Creative Studio#
The complete guide to the Creative Studio: generating images and videos with AI, telling a multi-scene story, orchestrating a production in the Node Canvas and assembling the result in the montage.
How to read this page
Start with Part I: it defines the words and common rules (access, credits, models). Then read the part for the engine you use: II for images, III for video (Light, Storyboard, Node Canvas), IV for the montage. Part V gathers limits, best practices and frequent questions.
Fundamentals#
What the Creative Studio is, what it produces, who can use it, the words used throughout this documentation and the path every generation follows.
1.1The three engines and the two modes#
The Creative Studio brings together three engines, picked from the selector at the top of the screen. Each produces one kind of content; all of them save their results in the project Library.
| Engine | What it produces | Modes | See |
|---|---|---|---|
| AI Image | Marketing visuals, flyers, consistent image sets | Light | Part II |
| AI Video | Video clips, multi-scene stories, voices | Light or Node Canvas | Part III |
| Video Montage | A final video assembled on a multi-track timeline | Single mode | Part IV |
Two distinct modes: Light vs Node Canvas
Light mode (universal)
a direct input field and a few settings. Designed for all accounts (free and paid) and all devices (smartphone, tablet, computer). It gathers single-shot generation (Text/Image → Video, Motion control) as well as the Storyboard, a step-by-step linear generator of 1 to 8 chained scenes, without any graph.
Node Canvas mode (advanced)
an infinite 2D workspace where each step is a card wired to others through typed ports (references, camera, context, prompts, voices, images, videos). Available to all plans (including free) on computer and tablet (screen ≥ 768 × 600 px). It features a Prompt Composer bar (⌘K / ⌘J) to plan complete graphs, collaborative team comments, and direct handoff to the multi-track montage timeline.
1.2Access by plan and device#
| Feature | Free plan | Paid plans or purchased credits | Phone and tablet |
|---|---|---|---|
| AI Image (Light mode) | ✓ Yes | ✓ Yes | ✓ Yes |
| AI Video: Single clips (Light mode) | ✓ Yes | ✓ Yes | ✓ Yes |
| AI Video: Linear Storyboard (Light mode) | ✓ Yes | ✓ Yes | ✓ Yes |
| AI Video: Node Canvas (Advanced Graph mode) | Yes, on a tablet or computer | Yes, on a tablet or computer | - |
| Video Montage (Multi-track timeline) | Yes (3 exports included / 30 days) | ✓ Yes | ✓ Yes |
“Tablet or computer” means a screen at least 768 × 600 px and a pointer (mouse, trackpad, or large touch surface): wiring cards requires sufficient screen space. On a smartphone, the Studio opens automatically in Light mode, with no mode selector.
1.3Glossary#
These terms have a precise meaning throughout the documentation and the interface.
| Term | Definition |
|---|---|
| Node | A step of the Node Canvas graph: what it receives (its inputs) and what it produces (its output). Examples: Prompt +, Reference, Video. |
| Card | A node as shown on screen: a rectangle with its title, settings and connection dots. |
| Port | A colored dot on a card's edge. Inputs surround the card; the output sits on the right edge. The color tells the data type. |
| Link | A grey wire connecting one card's output to another card's input. |
| Composition | A saved Node Canvas graph (available on all plans on desktop and tablet), listed in “My compositions”. |
| Montage | A timeline and its clips, listed in “My montages”. Independent from the open composition. |
| Track / Clip | A timeline row (main video, voice, music…) / a media item placed on a track, with a start and a duration. |
| Storyboard / Scene | The linear generation tool of Light mode (available to all plans and screens without nodes or graph) that splits a brief into chained scenes / one shot of that sequence, with its description, camera move and duration. |
| Credit | The billing unit of generations. The cost is shown on every button before launch. |
1.4The life of a generation#
Describe
Enrich (optional)
Pick the model and format
Generate
Retrieve
Duration, failure and cancellation
- An image takes a few seconds to a minute; a video, one to several minutes depending on the model, the duration and the provider's load.
- A video is only charged once delivered: a failed or canceled video costs nothing.
- A failed (or canceled) image, voice, lip-sync or paid export is refunded automatically.
- Relaunching an identical generation does not bill it twice: each launch is identified, and a double click finds the job already sent.
1.5Models and credits#
The tables below are the reference: formats, durations, number of reference images accepted and cost of every model offered in the Studio.
Image models
| Model | Tier | Max. references | Credits / image |
|---|---|---|---|
| GPT Image 2.6 Flare | Standard | 16 | 0.5 |
| GPT Image 2.6 Sunburst | Premium | 16 | 1 |
| GPT Image 2.5 Flare | Standard | 16 | 0.5 |
| GPT Image 2.5 Sunburst | Premium | 16 | 1 |
| Flux Pro 1.1 | Standard | 4 | 0.5 |
| GPT Image 2 | Standard / Premium | 5 / 6 | 0.5 / 1 |
| Nano Banana 2.1 | Premium | 14 | 1 |
| Seedream 5 Pro | Premium | 10 | 1 |
Video models
Each row gives, for one model: the length of a shot, the frame formats, the output qualities with their price in credits per second, and the number of reference images accepted. Omni Flash 1.2 is the default model.
| Model | Shot length | Formats | Quality: credits / second | Reference images |
|---|---|---|---|---|
| Omni Flash 1.2 | 3 – 15 s | 16:9, 9:16 | 360p: 0.33 · 720p: 1 · 1080p: 1.5 | 10 |
| Kling 4.0 | 3 – 30 s | 16:9, 9:16, 1:1, 21:9 | 720p: 1.25 · 1080p: 1.88 | 10 |
| Seedance 2.0 | 4 – 15 s | 16:9, 9:16, 1:1 | 480p: 1.75 · 720p: 2.5 | 9 |
| Seedance 2.5 | 4 – 30 s | 16:9, 9:16, 1:1, 21:9, 4:3, 3:4 | 480p: 2.8 · 720p: 4 | 30 |
| MiniMax H3 | 5 – 15 s | 16:9, 9:16, 1:1, 21:9, 4:3, 3:4 | 480p: 0.85 · 768p: 1 · 2K: 2.2 | 1 |
| Wan 3.0 | 3 – 30 s | 16:9, 9:16, 1:1, 4:3, 3:4 | 480p: 0.5 · 720p: 1 · 1080p: 2 | 10 |
Reading a video's price
- Price = shot length × credits per second of the chosen quality, rounded up to the next credit. Example: 5 s on Kling 4.0 (1080p) = 10 credits; 10 s on Seedance 2.5 (720p) = 40 credits.
- Unless you choose, quality is the base rate (720p; 768p on MiniMax). Kling 4.0 offers a choice between 720p and 1080p. Switching model drops a quality it does not offer.
- An end image can complete a start image (with no other reference connected): the video ends on it.
Other costs
| Operation | Cost |
|---|---|
| Generated voice (ElevenLabs, MiniMax) | 1 credit per 1,000 characters; listening to voices is free |
| Create a voice from a style | 8 credits per voice; refunded if creation fails |
| Lip-sync | 1 credit per started second of video |
| Extract a clip's voice (montage) | 2 credits per started minute |
| Storyboard | The scenes' price, plus 1 credit per completed storyboard (breakdown and assembly) |
| Montage export | Included over a rolling 30 days: Free 3, Starter 30, Pro 100, Agency 300, Enterprise unlimited; beyond that, 1 credit per 5-minute block |
| Canvas JSON export, organizing, listening | Free and unlimited |
AI Image#
The AI Image engine produces publication-ready marketing visuals: one visual explored in variants, or a consistent set for a whole range.
2.1Visual types and formats#
| Type | What you get | When to choose it |
|---|---|---|
| Hybrid | An image (photo, product, scene) composed with a text hierarchy: headline, subline, call to action. | Promotions, launches, events, products. |
| Typographic | A poster built from typography: large words, blocks, textures, no dominant photo. | Quotes, announcements, key figures, carousels. |
Available formats: 1:1 (square), 4:5 (feed portrait), 9:16 (story, reels) and 16:9 (landscape). You can attach images (product shot, portrait, logo): they act as visual references, within the chosen model's limit (chapter 1.5).
2.2Variants and Bundle#
Two ways to produce several images from one brief, billed at the same price per image.
Creation Modes: Variants vs Image Bundle / Set
Creative Studio offers two distinct generation modes: exploring multiple creative directions for a single post (Variants) or producing a unified, multi-image pack across your brand ecosystem (Bundle).
Generates comparative visual propositions for a single brief. Perfect for A/B testing different compositions, angles, and moods before sharing.
Generates a cohesive series covering your company's offerings and brand universe. The first visual acts as an anchor: subsequent images share its visual essence, lighting, color harmony, and art direction while rendering distinct scenes.
Action "Use as template" & Asset Library Grouping
In the Assets / Library page, each generated bundle is neatly grouped under a single card with a Set (N) badge and interactive slide navigation. Inside any asset's (...) menu, select "Use as template" to immediately lock it as the default visual reference in Project Settings > AI.
2.3Brand injection#
Your brand identity (defined in Brand Identity) can accompany every generation, so results stay recognizable without rewriting everything in the prompt.
What is added
- The brand's primary and secondary colors, and its fonts.
- The name, the official logo (reproduced faithfully, never reinvented) and the editorial tone.
- In the Node Canvas only: the products chosen in the Context card.
Where to enable it
Light mode
the composer's brand option.
Node Canvas
a Context card wired to the Context input of an Image, Reference or Video card. Without a Context card, nothing from the brand is added.
2.4Prompt library#
Copy these prompts directly into Creative Studio. They are engineered for high-converting visual hierarchy and clean aesthetics.
“Create an elegant advertising flyer for an exclusive 30% discount on our autumn collection. Minimalist style, bright clean background with subtle brand palette accents, modern promo badge in top-right, clear CTA 'Shop Now'.”
“Generate a product launch visual for our new mobile fintech app. Smartphone mockup in foreground with sleek UI, soft studio lighting, subtle background gradient, bold typography, and prominent brand logo.”
“Vertical 9:16 story flyer for a free webinar 'Master AI for Your Business'. Punchy typography readable on mobile screens, speaker portrait with clean cutout, 'Limited Seats' badge, date, and 'Swipe Up' button.”
“High-impact visual for a 48-hour flash sale. High contrast, dark dynamic backdrop with neon brand color highlights, bold 'FLASH SALE' headline, stylized countdown timer graphic, and urgent CTA button.”
“Cover visual for a LinkedIn carousel: '5 steps to automate digital marketing in 2026'. Clean corporate aesthetic, precise geometric icons, high contrast ensuring maximum feed legibility.”
“Cinematic low-angle dolly shot of a modern founder working in a sunlit glass office, typing on laptop, smiling as a success notification pops up. Slow forward camera motion, golden afternoon rim lighting, photorealistic, high definition.”
“Official poster for the Annual Tech Innovation Summit. Conference stage backdrop with blue and amber spotlights, clearly structured date and venue details, dedicated partner footer, prestigious keynote design.”
AI Video#
Three ways to produce video: one clip in Light mode, a multi-scene story with the Storyboard, or a complete, consistent production in the Node Canvas. This part presents them in that order, from simplest to most complete.
3.1Which tool should I use?#
| Need | Tool | Why |
|---|---|---|
| A clip of a few seconds | Light: Text or Image → video | The fastest: a description, a model, one click. |
| Reproduce a gesture or a dance | Light: Motion control | The motion of a reference video is applied to your character. |
| A 20–60 second story from one brief | Light: Storyboard | The AI splits the brief into chained scenes and delivers one final video. |
| Several shots with the same characters, products and voices | Node Canvas | Each element is defined once and reused everywhere; you control every shot. |
| Assemble, cut, caption, mix | Video Montage | A seven-track timeline and a final export (Part IV). |
3.2Light mode: the four tools#
| Tool | What you provide | Models | Duration |
|---|---|---|---|
| Text → video | A description | All | Per model (1.5) |
| Image → video | A start image, an optional end image, a description | All | Per model (1.5) |
| Motion control | A character image and a reference video whose motion it copies | Kling 4.0 | That of the reference video: up to 30 s (video orientation) or 10 s (image orientation) |
| Storyboard | A brief, split into scenes you review | Kling 4.0, Gemini Omni Flash | Up to 8 scenes |
Settings rules
- The format, duration and resolution offered depend on the model: a format it cannot produce is not shown (Gemini Omni: 16:9 and 9:16 only).
- Gemini and MiniMax use a single image as the start image. Kling 4.0, Seedance and Wan 3.0 accept several images: with two or more, they switch to “reference” mode, where none is forced as the first frame.
- An end image requires a start image; it cannot be combined with several references.
3.3The Storyboard in detail#
Built-in Light mode feature
The Storyboard turns a brief (“a 40 s video for our summer collection launch”) into a sequence of chained scenes, then into one final video. It is an integral part of Light mode: available to everyone, requires no spatial manipulation, and produces an immediately usable result.
Write the brief
Break down (free)
Review and adjust
Generate as a chain
Receive the final video
What happens depending on the model
Kling 4.0
consecutive scenes are grouped into blocks (at most 6 scenes and 15 s) rendered in a single “multi-shot” call: fewer joins, better continuity. A scene carrying its own image always opens a new block.
Gemini Omni Flash
one scene per call, each starting from the previous scene's last frame.
Seedance, MiniMax H3, Wan 3.0
not offered for the Storyboard until frame-to-frame chaining is guaranteed with these models.
Cost
Cost = price of each rendered block (duration × model rate, chapter 1.5) + 1 credit per completed storyboard. Example: 6 scenes of 5 s on Kling 4.0 (1080p) form two 15 s blocks, i.e. 29 + 29 + 1 = 59 credits (or 19 + 19 + 1 = 39 credits at 720p). The total shown before launch follows exactly this rule.
Two distinct features: Storyboard vs Node Canvas
The table below summarizes the fundamental differences in access, interface and usage between these two tools:
| Criterion | Storyboard (Light mode) | Node Canvas (Graph mode) |
|---|---|---|
| Access & Plan | Light mode: available to all plans (free and paid) | Node Canvas mode: available to all plans (free and paid) |
| Supported devices | All devices (Smartphone, tablet, computer) | Computer and tablet only (screen ≥ 1,024 px with pointer) |
| Working paradigm | Step-by-step linear: brief form broken down into scenes | Non-linear spatial: infinite 2D canvas editable at will |
| Nodes and wiring | No nodes, no ports, no wires | 9 interconnectable card types with typed ports and bezier wires |
| AI Planning | The AI suggests a sequence of scene descriptions in the list | Prompt Composer bar (⌘K / ⌘J) planning and laying out nodes on the canvas |
| Visual continuity | Automatic (last frame of a scene feeds the next) | @name shared assets, Reference/Context/Camera cards, manual extension |
| Creative control | Editing text, camera move and duration scene by scene | Full surgical control: regenerate one single shot, independent settings per card |
| Voice and audio | Sound generated automatically by video model (no custom voice) | Dedicated Voice card: catalog, voice creation from description, cloning, voice-over and lip-sync |
| Result & Output | One single assembled final video, saved to the Library | Saved compositions and individual shots transferable to the montage timeline |
3.4Node Canvas: principles#
Key distinction: Node Canvas vs Storyboard
The Node Canvas is an infinite workspace where a video production is described as a graph: each step is a card, and links tell which card feeds which. A character defined once in a Reference card can appear, identical, in ten shots; changing a prompt only affects the shots that depend on it.
Screen anatomy
Palette (right)
creates a card: Reference, Camera, Context, Prompt +, Prompt −, Voice, Image, Video, Note; and “New graph”.
Bottom bar
My compositions, JSON import and export, video download, send to montage, undo / redo, recenter, zoom.
Action bar (above the selected card)
on Image, Video and Reference cards (the latter only when a Prompt is plugged in): model, format, duration and quality for a video, variants, and the Run / Rerun button showing its cost in credits; during a generation, this button becomes “Cancel”. It writes the same settings as the card. The variants count (1 to 4 in the bar) creates copies of the card, offset on the canvas and launched together; a Reference has no variants (its name is unique).
Publish (in the media frame)
an Image, Video or Reference card holding a generated media shows a small “Publish” button in its frame: it opens the publish or schedule window for that media. An image's download is under the image, videos are downloaded from the bottom bar: each click triggers a direct file download.
Selection bar (top)
appears as soon as two cards are selected: Group, Ungroup, Delete, plus a batch Stop button whenever selected cards are currently generating.
Saving
Everything is saved automatically, less than a second after each change, in the open composition. Generated media are automatically re-hosted in durable project storage so their links never expire when reopening a composition. A local copy protects your work if the connection drops: it is recovered on reopening; closing the tab also sends the latest state to the server. If a teammate edits the same composition or montage at the same time, their changes and yours are merged automatically, card by card and clip by clip. A message asks you to choose between your version and the online one only if you both changed the same setting of the same item.
3.4.1Nodes and their ports#
Each node receives data through its inputs and produces one through its output. The table gives, for each, what can be plugged into it.
| Node | Inputs | Output | Good to know |
|---|---|---|---|
| Prompt + | - | Prompt + | No input. Can feed several nodes at once. |
| Prompt − | - | Prompt − | No input. |
| Reference | ImagePrompt +Prompt −Context | Reference | Its picture is uploaded in the card or comes from a connected Image node. Its name is the @mention. A reference cannot plug into another. |
| Camera | - | Camera | No input. Plugs into a Video node only. |
| Voice | ScriptVoice style | Voice / Audio | Upload a file, or generate from the connected script; the voice style creates a voice added to “My voices”. The cost shows before launch. |
| Image | ImagePrompt +Prompt −ReferenceContext | Image | GPT Image, Nano Banana and Seedream models; the number of Reference slots follows the chosen model. |
| Video | ImageEnd imagePrompt +Prompt −ReferenceCameraVoice / AudioContext | Video | Gemini Omni, Kling 4.0, Seedance 2.0/2.5; durations and references depend on the model. Voice mode: voice-over or lip-sync. |
| Context | - | Context | No input. Choose the brand and products; plugs into Image, Reference or Video. |
| Note | - | - | No input or output: free text to annotate your work. Never enters a generation. |
Port colors
A port's color tells the type of data flowing through it. Two pairs share a color: Prompt + and Text (green), Image and Reference (cyan).
| Port | What flows through it |
|---|---|
| Prompt + | The text of what to generate, with its @name mentions. |
| Script | The script: what the voice must speak (same green as Prompt +: a Prompt + feeds it). |
| Voice style | How the voice sounds (age, accent, tone): used to create a voice added to the list. Fed by a Prompt +. |
| Prompt − | What to avoid. |
| Image | An image: the one to rework (Image node), the clip's first frame (Video node), or a reference's picture. |
| Reference | A character, product or place to keep identical from shot to shot. A new slot opens with each connection, up to the chosen model's capacity. |
| End image | The last frame the clip must reach. Pink input, fed by an Image or Reference output (cyan). |
| Camera | The movement (static, close-up, dolly, pan, drone, orbit, zoom) and the camera type. |
| Voice / Audio | An imported or generated voice (ElevenLabs or MiniMax). |
| Video | The generated clip. It plugs into no node: it goes to the edit through the “Add to edit” button. |
| Context | The brand (colors, fonts, tone, logo), products chosen in a Context card. |
3.4.2Connecting and disconnecting#
- To connect, drag from a card's output (right edge) onto another card's input. A green ring announces an accepted link; a red ring a refused one, and a message says why.
- An input takes one link: plugging another replaces the first. Exception: Reference, which opens one more slot per connection, up to the chosen model's capacity.
- When a card has many inputs, the dots follow its outline (left, bottom, right, top edges) instead of overflowing.
- Loops are impossible: a card cannot depend on itself.
- To disconnect: hover a link and click its red cross, click a connected input, or use the cross in the card's “what it receives” summary.
- To move a link without disconnecting it: press and hold the connected input (the dot on the receiving card) and drag it onto another input. Released elsewhere, the link stays where it was.
- Plugging a Prompt + into a card where you had already written: your text is not lost, it moves into the Prompt + card (after what it already holds), with its @name mentions. Same for a Voice's script and style.
- Duplicating a card keeps its inputs (to try a variant: another model, another duration), not its outputs.
3.4.3Generate, track, cancel#
What actually goes to the model
Text
the connected Prompt +'s, otherwise the one written in the card.
@name mentions
each @name is replaced by the matching Reference's image (Seedance receives it as @Image1, @Image2…). A reference without an image blocks the launch, before any charge. Picking a reference with @ also wires it into the card's Reference input (or the cards a Prompt + feeds), up to the model's maximum: beyond it the mention is not added and a message says so. The two stay in step: erasing the mention removes the wire it placed, and cutting the wire turns the mention back into a plain word (a wire you drew by hand is never removed by the text).
#context mention
writing #context in a Prompt + (or in the prompt of an Image, Reference or Video card) creates a Context node and wires it into the cards that prompt feeds; if a Context node already exists on the canvas it is reused, and a card that already has its own context keeps it. Typing # suggests the mention. The link is placed when the mention appears: erasing the text or cutting the wire afterwards does not recreate it on each keystroke. The mention is not offered in Prompt −. The model receives the word “context” without the #, and the brand and products arrive through the connected node.
Start
an image becomes the shot's first frame; a video is EXTENDED, its last frame opening the follow-up. A Video card's “Extend” button creates the next card and the link.
Camera
the type and movement are appended to the text (“Camera: Type …, Movement …”).
Prompt −
added as an exclusion instruction (“Elements to exclude: …”); it is not a native model setting.
End image
requires a start and cannot be combined with several references; otherwise it is refused in the card, before any charge.
Voice
as voice-over, sending to the montage mutes the clip and places the voice on the Voice track; for lip-sync, run the sync with the current voice before sending.
A reference is not necessarily a character
Card states
| State | Meaning | What to do |
|---|---|---|
| Empty | Nothing generated or imported yet. | Generate, or import media. |
| Running | A generation (or import) is in progress; its animation shows in the card. Several cards can generate at once. | Wait, or Cancel (a canceled video is not charged). |
| Ready | The result matches the current inputs. | Use it, send it to the montage. |
| Needs refresh | An input changed since (prompt, reference, format…): the shown result no longer matches. | Regenerate whenever you want: nothing is relaunched or charged automatically. |
| Error | The generation failed; the message says why. | Fix the reported input, then relaunch. |
Canceling a generation
A running generation can be stopped from the card, from the action bar (“Cancel”) or, for a group launched in one go, from the group's Stop button, which cancels all its cards still running. The stop is real: the request is also canceled at the provider (images, videos, voices, lip-sync), not just hidden. The credits of the canceled generation are refunded once; a generation already finished is not refunded. The card goes back to its previous media, or to empty, with no error message. Canceling one variant leaves the others running. Right after a click on Run, a double-click does not cancel it by mistake.
Importing media into a card
Reference (image), Voice (audio) and Video (video, empty card with no input) accept a file. The type is checked before upload; progress and a Cancel button show in the card. Files can be up to 500 MB.
Trimming a shot, listening to a voice
In a finished Video card, “Trim” shows two handles (start, end): each handle shows the exact frame at that moment, which is also how you pick the last frame a follow-up will start from. The cut is made on our servers, with no credits, by any member with access to the project's content; the original shot stays in the library. A Voice card shows the waveform of the produced voice: clicking the wave moves playback there.
3.4.4Organizing: notes, selection, groups#
Note
the card with the T icon holds free text (“Scene 1: redo in 16:9”). It grows with its text, collapses and can be renamed, and never enters a generation.
Selection
a click selects a card; Shift or Ctrl + click adds or removes one; a rectangle drawn while holding the right button (or Shift + drag on the background) selects everything it touches. Dragging a selected card by its header moves the whole selection.
Group
Ctrl+G surrounds the selection with a frame. Its title (double-click to write it) annotates the part: “Scene 1: the beach”. Dragging the title moves the group; Ctrl+Shift+G or the title's cross removes it without touching the cards. A card belongs to one group; an emptied group disappears.
The Prompt Composer bar
To maximize workspace on the canvas, the prompt composer is collapsed by default upon arrival: it appears as a compact keyboard pill positioned strictly in the exact center of the bottom dock. Clicking the pill or pressing H (as well as ⌘K / ⌘J on macOS or Ctrl+K / Ctrl+J on Windows/Linux) smoothly morphs it open into the full Prompt Composer. Pressing H again or Esc contracts it back into its pill. Unlike the linear Storyboard which directly renders a fixed sequence, the Prompt Composer analyzes your text, identifies intentions (duration, aspect ratio, camera moves, mood, characters via @name) and automatically plans a complete graph of interconnected cards on the canvas (Reference, Camera, Context, Prompt + and Video cards). Each laid-out card remains fully editable, movable, replaceable, or regenerable individually.
In the composer, typing @ suggests references: the message's attachments (under the name their Reference card will take) and the canvas references that are already ready. A canvas reference mentioned by @name is sent with the plan like an attachment, within a limit of 4 references in total. Typing # suggests #context: the project's brand and products then guide the plan, without creating a node. Arrows to navigate, Enter or Tab to choose, Esc to close.
Mentions are colored in the text, in the Prompt card and in the composer alike: @reference in cyan (stronger when the reference exists), #context in amber, the Context port's color. A lone @ or # is colored as soon as you type it, confirming the trigger is recognized and the suggestion menu is open.
Canvas keyboard shortcuts
The canvas provides shortcuts to create cards on the fly, navigate, zoom, and copy-paste without leaving the keyboard (disabled while typing in a text field):
| Action | macOS | Windows / Linux | Role |
|---|---|---|---|
| Reference card | R | R | Creates a character or object Reference card |
| Camera card | C | C | Creates a Camera card (movement and shot type) |
| Context card | X | X | Creates a brand or project Context card |
| Prompt + | P | P | Creates a Prompt + card to drive shots |
| Prompt − | N/⇧P | N/Shift+P | Creates a Prompt − card to exclude defects |
| Voice card | V | V | Creates a Voice card for voiceover or lip-sync |
| Image card | I | I | Creates a pure Image card or shot start |
| Video card | M | M | Creates a Video card (generate or extend) |
| Text Note | T | T | Creates a free canvas annotation note |
| Copy | ⌘C | Ctrl+C | Copies selected nodes to clipboard |
| Paste | ⌘V | Ctrl+V | Pastes copied nodes with visual offset |
| Delete | ⌫/Suppr | Suppr/Backspace | Deletes selection (with undo toast) |
| Select all | ⌘A | Ctrl+A | Selects all nodes in the composition |
| Group / Ungroup | ⌘G/⌘⇧G | Ctrl+G/Ctrl+Shift+G | Frames selection or dissolves group |
| Undo / Redo | ⌘Z/⌘⇧Z | Ctrl+Z/Ctrl+Shift+Z | Undoes or redoes document changes |
| Prompt composer | H/⌘K | H/Ctrl+K | Shows or hides the prompt composer (morph) |
| Comments | ⌘⇧C | Ctrl+Shift+C | Opens or closes the comments panel |
| Recenter view | ⌘0 | Ctrl+0 | Recenters and fits view around all nodes |
| Zoom in / out | +/− | +/− | Center zoom (or mouse wheel on pointer) |
| Pan navigation | Espace/F | Space/F | Hold key and drag canvas |
| Clear selection | Échap | Esc | Clears active selection |
Cards without inputs (Prompt +, Prompt −, Context, Camera, Note) collapse into a pill with their round button, to free up space.
3.4.5Compositions, files and handing over to the montage#
New graph
saves the open composition, checks it is listed in “My compositions”, then opens an empty canvas. The montage timeline is untouched.
My compositions
open, rename or delete a composition (the open one cannot be deleted). Images and videos created by VIBE or by an automation are filed in the composition of their conversation or automation: the list shows this origin and an info panel links back to the VIBE conversation or the automations.
JSON export
free and unlimited: the composition only (cards, links, notes, groups, media). Re-imported, it opens as a new composition without touching the montage.
JSON plan
a “gliiz_story_plan” file (title, references, shots in order, with optional model and format) imports as a composition; a composition export also carries a readable summary of the shots (“story”). The same JSON pasted into the Composer prompt builds the graph as is, with no AI reinterpretation: the plan you wrote is the one that runs. Text that is not a valid plan follows the usual path (AI planning).
Send to montage
“Add to montage” on a Video or Voice card, or “Send all” from the bottom bar: shots go in canvas order (left to right, a follow-up after its source); a refused shot is named with the reason.
Working with others on a project
Invited members with the “Content” permission work in the same project, each on their own composition: opening or creating a composition only changes your tab, never a colleague's. The montage stays shared by the whole project. When several of you are on the same composition, everyone's pointer appears with their name and colour (the same as in the Team panel), yours reads “You”, and the Canvas shows who is online. This is not simultaneous co-editing of the same card: if you both change the same composition, your changes are merged automatically as long as you do not touch the same field of the same card (different cards, links or fields: everything is kept, and so is your undo history). The Canvas only warns you (with the option to export your version) when you changed the same field of the same card. The Comments button opens the composition's threads: a comment targets the composition, a card or a result, gets replies and can be resolved (“resolved” means handled, not approved); it never launches a generation. Type @ to mention a teammate: the list offers the people of the project, and the mentioned person, the person you reply to and the composition's creator are notified in the app and on Turbo.
3.4.6Tutorial: two shots with the same character#
Create the character
Write the first shot
Wire the video
Add a camera (optional)
Generate, then extend
Organize and edit
3.5Voices#
The Voice card produces a voice-over from text (its own, or a connected Prompt +'s), or takes an imported audio file. ElevenLabs voices (v3, Multilingual v2, Turbo v2.5) and MiniMax voices (Speech 2.8 HD and Turbo) all go through the same provider; if it does not respond, ElevenLabs is called directly as a fallback.
- Cost: 1 credit per 1,000 characters, shown on the button. Listening to voices is free.
- A voice wired to a Video card serves as voice-over (placed on the Voice track in the montage) or lip-sync (1 credit per second).
- An @name mention in the text is read as the character's name; it needs no image.
Voice creation & speech methods
Script & Catalogue
- Delivery and speed vary without creating a new voice
- Type directly or wire from a Prompt + into Script
Creation from description
- Joins “My voices” and reusable across the project
- Automatically refunded if creation fails
- Deleted by provider after 7 days without use
Audio cloning
- Sample and activation included in the 13 credits
- Consent or own voice mandatory (checkbox required)
- Free reuse if identical audio already cloned in project
In “My voices”, “Rename” only changes the name: its sonic identity, generated audios, and video links remain identical. Importing an audio file directly uses that recording for the current shot; it does not create a reusable clone capable of reading other scripts.
Voice and video: how they fit together
A voice wired to a Video card replaces the shot's native sound. Kling and Seedance then produce a silent picture. Gemini Omni cannot turn its sound off: it stays in the card, but sending to the montage sets it to zero and places your voice on the Voice track.
Separate voice (Voice-over)
Lip-sync
- Externally produced voice (ElevenLabs, MiniMax…): import the audio into the Voice card and wire it to the shot. No cloning or voice cost: only the video is charged (plus 1 credit/s for lip-sync).
- Duration management: if the voice outlasts the shot, the Video card warns immediately and suggests the duration to pick to prevent the voice end from being clipped in the montage.
Listen to the voices
Every catalog voice, with its sample: play, pause, resume. A missing sample is prepared on first listen (a few seconds).
What the Node Canvas prepares for you#
The canvas prepares media and talks to the models so a generation succeeds, or fails fast with the real reason.
- An imported image is prepared before upload: converted to JPEG if it is WebP or another format, brought down to 8 MB and 4096 px on the long side, enlarged to 320 px on the short side. An image that already fits is left untouched; an unreadable image (HEIC outside Safari, damaged file) is refused right away.
- A generated image feeds a video through its durable address in your library, never through a temporary link.
- When a model refuses the inputs (identifiable person, incompatible references), the card shows the refusal right away instead of retrying for minutes, and no credit is charged.
- Cancelling a card also stops the generation at the provider.
- On a short window (laptop, tablet), the canvas fits the screen: the bottom bar and the node palette stay visible; the palette scrolls if needed.
Video Montage#
The montage assembles generated shots and your own videos on a seven-track timeline, then renders the final video. It is independent from the composition open in the Node Canvas: switching compositions does not affect the montage.
4.1Feeding the montage#
From the Node Canvas
“Add to montage” (one card) or “Send all” (every ready shot, in canvas order). The same shot is never added twice.
From a file
“Import video” first opens a preview, then you choose automatic or manual editing (table below). Track buttons import straight onto the chosen track.
| Automatic editing | Manual editing | |
|---|---|---|
| What is done | Rough cut (silences and hesitations removed), captions, B-roll, music, isolated and processed voice | The video is placed on the main track; you edit yourself |
| Cost | A quote is shown before launch (Video Edit pricing) | Free; “Extract voice” from a clip: 2 credits per minute |
| Video length | 3 s to 5 min | Free (export up to 10 min) |
Import rules
- Files up to 500 MB (MP4, MOV, M4V for video). Size is checked before upload: a refusal never makes you wait for the upload to finish.
- Actual progress (optional compression, then upload percentage) shows on the preview.
- The first video of an empty montage sets its format (16:9, 9:16 or 1:1, the closest): it fills the frame instead of sitting between black bars. The format remains editable.
4.2The timeline#
The seven tracks
| Track | Content | A clip can move to |
|---|---|---|
| Main video | Shots, end to end | B-roll |
| B-roll | Shots shown over the main video; its sound continues. An added B-roll is placed at the playhead and muted by default (sound adjustable afterwards) | Main video |
| Voice | Voice-overs, isolated voices | Music, Sound effects |
| Music | Music bed, looped if shorter than the clip | Voice, Sound effects |
| Sound effects | One-off sounds | Voice, Music |
| Text and titles | Text over the picture, centered on its position; may overlap | - |
| Transitions and effects | Effects from automatic editing (motion, grading, transitions) | - |
Working with clips
| Gesture | Effect |
|---|---|
| Drag a clip | Moves it in time, or to a compatible track. It snaps to 0 s, the playhead and other clips' edges, and never lands on another clip of its track: it goes to the nearest free spot. The position shown while dragging is the final one. |
| Edge handles | Trim the clip: the right edge stops at the media's end and the next clip; the left edge moves into the source. |
| Moving a shot or a voice | Moving a shot of the automatic edit takes its effects, special texts and the voice-over placed under a generated shot. Moving a voice takes the caption words it transcribes; hand-written text stays where it is. |
| Split at the playhead | The second half does not replay the shot's entry fade or flash: that transition stays at the start of the first half. |
| Click a clip | Selects it and opens the action bar below it: split at the playhead, duplicate, disable, extract voice (videos), replace media, delete. |
| ← → (Shift: 1 s) | Nudges the selected clip by 0.05 s (or 1 s) |
| ↑ ↓ | Moves the clip to the neighbouring compatible track |
| Delete | Deletes the selected clip |
| Ruler (top) | Touch or drag to place the playhead |
What clips show
Audio clips show their waveform; video shots, thumbnails of their picture; images, the image itself. The right panel sets the selected clip's position, duration, text and volume; typed values follow the same rules as dragging.
4.3Exporting and finding your montages#
| Format | Render size | Use |
|---|---|---|
| 9:16 | 1080 × 1920 | Reels, TikTok, Shorts, stories |
| 16:9 | 1920 × 1080 | YouTube, website, presentations |
| 1:1 | 1080 × 1080 | Instagram feed, LinkedIn, Facebook |
- Rendering happens in the background and resumes tracking on reload. Only active clips count: a disabled clip stays on its track but is neither played nor exported.
- Maximum length of an exported montage: 10 minutes. A video in another format than the montage is fitted without cropping (black bars).
- Exports included per plan, then 1 credit per 5-minute block, refunded if the export fails (chapter 1.5).
- A finished export is filed in the Library as a video (Studio source, with the link to its composition): next to “Download the montage”, the “Publish”, “Schedule” and “View in Library” buttons open from the studio.
- A running export can be canceled (“Cancel export”): it is refunded and does not count among the included exports; neither does a failed one.
- “My montages” keeps every montage and its exports: reopening one restores it with its tracks; “New montage” starts from an empty timeline without touching the canvas.
Reference#
Technical limits, writing best practices, social network formats and answers to frequent questions.
5.1Limits and behaviors#
| Item | Limit | What happens beyond it |
|---|---|---|
| Video prompt text | ≈ 2,000 to 2,500 characters depending on the model | Shortened at a word boundary so the provider does not reject it |
| Prohibited content | Provider policies (violence, nudity, copyright…) | Refused with a message, no charge |
| Reference images | Model capacity (chapter 1.5) | Launch refused before any charge; no further slot opens |
| Clip duration | 3 to 30 s depending on the model | Brought to the nearest allowed duration |
| Storyboard | 8 scenes | Adding scenes is disabled |
| Composition | 300 cards, 1,000 links | Saving refused with a message; split into several compositions |
| Imported file | 500 MB; automatic editing: 3 s to 5 min | Refused before upload, with the reason |
| Exported montage | 10 minutes | Export refused before rendering |
| Provider overloaded | - | A clear message in the card or results area; relaunch whenever you want |
5.2Best practices#
- Specify lighting and ambiance : 'Soft studio rim light', 'golden hour sunlight', 'cinematic moody lighting'.
- Clear typography hierarchy : Specify one bold headline and one short CTA instead of multi-line text blocks.
- Break down video motion : One distinct primary subject motion per 5s scene produces the sharpest physics.
- Use reference images : Upload clean product packshots for high-fidelity asset composition.
- Avoid contradictory adjectives : E.g., 'vintage rustic yet ultra-futuristic neon' confuses the model layout.
- Do not paste huge raw copy : Do not paste multi-page articles into the prompt; focus on visual direction.
- Avoid tiny body text on mobile : Dense small body copy on vertical mobile video degrades visual impact.
- Do not cancel in-flight jobs : In-flight video jobs finish safely in the background even if your tab reloads.
5.3Social network formats#
Instagram Feed
1080×1080 px (square) or 1080×1350 px (portrait)
1:1 | 4:5Instagram Stories & Reels
1080×1920 px (vertical plein écran)
9:16TikTok & YouTube Shorts
1080×1920 px (vidéo dynamique)
9:16Facebook Feed & Pages
1200×630 px (landscape) or 1080×1080 px
1.91:1 | 1:1LinkedIn Post & Bannières
1200×627 px (banner) or 1080×1080 px
1.91:1 | 1:1YouTube Long-form
1920×1080 px (Full HD 16:9)
16:95.4Frequently asked questions#
›I can't see the Node Canvas.
The Node Canvas is open to all plans (including free). It simply requires a suitable screen: a tablet or a computer (screen ≥ 768 × 600 px). On a smartphone, the Studio opens automatically in Light, where the Storyboard remains available.
›Why does my card say “needs refresh”?
One of its inputs changed since it was generated (prompt, reference, camera, format…). The result stays visible; nothing is relaunched or charged until you click Generate.
›Is a canceled or failed generation charged?
No. A video is only charged on delivery; an image, voice, sync or paid export is refunded on failure or cancellation.
›Why isn't 1:1 offered?
The chosen model cannot produce it: Gemini Omni Flash only does 16:9 and 9:16. Pick Kling, Seedance, MiniMax or Wan for a square format.
›A voice won't play when I listen.
On first listen the sample is prepared (a few seconds). If the row says “Sample unavailable”, click again: the button retries.
›Does opening another composition change my montage?
No. The montage and the composition are independent: the timeline, its format and exports stay as they are.
›How do I reuse a graph in another project?
Export it as JSON (bottom bar), then import the file in the other project: it opens as a new composition, with its notes and groups.
Was this documentation helpful?

