Director ModeBETA

Say it once.
Then direct it.

Describe the video you want and the first take lands in about a minute, picture and sound together on whichever of 21 video models you pick. After that you stop prompting and start directing. One note at a time, each one a brand new take, with everything you already made still sitting right there.

Start directingWatch a note become a take

Included with the Pro and Teams plans. No separate video subscription.

Director ModeThe lighthouse
Take 1 · rendering
A lighthouse at dusk, waves rolling in
Product reveals✳Ads that move✳Music videos✳Openers✳Explainers✳Trailers✳Social cuts✳Scene work✳21 models, one set✳Anything you can picture✳Product reveals✳Ads that move✳Music videos✳Openers✳Explainers✳Trailers✳Social cuts✳Scene work✳21 models, one set✳Anything you can picture✳
The loop

A note. A take. Nothing lost.

Give a note the way you would on a set. Director Mode shoots a new take that remembers the one before it. Change your mind and the take you liked is still right there, one click back.

Take 1 · ready
A courier crossing a wet street at night

Give it a note

Each note shoots a new take. The old ones stay exactly where they were.

Take thread

1 of 30

Click any take to bring it back. Star the keeper so the cut knows which one you meant.

The rule of the room

Nothing you make is
ever overwritten.

Takes stack into a thread you can branch from, star, and come back to a week later. There is no undo to be afraid of, because there is nothing to undo.

Pick a tier

Fast while you explore. Cinematic when it counts.

The tier is a per-take decision, not a plan you are locked into. Rough something out on the quick one, then shoot the keeper on the good one.

Flash

The one you explore in.

Try three versions of a shot before lunch. Flash is where you live while you are still working out what the video actually is.

Clip length
Around eight seconds
Frame shapes
16:9 and 9:16
References
One opening frame
Best for
Drafts, tests, social cuts
The model market

Every model. One set.

21 video models from 11 labs, picked per take. Open the Model picker on any tier, read the spec sheet, and shoot the take on the model that suits the shot. The tier's built-in engine is always one tap away, and swapping models never changes how you direct.

Flash

7 models
  • Gemini OmniBuilt-in default
    Google DeepMind

    The conversational default — direct changes in plain language and the shot stays the shot.

    Up to 8s720pNative audioImage anchorEdits footage
  • Aleph 2.0Edits footage only
    Runway

    Runway's video-editing flagship — hand it footage and direct the change.

    Up to 30sSilent
  • Act-TwoEdits footage only
    Runway

    Runway's performance-transfer model — your character performs the footage's exact delivery.

    Up to 30sSilentImage anchor
  • Kling 2.5 Turbo
    Kuaishou

    Crisp, energetic motion at turbo speed.

    Up to 10sSilentImage anchor
  • Hailuo 2.3 Fast
    MiniMax

    Lifelike motion from a single frame, fast.

    Up to 10s768pSilentImage anchor
  • LTX 2.3 Fast
    Lightricks

    High-resolution drafts on a budget — up to 20 seconds of 4K.

    Up to 20s4KNative audioImage anchor
  • Pika 2.2
    Pika

    Playful, stylized five-second shots.

    Up to 5s1080pSilentImage anchor

Cinema Lite

1 model
  • Seedance 2.0 MiniBuilt-in default
    ByteDance

    The Cinema Lite default — the Seedance family's draft tier, about half the credits.

    Up to 15s720pNative audioImage anchorEdits footage

Cinema

10 models
  • Seedance 2.0Built-in default
    ByteDance

    The Cinema default — premium fidelity, footage-anchored edits, native audio.

    Up to 15s4KNative audioImage anchorEdits footage
  • Kling 2.6 Pro
    Kuaishou

    Cinematic motion with a director's eye for camera moves.

    Up to 10sSilentImage anchor
  • Kling 3.0 Pro
    Kuaishou

    Kling's flagship — multi-shot direction with native audio, up to 15 seconds.

    Up to 15sNative audioImage anchor
  • Hailuo 2.3 Pro
    MiniMax

    Best-in-class physics and human motion at 1080p.

    Up to 10s1080pSilentImage anchor
  • MiniMax H3
    MiniMax

    MiniMax's arena heavyweight — native audio up to 4K.

    Up to 15s4KNative audioImage anchor
  • Veo 3.1
    Google DeepMind

    Google's flagship — native audio and razor prompt adherence.

    Up to 8s1080pNative audioImage anchor
  • Sora 2
    OpenAI

    Native sound and sharp storytelling instincts.

    Up to 12s720pNative audioImage anchor
  • Grok Imagine 1.5
    xAI

    xAI's fast cinematic engine — lip-synced dialogue on a budget.

    Up to 15s1080pNative audioImage anchor
  • Happy Horse 1.1
    Alibaba

    Alibaba's arena climber — lip-synced dialogue in seven languages.

    Up to 15s1080pNative audioImage anchor
  • LTX 2.3 Pro
    Lightricks

    Open-source cinema — 4K-capable takes at the lowest premium price.

    Up to 10s4KNative audioImage anchor

Cinema Ultra

3 models
  • Seedance 2.5Built-in default
    ByteDance

    The Cinema Ultra default — up to 30 seconds in one continuous take.

    Up to 30s1080pNative audioImage anchorEdits footage
  • Sora 2 Pro
    OpenAI

    OpenAI's premium cut — longer, steadier, cinema-ready takes with native sound.

    Up to 20s720pNative audioImage anchor
  • FLUX 3
    Black Forest Labs

    Black Forest Labs' video flagship — 20-second sound-designed takes.

    Up to 20s1080pNative audioImage anchor

Names and marks belong to their labs. This roster mirrors the studio's live Model picker, so it changes as models ship.

The production

Build it scene by scene.

Open the next moment and it picks up exactly where the last scene's chosen take left off. Same world, same cast, next beat. Star a keeper in each scene, then play the whole thing through as one film or hand it to Final Cut.

  • Up to two hundred scenes in a single production.
  • Every scene keeps its own take thread, so a rework never touches the rest.
  • Reorder scenes on a timeline and trim, crop, or resize any clip in place.
Scene 01 · Opening4 takes
Scene
The cast

Capture a face once. Keep it for the whole film.

Pause any take, draw a box around whoever deserves an encore, and they join your library as exact pixels. A full turnaround comes with them, so they hold their look from every angle in every scene you stage them into.

Turnaround sheet

Portrait
Full
Side
Back
3/4 L
3/4 R
3/4 B
Staged in
4 scenes
Voice
Low, unhurried
Angles
Seven, plus a hero still
Bible
Rules that ride every take
21

Video models to pick from

11

AI labs on one set

200

Scenes in one production

30

Takes per scene

7

Angles per cast member

30s

In a single unbroken take

The rest of the studio

Everything a shot might need, one tap away.

Pick the model per take

Every tier opens a Model picker of 21 video models with a spec sheet and warnings against what you have staged. Each take renders on the one you chose, and the built-in default is always one tap away.

Start from your own pixels

Attach a photo as the opening frame, or a stack of references so the shot is built from the exact look you already have. Up to thirty on the top tier.

Lighting you pick by looking

Sixteen setups, each one photographed on the same stand-in so you choose a look by seeing it rather than by reading its name.

Bring your own footage

Upload a clip you already shot and direct it the same way. Every note becomes a new take, and the original is never touched.

Story Mode

Describe a film once. You get a cast, a style register, and a shot list you can edit, then roll the whole thing shot by shot.

Lyric videos

Upload your song and the words land on the picture in time with it, in a look you pick from a shelf of eighteen.

An audio desk

Score, voice, effects, and per-scene levels on one timeline, with ducking under narration so nothing fights the words.

Final Cut

Every starred take assembled into one film with transitions, narration, and a soundtrack, ready to download.

Perform

Attach an audio track and the person on screen performs it, or point it at footage you already have and re-voice that instead.

Six frame shapes

From 21:9 to 9:16, picked once for the production so every scene cuts together instead of fighting the format.

How it goes

Four moves, start to finish.

01

Describe it

Type the video you want and hit send. Attach a photo if the shot should start from something real. Take one lands in about a minute.

02

Direct it

One note at a time. Warmer light. Make it rain. Every note is a new take that remembers the last, and the thread keeps all of them.

03

Assemble it

Star the keeper, open the next scene, and keep going. Each scene continues from the take you chose in the one before it.

04

Cut it

Trim, score, narrate, and export the whole production as one film. Or grab the scenes in order and cut it somewhere else.

How it compares

Same models. A different way to work.

Runway, Higgsfield, and 8frame put many video models behind one subscription too. The difference is what happens after the first render.

And in chat

Not every video needs a whole studio.

Ask Theo for a clip in an ordinary conversation and it renders right there in the thread. When the idea turns out to be bigger than one shot, carry the prompt straight into Director Mode and keep going.

Start creating freeSee video in chat
A short video rendering inside a Theo conversationVideo in chat

Free · No sign-in · Always on

See it running, 24 hours a day.

OpenCharts TV is fifteen channels of never-ending video, every clip made in seconds with Director Mode. Tune in, switch between Shuffle and Story, and when a scene grabs you, remix it in the studio.

Watch OpenCharts TV
  • 01CommercialsBrought to you by nobody.
  • 02KidsGentle worlds, big hearts.
  • 03Deep FieldTransmissions from the far edge.
  • 04WildThe planet, up close.
  • 05MidnightRain, neon, and unanswered questions.
  • 06Channel ∞Anything. From anywhere.
  • 07RetroTelevision from a living room that no longer exists.
  • 08ComedyOne joke. Ten seconds. Full commitment.
  • 09CookingRecipes from every dimension.
  • 10Rest & RelaxationClose your eyes. Or don't.
  • 11Backgrounds & MusicSomewhere calm, all day.
  • 12PodcastTwo mics. Zero facts.
  • 13TerrifyingDo not watch alone.
  • 14LovePoetry you can see.
  • 15SportsAll the coverage. None of the sports.
Questions

The things people ask.

What is Director Mode?
An immersive conversational video studio. You describe the video you want, watch the first take render in about a minute, then direct changes in plain language. Each note becomes a brand new take.
Which AI video models does Director Mode support?
21 video models from 11 labs, picked per take: Gemini Omni and Veo 3.1 (Google DeepMind); Aleph 2.0 and Act-Two (Runway); Kling 2.5 Turbo, Kling 2.6 Pro, and Kling 3.0 Pro (Kuaishou); Hailuo 2.3 Fast, Hailuo 2.3 Pro, and MiniMax H3 (MiniMax); LTX 2.3 Fast and LTX 2.3 Pro (Lightricks); Pika 2.2 (Pika); Seedance 2.0 Mini, Seedance 2.0, and Seedance 2.5 (ByteDance); Sora 2 and Sora 2 Pro (OpenAI); Grok Imagine 1.5 (xAI); Happy Horse 1.1 (Alibaba); FLUX 3 (Black Forest Labs). Gemini Omni, Seedance 2.0 Mini, Seedance 2.0, and Seedance 2.5 are the built-in tier defaults; the rest are market picks you swap in from the Model picker. The roster grows as new models ship.
Can I switch models per take?
Yes. Every tier has a Model picker with search, filters, and a spec sheet for each model, including capability warnings against what you have staged. Each take renders on the model you picked, the tier's built-in default is always one tap away, and a market pick renders a fresh take from your note and references. The Cinema tiers can then edit that footage like any other take.
How is that different from writing a better prompt?
You do not rewrite a prompt for every change. You give a note the way you would on a set, and Director Mode shoots a new take that remembers the one before it. Nothing you already made is lost, so there is no cost to trying something.
Can I keep the same character across scenes?
Yes. Pause any take, draw a box around them, and they join your library as exact pixels, up to 100 characters. A full seven-angle turnaround rides along, so they hold their look from the front, the side, and the back in every scene you stage them into.
Can I make a whole film, not just one clip?
Yes. Build it scene by scene, up to 200 scenes with 30 takes each, with each scene continuing from the take you starred in the one before it. Then play the production through as one film, or send it to Final Cut for transitions, narration, and a score.
Can I direct footage I already shot?
Yes. Upload a clip and it becomes take one. Every note after that renders a new take anchored on your footage, and the original file is never touched.
How is Director Mode different from Runway, Higgsfield, or 8frame?
Runway, Higgsfield, and 8frame also put many video models behind one subscription. Director Mode runs the same frontier roster inside a directing loop: every note is a new take that remembers the last, nothing is ever overwritten, your cast holds its look across up to 200 scenes, and Story Mode, Lyric Video, Perform, and Final Cut live in the same studio. It also comes inside the OpenCharts plan you already use for documents, decks, code, and agents, so the credits are shared instead of being a separate video bill. The full comparisons are at /compare/runway, /compare/higgsfield, and /compare/8frame.
What does it cost?
Director Mode is included with the Pro and Teams plans and runs on the AI credits those plans already include, so there is no separate video subscription. Pro includes 5 million AI credits a month; a Flash take is 2,000 of them, and the Cinema tiers and market models cost more per take. Anyone can watch what it makes, free, on OpenCharts TV.
Soft clouds drifting across a bright sky

Step onto the set

Describe a video, watch it roll, then direct it take by take. Included with Pro and Teams, and nothing you make is ever overwritten.

Start directingRead the guide
Included with Pro and Teams · One plan for the whole studio · Your takes stay yours