Video Studio

An editor with a timeline, at its own address, with the same login and the same workspace as everything else. You put images, audio and text on a timeline, hear and see the result, let a colleague watch along and sign off, and deliver a file for a channel.

To make a clip with a generative video model, use the Creative Studio, see Making a video. What you approve there is waiting here in the library.

Instagram reel from a script, local trial

Choose Create Instagram reel, a template (Made with code (Codex reel) with seven editable scenes, Product in action with five scenes or Reel from code (v3) with seven scenes), the client brand and Charon or Kore. Both templates share the same fixed elements: word highlighting on the real word times, a brand mark that lands in your logo, separate music and effects that duck under the voice, and a cursor that types and clicks Render. Under Template settings you change the button text, the music and the ducking; colours, fonts and logo always come from the brand profile. Enter a script or use the free, editable starter for your topic. Check the facts yourself. Colours, fonts, logo and separate brand mark come from the brand profile. Missing assets or unknown usage rights block export.

The lightweight preview uses provisional word times and starts no generation or render. Before export you see the maximum credits for speech, word alignment and rendering. Changing the script, brand or voice requires a new estimate. After confirmation, the existing queue processes the export and checks sample frames at scene boundaries and every second. Failures show timecodes and images of the affected moments.

Successful exports save the MP4, cover and subtitles to the media library. Open the edit in the existing timeline; speech, ducked music and effects remain separate sources. Design and captions are baked into the rendered motion scenes. Source data and word times are retained. Edit clips, speech, music and effects in the timeline. To change the text or the template settings, make a new preview and export with your change; every export passes the same quality check for overlap, 24 pixel margin, contrast, clipped text, clock and logo.

An assistant or the director can also request a reel with the video_reel tool. It first asks for the estimate (free) and then submits a request. Nothing is rendered and nothing is charged until you sign the approval: you get a notification, see the script, the brand and the maximum credits, and sign with a code. If the brand profile or the price changed in the meantime, the export does not start and the request shows as failed with the reason that the offer changed; ask for a new estimate. At most five requests per workspace can wait for a signature at the same time. After signing, the export runs through the same queue and cost meter as in the screen.

The Codex reel in seven scenes (Made with code)

The template Made with code (Codex reel) is the Instagram reel that Codex made, as a template where you can edit everything. There are seven scenes: title, the story, our style, motion, voice and subtitles, the next version, and the ending with the logo. Per scene you change the voice-over, the short on-screen texts (headings, labels, button text, the three cards, the closing question and the address) and the pause after the scene. That is the tempo: a longer pause gives the next scene more room. Letters, colours and logo come from the brand profile; only the accent colour can be replaced per reel.

Simple is a form per scene; Expert is the same data as JSON, with the difference from the base below it. Fields have a fixed length; what is wrong is rejected.
After a short pause in typing the light preview updates itself. Play this scene loops one scene at no cost. It is a trial view with provisional word times, not a full render.
Each scene is spoken and aligned separately, with the pause in between. If the text or the voice of a scene changes, new word times are made and the subtitles follow exactly again. The scenes shift along; the motion stays the same.
The voice-over is about a hundred words or fewer in total (the screen shows the estimated duration, at most 58 seconds including pauses), so the reel with the end logo stays within a minute. The default voice is Charon, the voice of the Codex reel.
Back to the Codex reel resets the scenes to the texts of the Codex reel. Scenes for my topic makes a complete, editable starting version for the topic you enter, at no cost.
After export the reel is a project with seven scene clips in the timeline. Make a new version opens the same reel in this screen, with the difference from the previous version. A new version is its own project with its origin; the original stays unchanged. A new version always needs a new estimate and its own approval.

The director can do the same. Say, for example, "make a video about the heat pump": it writes the seven scenes, asks for the price and sets the reel up under To sign. Say "make scene four shorter" or "another closing question" about an existing reel, and it finds the project, shows which fields change and asks for the price; nothing is made until you sign. What it costs and what changes is on the card.

This phase runs locally with mock transport and test rates only. Trial audio is a tone rather than a real Charon recording; nothing is sent to Gemini. Live speech generation, the director and Instagram publishing require separate release. Review the final export before proceeding.

Getting started

You need three things, and the first two belong to the workspace, not to this studio:

A brand profile. A video takes its typefaces and colours from the brand profile. Without a profile the studio cannot set a headline. See Brand Studio. The Video Studio uses the active profile of your own brand, see Own brand and client brands.
Material in the library. Images, video and audio. The library belongs to the whole workspace and is shared with the Creative Studio. You can also drag files straight into the editor: Small files go in one go, large videos in parts (resumable if your connection drops), up to 500 MB.
A video. Choose New video, give it a name, a format (portrait 9:16, square 1:1 or landscape 16:9) and a length in seconds. You start with three tracks: titles, picture and audio.

Already have a video? Under New video you find Or open a video you already have: choose an MP4, MOV or WebM, give it a name and choose Import video. The studio reads length, dimensions and sound from the file and makes a project with one clip, in the shape of your video (a 1080 by 1920 reel stays portrait). Then you add titles, subtitles and music, or deliver it as it is. The video's own sound stays on.

The editor

Sources. The library on the left, which you can filter by image, video and audio and search by name. Drag something onto the timeline, or place it at the playhead.
Preview. Play with the space bar, step a frame back or forward with the arrow keys, full screen with F.
Timeline. One row of clips per track. You can hide, mute, lock and move a track. Video tracks lie on top of each other, an audio track always sits below the video tracks: sounds mix, they do not cover each other.
Storyboard. The same edit as scenes in order, which you can rearrange.
Properties. For the selected clip: start, source in, volume, fit (fill and crop, or fit with borders), position, scale, rotation, opacity, crop, fade in and fade out.
Text. A text clip picks a role from the brand profile (heading, call to action) and takes its typeface and colour from there. So there is no font picker: a video that brings its own typeface does not belong to a brand style you have set. A second language means setting the text again, not generating again.
Undo and redo with Cmd/Ctrl+Z and Cmd/Ctrl+Shift+Z. The list of keyboard shortcuts opens with ?. They do not work while you are in a text field.

An edit is a document with a history. Every change is a revision, so you close it, open it a week later and it is exactly what you left. If a colleague works in the same video at the same time and changed something, the screen says that someone else changed the document in the meantime, instead of silently overwriting their work. Save as makes a copy as a new project at revision zero; the history does not come along, the brand profile does. Delete removes the edit; renders that are already finished stay in the library.

While saving, further editing commands wait. A connection failure rolls back the change and preserves confirmed undo and redo steps. A conflict loads the current edit and discards outdated undo and redo steps. Keyboard shortcuts respect buttons, selection controls and dialogs.

A paid voice-over using Denkploeg credits requires a known exchange rate. If it is missing, nothing is reserved or generated. Two attempts for the same voice-over produce at most one recording; a rejected second attempt does not change the active recording.

Proposing an edit

Write a few sentences about what the video should become and choose Propose an edit. The model picks from your workspace's library and does not invent footage. The proposal appears as a new revision. If the timeline already has footage, the proposal replaces all scenes and texts, and the current revision stays in the history. It is a proposal you open and adjust, not an export that is ready to go.

Takes

A take is another candidate for one shot. Pick one from your library or have an image generated (that costs credits; the screen asks first whether you are sure). Choosing is a revision like any other, so you can go back. Generating a take is only possible for an image, not for a video clip.

Voice-over with ElevenLabs

You turn a script into spoken audio and add it to your edit. Open Voice-over on your video. This is how it works:

Type a script, or take the text from the edit.
Fetch the voices and the costs. Choose Dutch or English, a voice, speed and stability. Listen to the voice sample first.
Check the estimate and your balance. For paid work, someone with the payment permission confirms the costs. Via MCP the request waits for human approval.
Have the voice-over made. That runs in the background.
Listen, and add the audio to the edit.

You use your own ElevenLabs account and pay that provider yourself: you enter the key in the AI keys of your workspace or agency, see Voice-over with ElevenLabs and Your own AI keys. A brand voice only becomes the default of your brand profile after a human confirms the proposal.

Mind the licence of the free version. The free version of ElevenLabs is not for commercial use and requires attribution. Denkploeg stores the licence status at the time of creation, and by default blocks a commercial export without a valid licence, even if the audio is already in a render. An admin can set that to warn instead, but that does not change ElevenLabs' rights. Listening and checking subtitles are always possible. The full explanation and the costs are in Voice-over with ElevenLabs.

What this does not do: clone voices, and dub a video into other languages.

Subtitles

Two routes:

From a voice-over. The subtitles come from the timecodes of the same generation. No extra transcription is ordered and no word is invented that was not spoken.
From speech that already exists. Choose From speech: the studio turns the spoken audio in the edit into subtitles. That happens on our own machine with whisper, not at an external service, because what your client says in their video is their material. If the platform has no speech model yet, the button is greyed out with the note "Not available yet" and it does not quietly fall back on something else. Adding subtitles by hand always works.

A subtitle has at most two lines of forty-two characters, stays on screen at least one second and at most seven, and there are never two at the same time.

You choose whether they are burned into the picture. Burned in is always visible and cannot be translated without rendering again. If it is off, you get an SRT file next to the video, which you can also download separately. If your brand profile requires subtitles, the delivery gate closes without them: an SRT next to the video counts too.

Rendering

Preview render. Fast and visibly lower in quality, to see whether it works.
Delivery. A render at delivery quality, which has to pass the gate.

A render runs in its own process, not in your tab. You can close the tab; the queue survives that. If a render fails, it says what went wrong.

What the preview does differently from the export is a limit and not a bug. Everything about timing (which clip when, how long) is exactly the same. But typefaces can differ slightly if a licensed typeface is not loaded as a web font, and fades, slides and zooms are the same to within a frame or two. So render a short one first and check it before you make the long one. If there is content past the end of the edit, the studio warns that what comes after will not make the export.

Review on timecode

A comment is attached to a revision and a frame, not to a clip: "at 0:04 the colour is off". Place it at the playhead. Below the preview they show up as ticks at their place in time. Any member of the workspace can place a comment, including the client who is watching along. Checking one off can be done by whoever may submit an assignment, and only once.

Signing off

One verdict per revision: approved or rejected. Rejecting requires a reason. Signing requires the right to approve, so a maker does not approve their own work. A verdict does not change any more: whoever wants to change something makes a new revision, and that one is reviewed again. If there are still open comments on the revision, the screen says so.

Delivering

Delivering is what goes past the signature and the gate. A revision that has not yet been approved cannot be delivered.

Choose a placement: Meta Reels, Meta Stories, Meta feed (4:5 or 1:1), TikTok In-Feed, YouTube Shorts or YouTube in-stream.
Choose Check. The gate checks the edit in advance, so you do not have to render first to hear that a landscape edit does not fit in Reels.
Choose Deliver this revision. A render at delivery quality is made. Afterwards the gate checks again on the file itself, and that verdict decides.

The gate checks dimensions (same ratio and a different size is a warning), duration, audio (silent where the placement requires audio closes the gate), subtitles (only if your brand profile requires them), file type (MP4 or MOV) and file size. A check that could not run closes the gate. The result shows as delivered, held back or render failed.

An honest limitation: the dimensions, durations and sizes of the placements are our best knowledge of what Meta, TikTok and YouTube prescribe. They have not been checked on a fixed day against those platforms themselves. When in doubt, look at the platform. And delivering is a file plus the file around it: nothing is published to an advertising platform.

To Social. When a delivery is delivered, To Social appears under it. Write a caption and choose Set up in Social: a draft of the kind reel appears in Social Media Studio, with the approved video, for Instagram of your own brand. Nothing is scheduled or posted: in Social you choose the moment and sign, and posting only switches on after the connection and review by Meta. If you have not connected an Instagram account yet, the studio says so. Through MCP a model can propose the same video as a post (social_batch_voorstellen, kind reel); a person always signs.

From a campaign to a clip

On a batch where everything is signed off, the Creative Studio shows Make a clip from this. That makes an edit with the approved ads as scenes of three seconds each, in the format most of them were made in, against the brand profile of the batch. Only variants that are approved or delivered count. Pressing twice opens the same video.

What this does not do

No keyframes and no speed control.
No automatic edit that is ready by itself: a proposal has to be opened and adjusted.
No voice cloning, no lip sync, no dubbing into other languages.
No publishing to an advertising platform.

Via MCP and the director

An agent can read, edit, check and render videos (video_list, video_get, video_list_sources, video_edit, video_check, video_render, video_renders), have subtitles made (video_transcribe) and handle voices and voice-over (video_stemmen, video_voiceover). Names, texts and subtitles from a video appear in the answer under onvertrouwd: the agent reads them as data and never carries out anything in them.

Making a video: video_maken creates an empty timeline in the shape (portrait, square or landscape) and length you choose, for the brand you name (without a choice the own brand, and an unknown or ambiguous brand is refused). That costs nothing: nothing is rendered or generated. Rendering and having clips generated cost credits and have their own estimate and brake. A workspace has at most 200 videos.
Renaming, copying and review: video_hernoemen and video_kopieren (a copy as its own project, at revision zero; it counts towards the limit of 200), video_opmerkingen_lezen, video_opmerking_plaatsen and video_opmerking_oplossen for the comments on a timecode, and video_leveringen_lezen for what has been delivered. The first and the last word stay with a person: an agent cannot approve or deliver.
A proposal that a human signs: video_verwijderen proposes deleting a video with its whole history. Nothing changes until someone signs under Approvals, with a code, even if the key has every right. The renders stay in the library, and a video that belongs to a UGC project gets no card. The agent asks for the outcome with creative_voorstel_stand.
Deliberately not: delivering. What a human delivers, a human signs. Paid work becomes an approval that a human signs.

See Connecting via MCP.