Flash or a handful of long ones on Ultra: same runtime, a very different plan and a very different bill. Pick the tier first and the rest of the interview sizes itself around it.

The tier is part of the production. Your pick is saved with the story, so reopening it days later still rolls on the engine you planned for rather than quietly falling back to the default. You can change tiers after the plan lands — the controls stay on the rail — and Theo tells you what moved: a quality or length pick the new tier can't render is shifted to its nearest option, and if the new shot length no longer matches what the plan was budgeted for, it says that too. It never rewrites the plan. Your shot list, your scripts, and your cast stay exactly as they are — only the shots you roll from then on change. If you want a plan built around the new length, plan the story again.

Cast the people you already have. If your Character Studio already holds characters, the interview adds one more step: your library as a grid of portraits, tap the ones this film is about. Whoever you pick is written straight into the plan as a real part with that character already attached — so they're cast before a single render, there's nothing to wait for, and the face is the one you already approved. Skiphands the casting back to Theo, and with an empty library the step doesn't appear at all.

The rest of the cast is drawn before anything rolls. The moment the plan lands, Theo starts drawing every part nobody is cast in yet — a canonical portrait plus a full turnaround from the other angles, straight into your Character Studio. You don't press anything; it just starts, and anyone you cast yourself is skipped because they're already drawn. Those images anchor every shot the character appears in, which is what keeps a person recognisably the same person across a twenty-shot film.

Shots wait for the cast — on purpose. Generatestays disabled until every character is drawn. Rolling early doesn't fail, it does something worse: the engine invents each person from the written description, and invents them differentlyin every shot — which is the one problem planning a whole film is meant to solve. The rail tells you how many are left. If a character genuinely can't be drawn, you get a Roll anyway escape rather than a dead end.

Look before you spend. The Casttab on the right edge holds the whole company, and it opens by itself when you arrive at a planned story — the cast is both what's being drawn and what rolling is waiting on, so it's the first thing worth looking at. Each character is a portrait with the slot number the script refers to them by ("image 1") and how many angles have landed; tap one to see it big, flip through every angle, redraw it, or open the full Character Studio. Hide(or Escape) folds it back into the tab when you want the whole stage — and the room moves out of its way while it's open, so the play controls and the top bar are never underneath it. The shot strip and the script stay live behind it either way. Worth doing before you roll: eight renders is a real spend, and a wrong face is much cheaper to catch here.

Bring somebody in, or recast a part. Character Studio in the top bar opens your whole library, and inside a story Cast in this story asks where that person fits: an existing part — which recasts it on the spot — or a brand-new one named after them. It works the other way round too: tap a part in the cast sheet and Recast… swaps in someone else, and Cast someone at the end of the sheet adds a part from your library. All of it is instant and free, because nothing is being drawn — the part simply points at a character you already have, and the wait before you can roll clears the moment every part has somebody in it. Casting never touches the character: they stay in your library for the next film, and so does anyone they replaced.

Arrange who's who. Tap a part and you can rename it, mark it a lead or supporting, and move it up or down among its own group — the sheet then shows your leads first, so eight names stop reading as eight equal ones. It's for you rather than the engine: the order and the labels don't change a word of any script, so rearranging your company is always safe. Uncast empties a part without deleting the character, and Remove role takes the part out of the film — it tells you how many shots mention it first, because those scripts renumber around the gap.

Every shot has a full script. Pick a shot from the strip along the top and its complete direction opens in the left rail — the style register, the cast, the lighting, and a time-anchored beat script that choreographs the shot second by second ([0-3s] establish, [3-6s] react, and so on). It's the same comprehensive format a director would write by hand.

Adjust a shot by saying what you want different. When your workspace has it enabled, each shot gets its own small composer under the script. Watch the take, then tell Theo the one thing that's wrong — "hold on her face a beat longer", "make the ending warmer", "she should already be walking when we cut in" — and it revises that shot. Theo rewrites the structureof the shot rather than the whole script, so the film's look, your cast, their slot numbering and the shot's length all stay exactly as they are.

It tells you what it touched. Every revision reports back in one line — "Theo changed beat 3.", "Theo added a beat and changed the lighting."— so a small note can't quietly rewrite the whole shot behind your back. If it read the shot and decided it already did what you asked, it says that too rather than pretending to change something. Undo puts the shot back exactly as it was.

Adjusting never rolls anything. A revision only changes the plan — no credits, no render. Read the new beats, adjust again if you want, and when you like it hit Roll another take. Two decisions, so you're never charged for a take you didn't ask for. A few things a revision deliberately can't do — recasting, changing who's in frame, changing how long the shot is — and rather than half-ignore you it does the part it can and says in one line where the rest is done.

The raw script is one tap away. Scriptunder the composer opens the beat script as a text box — change the line that's wrong, hit Save, and your version is what rolls from then on; Revertbrings Theo's back. A hand-edited script wins over everything afterwards, which is exactly why an adjustment clears it: the composer tells you before it does, and Undo restores your edit along with the beats. While you have unsaved script changes the composer waits and says so, so a revision can never land on top of them. When you'd rather just watch, Hide folds the whole rail away and a Script pill appears beside the play controls to bring it back — each production remembers which way you left it.

Every take of a shot, side by side. Roll a shot more than once and its takes appear as thumbnails under the script. Tapping one SCREENS it in the room — that's a preview and nothing more. The star is the decision: it picks the take that shot contributes to the play-through, the download, and the final cut. Two separate actions on purpose, so comparing takes can't quietly rewrite the film. Tap a star again to unpin it and the shot goes back to using its newest finished take.

Another take is another attempt at the shot. Rolling a shot again re-renders it from its own script — it doesn't edit the take before it. That's deliberate: on the premium tiers an ordinary edit works by watching the previous take's footage and changing one thing on top of it, which is right when you're iterating on a single shot and wrong when you want a second go at a planned one. So every take of a shot is an independent attempt at the same direction, and fixing the line that's wrong in the script and rolling again gives you the shot you asked for rather than a variation of the one you didn't want.

The last card is the finish line. The shot strip ends with a Final cut card. It stays greyed out while shots are still to roll and tells you how many are left — the finish line is visible from the start rather than appearing once you get there. Once every shot has landed it opens the Final Cut builder, which joins your starred takes into one finished video with transitions, narration, and a score, then hands you a download. Worth knowing: the cut re-uses your takes as they are rather than re-rendering them, so it comes out at the quality you rolled the shots at. If you want a higher-resolution master, choose the bigger quality setting on the shots BEFORE you cut — that costs more credits per shot, which is why it isn't automatic.

Rolling the film. Generate rolls the shot you're reading; Generate all walks the rest of the plan in order — one shot at a time, start to finish: it dispatches a shot, waits for that shot to land, then moves to the next, all the way to the end of the plan. Shots render one at a time because that's a real limit, not a queue we're hiding, and both buttons show what they'll cost before you press them. A long run is a long wait, so keep the tab open: closing it pauses the run rather than losing it, and reopening the production picks up at the first shot that still needs rolling. If a shot fails, the run stops there instead of charging for everything after it.

It's an ordinary production underneath. Each shot is a scene, so everything else in Director Mode keeps working: re-roll a shot for another take, star the one you want, trim it in the timeline, play the film through end to end, download the scenes, or cut them together with Final Cut.

The third way in. Pick Lyric video on the landing and four guided steps take it from there: describe the video, pick imagery and characters, paste the lyrics, add your song. Theo then builds a full music video around the track: every line timed to the music, a visual shot planned for every stretch of the song, and a karaoke overlay when you watch it back.

Bring your own faces and imagery — or skip it. Step two is optional. Feature up to three saved characters from your library — Theo weaves them into the shots where they serve the song (never lip-syncing the vocal) and anchors each one on the exact reference you saved — and add up to three reference images to anchor the mood, palette, or setting. Skip the step and Theo designs the visuals from your description alone.

The song has to be yours. Upload the track (MP3, WAV, M4A, AAC, or OGG — thirty seconds to ten minutes) and confirm you hold the rights to use it. That affirmation is required before anything uploads: the video is built around the recording, so the recording has to be one you're allowed to use. A song made with Theo Symphony is always safe here.

The lyrics are timed against the real audio. Paste the words (section tags like [Chorus]help but aren't required) and Theo listens to the track and times every line — word by word when the audio allows it, evenly spread across the song when it doesn't, and the setup tells you which you got. The timing drives everything downstream: where shots start and end, and when each karaoke line lights up.

The tier decides the shot count — and the bill. The same Flash / Cinema / Cinema Ultra tabs as everywhere else, and here the trade is at its sharpest: the tier sets how long one clip runs, and the song's length divided by that is your shot count. A four-minute song is thirty-odd shots on Flash or about eight on Cinema Ultra — which is why songs past four minutes are steered to Ultra automatically (a recommendation, never a lock; your own pick always wins). The setup quotes the real shot count and credit cost before you plan anything.

One look, planned across the whole song. Describe the look you want — the world, the palette, the mood — and Theo plans a shot for every window of the timeline, each carrying the same visual register so the video reads as one piece. Repeated choruses are detected from the lyrics themselves and reuse their visuals: roll the first chorus and every repeat replays it for free, the way real music videos return to the hook.

Shots roll like Story Mode. Every shot's direction is editable before it rolls. Roll one at a time, or Build the video to walk the plan in song order — one shot at a time, stopping if one fails rather than charging for everything after it. Keep the tab open during a run; closing it pauses the build and reopening picks up where it stopped.

Watch it, then take it with you. Once every shot has footage, Watch the videoscreens the whole thing in the room — every clip cut on the song's own timeline, the real track playing underneath, and the lyrics as a live karaoke overlay. Export the MP4 then renders it as one downloadable file with the song and the karaoke lines burned in. The export is free — the shots already paid for the footage, and the song is your own.

Make the words look like yours. Tap Design the lyricsand the design studio takes over the screen, built to fit it: the live stage always fits your window, the style gallery sits in a collapsible rail on the left, and every control lives in grouped panels on the right, so nothing hides below the fold. The stage plays your own words (sweep, entrance, effects and all) in any of the lyric styles, grouped into house looks, statement looks, and elegant looks, so you pick one by watching it instead of reading a swatch. Preview over your own footage once a shot has rolled, over a still Theo paints from your concept, or over a ready-made backdrop. Theo picks a signature look per video so two productions don't come out matching, and your own pick always wins. What you see on the stage is exactly what burns into the MP4 — and Classic is always there when you want the original lower-third look.

Dress the text itself. Beyond picking a look, the inspector tunes the type: a roster of faces (each shown set in itself), a drop shadow, an outline, a glow, or a backing plate behind every line, the sweep tint each word lights up in as it's sung, an all-caps toggle, plus size, alignment, tilt, and scale. Every change applies live on the stage, and returning a control to the preset's own default hands the look back untouched.

Open with a title, close with the credits. The studio's Title & creditstab adds a title card over the intro — the song's name (prefilled from your upload), the artist, and an extra line if you want one — and a credits card over the final seconds for writers, producers, copyright, whatever the song should carry. The card previews on the same stage as you type, each one toggles on and off without losing what you typed, and both show up in the watch-back and the export alike.

Mix it like a record. Open the Audio Studio on a lyric video and the song rides its own Songfader: pull it up or down, let a scene's own audio through where you want it, drop in extra sounds, and ride the master. The watch-back previews your mix live, and the exported MP4 is mixed the same way — the preview and the render never disagree.

Availability may depend on your deployment — the Lyric video tab appears on the Director Mode landing when it's enabled for your workspace, and the design tools (lyric styles, title & credits cards, the mixer) may be enabled separately.

Edits work best as one focused change per take: "Change the lighting to golden hour." "Make the umbrella red." "Slow the camera push." Ending with "Keep everything else the same" pins the rest of the scene in place.

For a fresh generation, describe the subject, action, setting, camera move, and mood. Want one continuous shot? Say "single continuous shot, no cuts."

Direct the sound too. Takes render with native audio, so describe it: "Audio: rain on glass, distant thunder, no music." Say "no dialogue" when you want a silent performance. For text inside the shot, put it in quotes: "a neon sign that reads ‘OPEN LATE’."

Time the beats. Timestamp ranges choreograph the clip: "[0-3s] establish the scene. [3-6s] the key action. [6-8s] hold on the final frame." The sparkle in the composer holds these pro tips — tap one to drop it straight into your note. And when you'd rather not structure it yourself, tap the wand: Theo rewrites your note the way the engine reads best, with a one-tap Undo and nothing sent until you roll it.

Type @ to bring something in. When your workspace has Elements enabled, typing @mid-direction opens one picker over the composer holding everything you can anchor a take on: your saved characters, this scene's finished takes, and an upload. Keep typing to narrow it, arrow keys to move, Enter to pick. Choosing a character leaves their name in your sentence— "@sar" becomes "Sarah walks to the door" — which is exactly what the engine reads best. Choosing a take grabs its final frameand stages it as a reference, which is the fastest way to say "carry on from how that one ended." Everything lands as an ordinary reference card, so roles and the per-take limit work exactly as they always have.

Take settings.The chip beside the composer's controls holds the knobs for the take you're about to roll: the frame, quality, length, and — where the engine supports it — encode quality, which trades a bigger file for a cleaner picture on detailed or fast- moving shots. Each menu only offers what the engine you're on genuinely renders, and switching engines moves a pick onto the new menu and says so rather than silently downgrading it.

Need more room? When a note grows past a few lines, an Expand button appears in the composer — it opens the writing room, a full-size page for detailed scripts and beat sheets. Drop a .md or .txt file to import it straight into the note, keep attaching images and characters from the footer, and press ⌘/Ctrl+Enter to roll the take. Closing the room keeps your draft.

"Moody lighting" means something different to everyone — including the engine. When your workspace has the Lighting Studio enabled, a lamp buttonsits in the composer's controls row. It opens a visual gallery: every lighting style is shown as a real photograph of the same stand-in subject lit that way, so you choose a look by seeing it — not by guessing what its name means.

Four decisions, one at a time. The studio is organised as collapsible sections — lighting style, colour & mood, where the scene lives, and fine-tune — and each header shows what's currently picked, so you open only the decision you're making and a collapsed studio still reads as your complete recipe.

Start from a style. The style section holds the setups a cinematographer reaches for by name — Rembrandt, split, butterfly, loop, noir, silhouette, high key, golden hour, overcast, hard midday, window light, neon practical, firelight, moonlight, stage spotlight, and classic three-point — grouped as Portrait, Dramatic, Natural, and Stylized. Tap a tile and it fills the big preview up top; drag the preview (or tap the orbit dots) to walk all the way around the subject — front, three-quarters, either profile, and directly behind. The orbit wraps, so you can keep circling. The other sections pick the colour & mood (noir, golden hour, teal & orange, bleach bypass, …) and where the scene lives (studio, interior, exterior day or night, stage, vehicle).

Your picks blend together. Pair a style with a colour or a scene and the big preview doesn't stay a preset photo — it blends everything you picked into one new photograph of the same stand-in, so "Rembrandt · teal & orange · night exterior" is something you can seebefore you commit. The caption names the blend while it's being prepared, and if a blend isn't available the preview quietly keeps the base style photo.

Fine-tune the lights. Open Fine-tune for a top-down set diagram — your subject in the middle, the camera at the bottom, and each light as a dot you drag around them. Every light has a job — key, fill, rim, background, or practical — and plain-language controls: Height (below, eye level, high, overhead), shadow edge (hard to soft), distance (near to far), brightness, a warmth slider from candlelight to sky-blue, and a row of colour gels. The studio keeps a running contrast ratio in view — the number that decides whether a shot reads as flat, natural, dramatic, or hard-contrast. Once you fine-tune, the preview marks itself Adjusted, because the photo now shows your starting style rather than your exact rig.

The photos show the style, not your scene. Previews are photographs of a stand-in subject so every style is instantly comparable. They are not renders of your take — what the engine receives is the precise lighting direction the studio writes, and the collapsible "The direction this writes" panel shows you those exact words before you commit.

Lock it onto the take. Use this lightingturns the design into precise cinematography language — it leads with the style's own name and defining look (a Rembrandt take is told to carve the triangle of light, not just where the lamps sit), spells out direction, height, quality, ratio, colour, and environment, and tells the engine this light wins over whatever the scene itself would imply. It rides alongside your direction note. A chip above the composer shows what's applied ("Rembrandt · 3 lights"); tap it to reopen the studio, or ✕ to drop it.

It sticks. The rig rides every take you roll until you change or clear it, so a whole scene can be lit consistently. Retrying a take re-uses the lighting that take was shot with, a reopened production picks up the rig from your most recent lit take, and editing an existing take with a new rig reads as a relight: change the light, keep the shot. Perform takes are the exception — they render at the framing of the photo or footage you attached, so the lighting design doesn't apply there.

When your workspace has the premium tiers enabled, tabs rise off the composer's top-left edge. Flash is the fast, conversational default: takes land quickest, and an edit re-renders the same shot with just your change — built for rapid back-and-forth iteration, and the lightest on credits.

Cinemais the premium tier: sharper fidelity, richer native audio, and longer shots (up to ~15 seconds) with multi-shot pacing. Cinema edits work differently — instead of re-rendering the same shot from memory, the engine watches the prior take's actual footage and applies your change on top of it, preserving the motion, camera work, characters, and setting. That makes it the strongest choice when characters or reference images anchor the scene. Cinema takes render more slowly and cost about 4× the credits of a Flash take.

Cinema Ultra is the same pipeline tuned for length. One continuous take can run up to 30 seconds instead of 15, it accepts far more reference images in a single shot, and it can edit anytake you've made — including the high-quality ones Cinema can't re-open (more on that below). It is not simply "more Cinema": the trade is the top of the quality menu, because Ultra finishes at 720p and doesn't offer 1080p or 4K. It is the slowest tier and the most credits — a full 30-second take runs roughly 3× a standard Cinema take, while the default 10-second pick lands much closer to one. Reach for it when the shot needs to breathe; stay on Cinema when you want maximum resolution.

Set the length & quality. With either Cinema tier selected, a settings button appears in the composer's controls row showing your current pick (for example 720p · Auto). Open it to choose the quality and the length. The menus are per tier, so you only ever see picks that tier can actually render: Cinema offers 480p, 720p, 1080p, or 4K and Auto or a fixed 4–15 seconds; Cinema Ultraoffers 480p or 720p and Auto or a fixed 4–30 seconds. Higher resolutions and longer clips look better but render more slowly and cost more credits (on Cinema, 1080p is about 2× a 720p take and 4K is the most). Your pick rides each take, so a reopened production and a retried take both remember it — and if you switch tiers with a pick the new one can't do, Theo moves it to the nearest option and tells you what changed rather than silently re-rendering something else.

One thing worth knowing about quality. A Cinema edit works by watching the prior take's actual footage, and Cinema only accepts footage up to 720p as that source. A shot you rendered at 1080p or 4K is perfect to screen, export, and cut into a final cut, but it can't be handed straight back to Cinema for the next edit. Theo catches that before you spend anything and offers the ways through right there. The cleanest is Cinema Ultra, which has no such limit — it edits your take at full quality, exactly as you shot it. Otherwise, make a 720p copy of the shot and edit that (your original take stays exactly where it is), or edit the take on Flash. Worth knowing because it isn't obvious — the quality setting in the composer picks the resolution of the nexttake, not the one you're editing, so turning it down there won't unlock the edit. If you already expect to keep iterating on a shot in Cinema, staying at 720p keeps that loop seamless.

Mix freely. The pick is per take: switch engines mid-production any time, and any engine can edit takes rendered by the others. Takes made on a premium tier carry a small Cinema or Cinema Ultra chip in the thread, a reopened production defaults to the engine you used last, and the ⓘ tab beside the tabs holds this same guide right on set.

Already have the audio? Attach it with the waveform button in the composer and the take is driven by that track instead of being described into existence. Perform reads what you attached alongside it and picks one of two shapes for you — there is no extra mode to choose.

A photo speaks it. Attach an audio track plus a photo of a person (or stage a saved character) and that person performs the audio — lip sync, expression, and head movement, in their own likeness. Your direction note isn't the script here: the audio is. Use the note for staging instead — "slow push in, warm key light" — or leave it empty and just roll.

Or dub a take. Hover any finished take in the thread and tap the waveformicon — "Dub this take" arms that footage, switches you to Perform, and opens the audio picker in one tap. The new take is that same footage with your audio taking over the speech: same shot, same performance, new words. It works on generated takes and on your own uploaded clips alike, and the original take is never touched.

Formats and length. MP3, WAV, M4A, AAC, or OGG, up to 25 MB and 60 seconds. The composer plays the track back before you spend a take and shows its length, because that's what you're billed on. Tracks over 30 seconds render at 720p rather than 1080p — the whole performance is worth more than the extra pixels.

Framing.A Perform take keeps the shape of whatever you attached — your photo, or the take you're dubbing — so the 16:9 / 9:16 pick doesn't apply to it. Everything downstream still works normally: Perform takes star for the playthrough, export with the rest, and can be edited afterward on Flash or Cinema like any other take.

Credits. Perform is billed by the second of audio rather than per take, and dubbing existing footage costs considerably less than animating a still. A short line is cheap; a full minute is not. The credit breakdown lists each Perform take with its length so the number is never a mystery.

A quiet thread of takes lives right under the composer — newest first, older takes fading away as the story grows. Click any take to screen it, download it, or unmute its audio. Nothing is overwritten: a bad note never costs you a good take.

Preview the full frame. Everything happens in the Immersive room — the take plays edge-to-edge across the whole canvas. Tap Preview (top-right, on desktop) to screen the selected take uncropped at full quality on an elegant centered screen — ringed by mood lighting sampled live from the footage itself, like a bias light behind a television. Esc drops you straight back on set.

Edit from any take. Hover a finished take and hit the pencil to make it the base for your next direction — perfect for branching ("actually, go back to Take 2 and try it at night"). Each scene holds up to 30 takes; your five most recent productions sit on the landing, and See all opens the full library in a drawer.

Or keep it rolling. An edit CHANGES a shot; sometimes you just want more of it. Open a finished Cinema take's details and Continue this shot rolls the next take as a continuation instead — same subject, same setting, the action carrying on past where the clip ended. It arms the take, seeds a neutral note you can replace with what should happen next, and switches you to the engine that shot it. Continued takes are marked in the thread, and the original is untouched as always. You can continue a continuation a few times over; past that the studio says so rather than letting the shot drift further from where it started with every pass — start a new scene from the last frame instead.

Name your production. The title starts as your first direction — double-click it (or tap the pencil) in the top bar to rename it. The new name follows the production everywhere: on set, on the landing shelf, and in the library drawer.

Clean up the shelf. Hover a production card and tap the trash twice to delete it (the second tap confirms), or use the trash pill on set to delete the production you're in. Deleting removes every take with it.

The scene strip above the composer turns one shot into a full production. Tap + Sceneto open the next scene (up to 200 per production), click a chip to focus it, double-click to rename it ("Opening shot", "The reveal"), and use the arrows on the active chip to reorder. Each scene keeps its own takes — up to 30 — so you can iterate every moment independently.

Continue into the next scene. Tap + Sceneand the new scene picks up from the current scene's chosen take: Theo grabs that take's final frame as the visual starting point and carries the same characters, setting, lighting, and style into the next moment — then layers your new direction on top. A "Continuing from…"chip sits above the composer (it shows a brief "capturing…" while it grabs the frame). Open the chip to pick a different scene to follow — say, jump Scene 3 back to Scene 1 — or clear it with ✕ to start the scene fresh.

Let Theo write the next scene. Don't know what happens next? Leave the note empty and tap the wand. Theo reads the story so far — every scene in order, and the exact frame the last one ended on — then writes the next shot for you: a new beat that follows on, with a camera move and a sound cue, ready to roll. It never sends on its own: the direction lands in the composer as a draft, so you can edit it, tap Suggest another for a different idea, or Undo suggestionand write your own. Once there's text in the note, the same wand goes back to being Rewrite my direction.

Star the keeper. In the take thread, tap the star on the take you want that scene to contribute. Unstarred scenes automatically use their newest finished take, so a playthrough is always ready.

Play it through. With two or more playable scenes, Playscreens the whole production as one continuous sequence — each scene's chosen take, dissolving smoothly from one to the next, with a scene rail to jump around.

Take it with you. Downloadexports every scene's chosen take, numbered in order ("01 — Opening shot.mp4"…) — up to 30 scenes arrive as a single zip; bigger productions download file by file. Drop them straight into any editor to finish your ad, trailer, or film.

When your workspace has the timeline enabled, the numbered scene chips become small thumbnails— same place, bottom-left under your takes, but now you can see what each scene actually is. The rail scrolls, and it follows whichever scene you're working on, so your most recent moments are always the ones in front of you.

Open the timeline. The Timelinebutton beside the rail slides the full editing bay up from the bottom of the screen: every scene as a card, laid out in reading order, so a forty-scene production is one glance instead of a long scroll. Escape (or tapping outside) drops you straight back on set — it's a place you visit, not a bar that lives on top of your footage.

Drag to re-order. Inside the timeline, pick a card up and drop it where it belongs — the new order is the order your production plays and exports in, and it saves instantly. The ◀ ▶ buttons on each card do the same thing one step at a time if you'd rather not drag.

Cut a scene.In the timeline, open a card's Cutaction to trim, crop, and time that scene without re-rendering it. Drag the two handles on the strip under the preview to set where the scene starts and ends; drag the corners of the box on the frame to re-frame the shot; and set a playback speed (0.5× to 2×) or mute the take's own sound. Everything previews live, and Reset puts the scene back the way it was.

Nothing is destroyed. A cut is a note on the scene, not a new render — it costs no credits, your original take is untouched, and you can change or clear it any time. Cards show a small scissors badge when a scene carries one, and the cut applies wherever the production plays as a film: the playthrough and your Final Cut. Because a trim is measured against one specific take, starring a different take in that scene clears the cut rather than applying it to footage it was never made for.

One thing to know about downloads. The per-scene Download always gives you the full, untouched clip. Your trims and crops are applied when the studio assembles the film — so use Final Cut when you want a single file that reflects every cut you made.

When your workspace has the Audio Studio enabled, an Audio button sits in the top bar of every production. It opens a real scoring desk: four faders, a timeline of your scenes you can write music, narration and sound effects straight onto, and a player that hears the mix against the picture. This is the final audio — not a setting on one render. Whatever you balance here is what your playthrough sounds like and what every Final Cut ships with.

It opens on the film you already have. The desk starts at the balance the studio was already using, so the first thing you hear is your production as it stands today — not everything flattened to zero. From there you're adjusting something that already works rather than building a mix from silence.

Four faders. Scenesis your takes' own sound — dialogue and ambience. Score is music, Narration is the voiceover, and Effects is anything you place yourself. Each one has a level in decibels, a mute, and a solo; a Master fader sits under all four. Double-click any fader to snap it back to unity.

Solo is how you listen, not what you ship. Soloing a bus is the fastest way to hear whether the score is fighting the dialogue — and it is deliberately never saved with the mix. Leave something soloed, close the studio, and your film still plays in full. Mute, on the other hand, is a decision, so it is kept.

Write music, narration and effects here — or bring your own. Add audio at the playhead covers all four lanes. For score, pick a mood and describe it the way you'd brief a composer ("low strings under a slow reveal, no percussion"). For narration, type the line, choose a narrator, and add a note on the delivery where that voice can act on one — no need to open Final Cut to record a voiceover any more. For effects, open the sound library and describe the sound you want. And on any lane you can upload your own MP3 or WAV instead. Everything lands as a cue on the lane you chose, right where the playhead was.

Replace a lane in one move. If a lane already has something on it, you'll be offered replace alongside add— so swapping Theo's score for your own track is one action rather than a delete followed by an upload you have to line back up.

The sound library keeps what you liked. Describing an effect gives you several takes at once. Play them, keep the one that works, and it goes on a shelf you can drop onto any scene as many times as you need — the same door slam in three places costs one render, not three. Removing a sound from the shelf never touches the cues you already placed from it.

A cue belongs to a scene, not to a timecode. Every cue is pinned to one scene plus an offset into it — so when you re-order your scenes, the sound moves with the picture instead of staying behind at second 42. If a cue's scene later leaves the cut, the studio says so rather than silently dropping it.

Move it, trim it, see inside it. Every cue draws its own waveform, so you can see where the hit lands before you place it. Drag the block to move it; drag either edge to trim the head or tail. Blocks snap to the playhead, to scene boundaries and to each other. Select a cue and you also get its own strip: level, a fade in and a fade out, loop it to fill a longer stretch, nudge its offset, move it to a different scene, or remove it. Select a sceneinstead and you get that scene's own sound — its level relative to the Scenes fader, or muted outright.

Zoom in, and drag the playhead where you want it. Pinch on a trackpad — or hold ⌘ (Ctrl on Windows) and scroll — to zoom in around whatever your pointer is over, right down to a fraction of a second on a long production. The small − and +buttons above the lanes do the same in steps, and the expand button beside them puts the whole film back on screen. Once you're zoomed in, hold Shift and scroll (or swipe sideways) to pan along the timeline. To move the playhead, press anywhere on the ruler or a lane and drag — it follows your finger and keeps playing from wherever you let go, so you can find the exact frame a sound should land on without clicking around for it.

Hear one sound on its own. Any cue — or any whole lane — can be auditioned without moving the film's playhead, so checking what you just made doesn't mean scrubbing the production to find it. Like solo, it's purely how you listen and is never saved.

Duck the music under the voice. Turn on Duck under narration and the score pulls down wherever the voiceover is actually speaking, then comes back up in the gaps. Set how far it drops, and tick Effects too if you want them to make room as well. There are no keyframes to place — it follows the narration you already have.

Hear one scene, or the whole thing. The transport switches between Whole film and This scene, so you can loop the join you're worried about without sitting through the film every time. This scenetakes whichever scene the playhead is sitting in when you press it, and then names it — press it again somewhere else to move on. Either way you're hearing the real mix over the real footage.

What you hear is what renders. Every fader is an absolute level on its own source, and the studio hands the exact same numbers to the render — so the mix you approve is the mix that comes back. Two honest caveats: monitoring cuts hard between scenes where the finished film will dissolve, so a join sounds sharper here than it will; and the timeline lays your scenes end to end, which makes the total slightly longer than the assembled cut, because every dissolve overlaps two scenes. If your browser can't give the studio a full mixing graph you'll see a Simple monitoring badge — playback still works, and the render is unaffected.

Nothing is saved until you say so. Experiment freely: Undo (or ⌘Z) steps back through your changes, and closing with unsaved work asks first. Save mixis what commits it to the production. Three things in the studio cost credits, and they're the three that actually make new audio — composing a score cue, rendering a sound effect, and recording a narration. Placing a track you uploaded, moving a cue, and every fader, fade, duck, audition and preview are free. A render that fails costs nothing.

How it meets Final Cut. Anything you put on a lane here replaceswhat the final cut would otherwise have used for it — write a narration in the studio and Final Cut won't also lay its own voiceover underneath; score it here and its score steps aside. Final Cut's Sound step still decides what goes in for anything you haven't taken over, and the balanceis yours once you've mixed: the studio tells you so on that step, and your levels are what the render uses instead of the automatic ones. A production you never mixed keeps behaving exactly as it always has.

One level the studio may move for you. A film that already had a Final Cut score opens with its Score fader tucked well down, because that's where the automatic balance had it sitting under the picture. The moment your own score takes that lane over, that number is describing a track the film no longer uses — so the studio brings the fader back up to unity and tells you it did. If you'd already set that level by hand, it leaves it exactly where you put it.

When your workspace has Final Cut enabled, a clapperboard button sits beside the scene strip on any production with a playable scene. It opens the edit bay — five steps (Scenes, Look, Sound, Script, Screen) that turn your takes into one cohesive, professionally edited video you can download as a single file. Click-first throughout: you can build the whole cut without typing a word.

Your footage is always on screen. Step one is a bay of live tiles — every scene playing, not a checkbox with a label, so you can see exactly what's going in. Tiles play the scene as cut, so a trim you set earlier is what you watch here. Once you move on, the scenes shrink into a film strip that rides along the top of every remaining step — choosing a tone or a narrator never means losing sight of the footage it applies to. (A few tiles play at a time, nearest first, and hovering one always wakes it. Nothing autoplays if you've asked your system for reduced motion.)

Order, include, trim — in one place. Drag a tile to re-order the cut, toggle a scene out (it drops to a shelf below, one tap from coming back), and open Trimright on the tile to set that scene's in and out points, framing, speed, or mute without leaving the bay.

Direct each transition. Between every pair of tiles sits the join itself. Leave it and it follows the cut's overall style; set it and that one boundary does exactly what you said — a hard cut, a dissolve, a fade through black, or a wipe. Pinned joins are marked, so you can always tell what you chose from what you inherited, and clearing one hands it back to the deck-wide style.

Say what the cut is for. On Look, pick the kind of video — an ad, a story, a social reel, a trailer, a product demo, or a documentary — and the tone. If none of the moods fit, choose Customand write the tone in your own words ("dry, deadpan, a little tired"); Theo quotes you rather than paraphrasing. Above it all sits the goal— one line on what this video needs to do ("convince a skeptical CFO this pays for itself"). It's the single strongest piece of direction you can give: every line of the script is written to serve it.

Cast the narrator. Sound opens a cast of voices grouped by character — documentary, authoritative, warm, energetic, intimate — and each one auditions: tap it and hear a line before you commit. Some narrators also take delivery notes ("slower, almost whispered"), and the ones that can are the only ones that offer the field — you're never typing direction into a voice that can't act on it.

Score it your way. Pick a mood for the music, or write your own: the preset you were on seeds an editable brief, so composing a custom score is an edit rather than a blank page. Describe it the way you'd brief a composer — "sparse piano, no drums, builds in the last third" — and Theo scores the cut to it.

Let Theo write the voiceover. With narration on, Theo watches your actual takes and writes a script paced to the runtime, shaped to your video type and tone, and aimed at the goal you set. Approve it as-is, edit any line yourself, or type a quick note ("warmer, less salesy") and let Theo refine it — with a one-tap Undo. You can also paste a reference script or upload a reference video to steer the writing.

Screen it before you commit. The last step prepares the real narration and score, then plays the whole edit right there — your scenes dissolving in order with the actual voiceover and music layered on top. What you hear in the preview is exactly what ships in the render. Change a narrator, a delivery note, or the score brief and the preview knows it's stale, so you always screen the cut you're about to make.

One file out. Create final cut sends the edit to the studio, where transitions, audio ducking, and the final mix are mastered into a single MP4. Rendering continues in the background (one final cut renders at a time per production) — when it lands, play it on set and hit Download. Your takes and scenes are never touched: a final cut is a new export, so you can re-cut the same production as many ways as you like.

Every cut, on one shelf. Once a production has rendered a cut, a Final cutsbutton appears in the top bar beside your credits. It opens the shelf: every version you've made, newest first, each with its own thumbnail, runtime, and the recipe you picked ("Ad / Promo · Cinematic · 3 scenes"). Hover a card to preview it, tap the arrow to download it on the spot, or open it to watch full-size and remove it. The clapperboard beside your scenes still MAKES a cut; the top-bar button is where you FIND them.

Keep what matters. A production holds its eight most recent cuts — once you go past that, the oldest finished cut makes room for the new one. Download anything you want to keep for good.

Found a character you love in a take? Tap the person icon on the playing take, drag a box around them on the frozen frame, give them a name, and they're saved to your character library — the exact pixels, not a description.

Bring them back. Character Studio sits in the top bar beside your credits — on the Director Mode home screen and inside any production — and opens your full library in one tap. (The people icon in the composer still previews your three most recent characters, See all opens the same library, and the Characters tab on the right edge is still there.) From the library you can rename, delete, use a character, or open their studio: Useon the home screen starts a fresh story, and inside a production it stages them for the next take. Staged characters count toward the composer's reference-image limit. In a Story production there is no composer to stage into, so the button reads Cast in this storyand asks which part they're playing instead.

Everything they carry comes with them. A staged character is not just their photo. Their Character Studio angles, their supporting photos, their voice, their identity notes and their character-bible rules all ride along to the render — one staged card, the whole kit — while still using a single image slot. The card itself tells you what's going with them ("7 angles · 3 photos · voice · 4 rules"), so you can see at a glance whether a character is ready or still needs their sheet rendered.

The full turnaround rides their first appearance. Every angle is sent on the take that INTRODUCES a character to a scene, so the engine gets the complete look of them up front; after that they anchor on their main image, because a scene that already knows who they are should be about the shot, not the character sheet. That decision is made per character: bring somebody new into a scene you've already shot three takes of and they still get their full introduction, while the faces already in it carry on as they were.

No sheet yet? Your photos stand in. A character you built from photographs sends those supporting shots instead, so a real person is recognisable from the very first take — you don't have to render a turnaround before using them. Rendered angles take over once you have them, since they're on-style and deliberate.

The cast, on set. The Casttab on the right edge of any production lists everyone who has appeared in it — how many takes they're in, with quick actions to stage them again or add someone new from your library. Remove from scene rolls a fresh take of the current scene without them: the character is erased and everything else stays the same, and nothing already shot is lost.

Start from photos of a real person. When your workspace has it enabled, New character in the characters drawer (or the + tile on the cast strip) opens an uploader: drop in up to eight photos of the same person and Theo builds the character from all of them. The first photo is the primary— it's the face shown everywhere and the one every render is checked against — and you can crop it to the head and shoulders right there. The rest are supporting references. Before this, the only way a character could exist was to freeze a take and crop someone out of it, so there was no way to bring a real person in at all.

What makes a good photo set. Variety is what holds a likeness: a straight-on face shot, a three-quarter angle, a profile, at least one full-height shot, and a couple of different expressions. Keep them recent and lit consistently, and skip anything where sunglasses, a hand, or heavy shadow hides the face — the engine can only hold what it can see. One photo works; a good set works much better.

Every character you save gets a cinematic hero asset rendered automatically: your exact character re-lit on a dark studio backdrop, like a movie poster. To step onto their own set, hit Character Studio in the top bar of any production and tap Open studio on their card — or take the shortcut on the save toast. The character fills the room edge-to-edge, with your tools docked along the bottom, the same immersive feel as directing takes. Your original capture stays untouched: it's always the first tile on the sheet and the face of the character everywhere; generated assets never replace it.

The character sheet. Render the hero plus the full continuity turnaround — portrait, full body, back, strict side profile, and the three-quarter angles either side plus the over-the-shoulder three-quarter back — every view locked to your original character so you (and the engine) can hold them from any camera position. Every render is explicit: tap an empty tile to generate it, hover a filled one to regenerate or delete it, and tap any asset to put it on the stage. If a render drifts too far from your character, the studio discards it instead of saving a stranger; run it again.

Render every angle. One button fills the whole sheet in a single pass, rendering the missing views one after another with live progress and a count of what's left. It tells you what it will cost before it starts — a full sheet is several renders — and a view that fails is reported rather than silently skipped, so you can run just that one again. Already-rendered views are left alone.

Light them. The Add lighting button in the top bar — it reads Change lighting once a rig is set — or the lamp in the composer opens the same Lighting Studio you use for takes — the visual gallery where every style shows itself as a real photograph. Applying a rig here replacesthe studio's default portrait lighting rather than stacking on top of it, so a noir key or a golden-hour wash genuinely governs the render. It applies to whatever you generate next — a hero, a turnaround view, or a variant — and clearing the chip returns to the default look.

Give them a voice. The voice chip in the top bar (or the Voices tab on the right edge) opens the audition drawer. Browse by Male / Femaleand by style — Commanding, Calm & steady, Warm, Smooth, Energetic, Dark & mysterious. Tap any row to hear a short in-character audition, then tap Use to give it to your character; their current voice stays pinned at the top with a one-tap Remove. The voice rides along on every take the character is staged in, guiding how they sound in the rendered audio.

Identity notes. Tap the note chip in the top bar to keep a line or two about the character — wardrobe, personality, era ("always in a red leather jacket, dry humor"). Notes fold into every take they appear in, so those details genuinely shape renders across productions.

Create variants. Type (or dictate) a change into the composer at the bottom, like "winter coat", "battle-worn", or "gala tuxedo", or attach an image of the look you want them to take on; their identity stays locked while the styling follows your reference. The preview shows the original and the variant side by side so drift is obvious at a glance — then Save as new character. Variants join the cast strip beside the base character, each independently stageable in your takes.

Keep it tidy. Hover any filled sheet tile to download the asset, and use the trash pill in the top bar to delete a variant (the base stays) or retire the whole character — the second tap confirms.

Use in Project. One tap opens a centered picker of your productions — living thumbnails that play each one's latest take — plus a Start a new production tile. Pick a destination and the character arrives already staged in the composer, their full character sheet anchoring every take.

A character bible is rules, not a story: short, non-negotiable laws about who the character is — "Always wears the cracked aviator goggles", "Never raises her voice", "Walks with a slow, deliberate limp." When your workspace has it enabled, the Bible tab on the right edge of the Character Studio (or the book chip in the top bar) opens the drawer. Add rules yourself, or describe the character in plain language and Theo drafts the rules — you review each one, accept what fits, and nothing saves until you hit Save bible. Each rule carries a category chip — Identity, Look, Behavior, Speech, World — tap it to recategorize.

Rules ride into every take. Stage the character in any production and their bible injects into the render automatically. On fresh shots, Theo goes further: he restructures your direction around the rules— wardrobe laws shape what's visible, behavior laws shape the action, speech laws shape the dialogue — while the character's multi-angle renderings keep their identity locked. Your note in the take history stays exactly as you wrote it.

The Introduction. Once the bible is saved, render the sealed-bible film: a pitch-black silhouetted room, one spotlight, one microphone — your character walks up out of the dark, says hello, and dictates the rules of their life in their own voice. It renders on the Cinema engine at 1080p for a full 15 seconds, in the background — keep working and the film lands in the drawer, ready to screen full-bleed on the studio stage. Re-render it whenever the bible evolves.

You can bring your OWN clip onto the set. Tap the clapperboardin the composer (or drop a file straight onto it) to add an MP4, MOV, or WebM clip — up to 100 MB and about 15 seconds. The clip lands in the scene as an Uploaded take: playable immediately, free to register, and never modified or re-encoded.

Then direct it like any other take: "make it rain", "swap the car for a bike", "relight it as golden hour". Each note renders a NEW ~10-second take built from your footage — your original stays untouched as a base you can always branch from, and every follow-up edit chains conversationally exactly like generated takes.

• Short clips work best. Edits render as ~10s takes, so a tight source clip is what the studio can faithfully rework end to end.
• Capture characters from it. Pause your uploaded clip on the stage and use the capture button to save anyone in it to your character library — then stage them in brand-new scenes and productions.
• Continue its story. Start the next scene from your clip and Director Mode carries its final frame forward as the visual anchor.
• Region note. In some regions the engine can't edit uploaded footage yet (generated takes still edit everywhere) — you'll see a clear message if that applies to you.

Every render passes through a provider-side content check, and that check famously misreads realistic faces — even fully AI-generated ones — as photos of real people. Because the check isn't perfectly consistent, Director Mode now retries a flagged take automatically (a fresh submission is a fresh check) before ever showing you an error. The take simply keeps rolling; retries never cost extra credits.

If a take still fails flagged, the message tells you the retries already happened. From there: try a more stylized or different-angle shot of the character, or — on a Cinema take — tap Try on Flash right on the failed take. It re-runs the same direction and anchors on the Flash engine, whose content check is entirely separate and often waves the same take through.

Why edits can trip it: an edit hands the engine your prior take's footage and reference images along with the note — so even a harmless direction like "just change the logo" re-runs the check on everything attached. That's why the fix is a retry or an engine switch, not a different wording. Only a direction the filter blocks for its own phrasing (violence, explicit content, named real people) needs a rephrase.

Director Mode is rolling out gradually and runs on a preview engine — output is 720p today, and clips top out at ~10 seconds. In some regions, editing applies only to takes generated inside Director Mode (not to footage you upload). Occasional retakes are part of the craft: if a note lands oddly, run it again or rephrase.