Studio Overview

Studio is where the work gets made. An infinite canvas paired with a conversational agent that carries your brand context, sees your references, and generates images and videos directly onto the canvas.

​

The canvas

An infinite canvas where every generation, reference, and asset lives as a moveable node. The Studio home lists your Projects (each one a canvas) and, beside them, Presentations. When you open the same project in two tabs in one browser, Studio autosaves from one tab at a time. Choose Edit anyway to move editing to the tab you’re using. If you close the editing tab, the remaining tab can resume saving automatically. Press anywhere on a card and drag to move it, including the middle of a picture or a clip. A selected card shows one rounded border; drag a corner or an edge to resize it (the pointer changes there). A card that shows a picture or a clip keeps that picture’s shape while you resize, so no empty bands appear beside it. Add reference (the @ button in the left toolbar) opens the reference picker. With one Image Generator selected, it opens that generator’s own picker, so the pictures you tick attach to it, the same as its +. With nothing selected, the pictures land on the canvas; drag a line from one to a generator to connect it. Generator nodes do the work:

• Image Generator — prompt, model picker, aspect ratio, output size, references, and a Generate button in one node; the shape and the size are set right on the node (the ratio picker draws each shape), and the node itself takes the shape of the picture it makes, so a result always shows whole. Outputs land beside it and can chain into further edits. Once a picture is made it fills the card. Hover it and the prompt and the card’s controls appear over the picture: click the prompt to edit it in place (the picture blurs behind the text), change the model, shape or size, and press Generate to make it again on the same card. The small arrows in the corner of the prompt box open it large over the card and shrink it back. Generating again keeps the earlier pictures: the new one goes first in the row of thumbnails under the card, and the earlier ones stay behind it; to send several results onward at once, tag them in See all results
• Video Generator — model-aware video generation with per-model durations and resolutions. A finished clip fills the card in its own shape, plays without sound while the pointer is over it, and has a sound button and Open preview for the full player. Hover it and its inputs, prompt and controls appear over the clip, as on a picture card: click the prompt to edit it in place, press Generate to make it again, or open the prompt box large with the corner button
• Workflow — a saved multi-scene recipe placed on the canvas, ready to run
• Assistant — one written answer from a text model, right on the canvas: write a prompt, connect text or pictures, pick the model and press Run. Export as text lands the answer in a Text node; Export as list splits it into a List you can feed to a generator, one item each. To make it write a saved way, type / in the prompt and pick a Skill: it sits next to + and your pictures under the prompt, and every run follows its latest version (✕ removes it). A long skill adds a few credits, and the price on Run includes it. It writes; it doesn’t generate, so a picture recipe gives you the prompts, not the pictures. Workflows and Memo’s built-in agent playbooks run in Studio chat instead
• Background Remover — connect any picture (an upload, a brand asset, a generator’s output) and press Remove to get the subject on a transparent background. Pick the model in the node: BiRefNet v2 (1 credit, fast and clean on products and objects) or Pixelcut (3 credits, finest edges on hair and people). The two icons on the cutout open it full size and download it as a transparent PNG. The cutout feeds generators, lists and video start frames like any picture. If you connect a different picture, the node asks you to run it again.

Content and reference nodes feed them:

• Image / Video — generated results with status, playback, and action buttons
• Upload — a picture you dragged in, pasted, uploaded or picked from your library. It shows its upload progress until the file lands, then sits at the same size as a generated picture and fits its own shape; select it to resize from any corner (the shape stays locked), hover for Open, Download (your original file) and the pixel size, and double-click the name above it to rename
• Brand asset — a picture from your Brand Kit, on the same card as an upload
• Text — a text or prompt box on the same card as a generator: a title row (double-click to rename), then the text; drag its output into a generator’s prompt input. Resize from any corner
• Note — a sticky note for instructions and reminders on the board, for example a guide for someone opening the board for the first time. It feeds nothing. Select it to pick its colour (gray, yellow, blue, red or green), duplicate or delete it; resize from any corner
• List — a set of items of one type: images, texts or videos. Connect anything to its input and the list works out the type for you, then fills itself in: a generator adds each of its results, a single image, text or video adds one item, and a Group adds one item per member (its image output for the images, its text output for the texts). A list holds one type at a time, so a source of another type cannot be connected once it has items. It is a snapshot you can curate: delete what you don’t want, and use Pull latest from source to bring in new results. Downstream, As references feeds every item into a single generation. As list depends on the list: on a text list it lays out one generator box per item, each wired from the list — drag from the handle and pick Image Generator or Video Generator, or press Generate on the list to lay out and run them (the button shows the total; rows that already have a result keep it). Each box has its own prompt, model and settings, so you can compare the options side by side and re-run one at a time; a box whose prompt you have not touched follows an edit to its row, and a box you have edited keeps your words. The row’s toggle marks it as a video, and the Assistant does the same when a line starts with [video]. On an image list, As list generates one output per picture inside one box, with one shared prompt
• Description — prompt descriptions attached to generations
• Brief — structured creative briefs with per-deliverable settings
• Group — combine nodes into a named, color-tinted frame (Cmd+G to group, Cmd+Shift+G to ungroup). Click the title to rename it. Selecting a frame shows its toolbar: pick a color, ungroup, arrange members (grid, horizontal, vertical, moodboard, or workflow order — the frame tightens around the result), duplicate the whole frame (Cmd+D), delete it with its members, download every image as a zip, or set a review status. Dragging a member around inside the frame keeps it in the group; dragging it out removes it. A group also has two outputs on its right edge: Members as references wires every image in the group into a generator’s reference slot, and Members as prompt text wires its text notes into the prompt. A wired group’s images show as thumbnails in the generator’s reference row, just like single wired images, and they are all sent. To send only some of them, @-mention those members in the prompt: the others are then left out, and on a Video Generator their thumbnails dim with an @ badge and take no reference slot. Mentioning the group itself, through its (all images) entry, sends every member again

You can arrange, compare variations, build sequences, or lay out a full campaign — all on one canvas. References from your library. Every reference slot on a generator has a +. On a Video Generator the video slots (Reference video and Video references) list the videos saved in your Inspirations and the clips you made in Memo. This is how a motion you saved reaches a model that copies it, such as Genjutsu: pick the model, press + on Reference video, choose the saved video, then add the image it is applied to. Connecting a Video node to the slot works too. A clip on the board. A video made by the agent or the chat plays without sound, in a loop, while it is on screen. Hover it and use the button in its bottom-left corner to pause it, or to play it again. The play button in the middle, or a double-click, opens the clip with sound. There, under the prompt, Inputs shows what the clip was made from: the start image, the end image, the reference video and the reference pictures. Select one to open the file. Clips made before October 2026 show their start image only. Keyboard shortcuts:

• V — switch to the select tool
• H — switch to the pan tool
• Space + drag — pan the canvas without leaving select mode
• Cmd+U — upload pictures to the canvas (up to 50 MB each)
• Cmd+V — paste a copied picture onto the canvas
• Cmd+G — group selected nodes
• Cmd+Z / Cmd+Shift+Z — undo/redo (50-step history)
• Backspace / Delete — delete selected nodes

A full list lives in Keyboard Shortcuts.

​

Starting from a Brand Kit

The quickest way into Studio is Create with this brand at the top of any Brand Kit (on a Cinema Kit it reads Create with this kit). It opens a new project named after the kit and the day, for example Ledger · 26 Oct, with no form to fill in; click the title in the canvas header to rename it. A new, empty project opens around the conversation: the chat is already open, the kit is chosen in the composer and the cursor is in the message box. Under What are you working on for Ledger? there are three ways to start:

• I have an idea — puts the cursor in the message box; describe it in a sentence
• Help me brainstorm — switches the composer to Brainstorm, where Memo explores ideas with you and generates nothing
• Guide me with questions — Memo asks one short question at a time, four at most, then suggests a direction and offers Switch to Create. It runs in Brainstorm, so nothing is generated before you have a brief to check

The card on the canvas says the same: Tell Memo what you need opens the chat (also after you have closed it), and Add an image generator instead puts a generator on the canvas if you prefer to build by hand. A project that already has work in it opens as you left it, with the chat closed until you open it.

​

The chat agent

Open the chat from the little character in the bottom-right corner. The character moves up and settles beside the conversation title. Close Memo chat returns it to the corner. With reduced motion enabled, it changes position without travelling across the canvas. The right panel is your creative director. It’s a conversational agent that can:

• Generate images from your text descriptions
• Generate videos with duration, camera, and motion control
• Search your inspirations and brand assets for relevant references
• Present creative briefs when you ask for a draft, including a single image, and for multi-deliverable campaigns — whether parallel (independent images) or sequential (a storyboard that chains each scene onto the previous one). See Creative Briefs for the sequential toggle
• Save work as brand assets — feed generation results back into your Brand Kit

The agent keeps the current task’s goal, choices, references and selected skill across follow-ups and reloads. It uses recent conversation plus a compact task record, and can retrieve earlier details when needed. An unrelated request starts a new task; changing the Brand Kit prevents the previous brand’s task from carrying over.

​

Choosing your Brand Kit and references

Choose your Brand kit beside Create or Brainstorm. Memo remembers it for the conversation and uses its brand-wide rules. When you attach a kit asset, the rules for its collections apply automatically to generations using that asset. You do not need to select a separate collection mode or review a context panel. You can also name a collection naturally: “Use our Summer launch references for three outdoor ads.” Memo looks for it within the selected Brand Kit, applies the chosen collection’s guidance and selects relevant references. It asks when plausible alternatives would change the result. Naming a collection does not attach every image. In a large collection, Memo searches for relevant names, handles and descriptions instead of loading the entire library. Skills are selected for the task from the workspace and chosen kits. You can also choose one directly with /. Attaching a collection’s asset applies its relevant rules; it does not run every skill in the kit.

​

Following up

“Make the background warmer” keeps the active task’s references and settings, even in a long message. “Write a poem” starts a different task, even though it is short. New attachments replace inherited task references by default; say “add these to the existing references” to combine them. Removing a reference or changing a previous preference updates the task. In longer chats, Memo can recover earlier instructions and the current saved brief, including edits you made to its scenes. If a detail cannot be recovered, it should say so rather than inventing it. A requested draft stays editable on the canvas and does not start generation. When revising an existing brief, Memo reads its current saved scenes along with the selected collection’s reference candidates. It applies small changes to the saved scenes and keeps the unedited text, formats and references. You do not need to repeat your canvas edits in the chat. If the quoted text has changed or is ambiguous, Memo rereads the brief before trying again. Memo asks about missing inputs, conflicting requirements or meaningful creative alternatives. It reuses quality, format and other choices you have already settled. Selecting a skill or collection does not approve generation or credits. A choice card from an earlier task may need to be replaced before it can be used.

​

Image prompts

The image model receives the scene, the role of each chosen reference and the applicable brand constraints. Chat history, skill-selection notes and internal planning instructions stay in the chat agent. Exact requested label text and necessary visual details are kept. An internal campaign name labels the brief; it only becomes text in the image when you ask for that. For an edit, describe what should change and what must stay. If a mandatory rule conflicts with that preservation request, Memo should ask about the specific conflict before preparing the edit. For example, a rule that always requires a gradient may conflict with keeping a photograph’s facade unchanged. Resolve the creative request or update the rule first; a chat instruction does not silently disable a mandatory rule.

​

Choosing an image model

Memo considers the model’s supported inputs, visible text, format and reference needs before cost. Some models accept references only for style; those inputs do not promise to preserve a product or person. A model accepting image edits still cannot guarantee exact lettering, logos or geometry. Your selected model and a skill’s model settings stay in effect. Memo can explain a relevant alternative, but routine follow-ups do not need another model-selection question. Use the generation quote for the final credit price; quality and additional references can change it. Exact brand typography or an unchanged logo may need a separate design step.

​

@mentioning brand assets

Type @ in the chat to mention a brand kit or specific asset. The agent loads your colors, typography, guidelines, and reference images as creative constraints. Mentions appear as styled chips with thumbnails in your message.

​

Attaching a video

Type @ and pick a saved video from Inspirations, or a clip from Generations. What the agent does with it depends on the video model selected in the chat:

• Genjutsu, Genjutsu Object Swap and Kling 3.0 Motion Control work from a video: the one you attached becomes their reference video. They also need the image the motion is applied to. That is the image selected on the canvas, a character or product you mention, or an image attached the same way.
• Seedance and Wan 3.0 use the attached video as a motion reference, when the clip does not start from an image.
• Other models take no video. The agent tells you the video was left out.

A clip already on the canvas attaches the way an image does: select it, and it appears above the composer with a play mark. The agent gets the clip’s first frame and a short account of what happens in it (the action, the camera move, the pacing, whether there is sound), written once when the clip finished rendering. Clips made before 13 October 2026 get that account the first time you attach them, which can take a few seconds. The clip itself never goes to the chat model, so to work from its motion keep one of the video models above selected. A YouTube or Vimeo link, or a TikTok saved as a link, has no file behind it, so it cannot be used. The chat has no button to attach a file from your computer. To use one, upload it in the Video tool, or save it to your Inspirations first (with the extension, or through your connected agent) and attach it with @.

​

Streaming

Memo’s mark anchors each reply. Text arrives at a steady reading pace, with headings, lists, links and tables. Use Expand chat for a wider reading view and Restore chat width to return. New replies show a work trail in the order the updates arrive: short status labels, commentary and tool outcomes. It opens while Memo works and folds into Worked for …s when the reply finishes. Open it to revisit the steps; if you choose to keep it open, it stays open. The trail is saved with the conversation. Older replies keep their previous reasoning and tool disclosures. You can draft your next message while Memo works. Sending waits until the current request finishes. Scroll up without being pulled back down, then use Jump to latest to follow again. Stop keeps received text and leaves a stopped notice. Work already submitted may still finish, so check the canvas before trying again. Finished and stopped text can be copied. Reduced-motion preferences disable animated streaming effects. Open a tool step for its available details and results. Queued or unconfirmed work stays distinct from completed work; failures are marked explicitly. Reference previews and brief controls remain available. Credit limits and chat billing are unchanged.

​

Creative plans

Creative choices appear in an expandable Creative direction card. Each choice has a short heading and a full description, so you can scan the subject, composition, lighting and style while keeping the references and output settings in view. These are proposed choices, not completed actions. For storyboard generations, Generation plan shows the progress reported by each frame’s generation. Queued work stays pending, failures remain visible, and interrupted updates stop animating. Successful plans collapse when finished; open the header to review them again.

​

Running a whole workflow

Hover a card a workflow starts from to see Run workflow above it.

• Run workflow appears on a card a workflow starts from: an uploaded picture, a text, or a generator nothing else feeds. It runs every card connected to it.
• To run one card alone, use its own Generate (or Run) button, on the card.
• To make one card again together with everything after it, press Generate on that card, then Run workflow: the cards after it are out of date, so they run.

Before anything runs, a sheet lists every card in order, what will happen to it and its price, with the total and your balance. Nothing runs until you press Run. Cards that already have a result are kept and cost nothing; turn on Make everything again to remake them. When a card runs again, the cards after it run again too, so they use the new picture. If the total is more than your balance, the sheet offers Get more credits instead. Cards run in order: a card starts when every card feeding it is done, and cards on separate branches run side by side. A List in the middle of a workflow takes in its source’s new results before the cards after it run, and an Assistant writes its rows into a List it feeds. A bar at the bottom shows progress and a Stop button; Stop lets the cards already running finish and starts nothing new. If a card fails, only the cards that need it stop. They are marked Not run, say which card did not finish, and are not charged. The other branches finish. Press Retry on the bar to open the sheet again and run what is missing. Keep the tab open while a workflow runs: the open tab drives the run. If you close or reload it, finished cards keep their results, a picture or video that was already being made still lands on its card when you come back, and Run workflow picks up the rest without paying for anything twice. Run workflow covers Image Generator, Video Generator, Assistant and Background remover cards.

​

Asking the chat to set up nodes

Ask for a setup you want to run yourself, for example “wire up a text prompt into an image generator, that picture into an Assistant that writes six camera angles, the Assistant into a List, and the List into a generator”. The chat places the nodes on the canvas, connects them, and stops: nothing is generated and no generation credits are spent until you press Generate or Run yourself. It can place Text, Assistant, List, Image Generator and Video Generator nodes, attach brand assets or canvas pictures as references, and give a List its rows when it already knows them. It connects them the way you would by hand: a prompt into a generator or an Assistant, a picture into a generator or an Assistant, an Assistant or a generator into a List, and a List into a generator one row at a time. A List holds one type, so the chat will not mix prompts and pictures in the same List. The reply ends by saying what is on the canvas. The chat builds a new setup beside your existing work; it does not yet change, rewire or delete nodes that are already there, and it does not place Note, Background Remover, Scene Builder or Workflow nodes. Director and Colour nodes have been removed. A board that still holds one opens with it converted: a Director becomes a Text node with its prompt (or a List when it held several takes), and a Colour node becomes a Note listing its palette. To use brand colours in a generation, pick them from Brand Assets or type them into the prompt.

​

Projects

Each Studio canvas is a project — its own canvas, its own chat history, its own references. Projects save automatically (2-second debounce on every change, plus a save on unmount). If your workspace has no Studio projects yet, Studio opens on a focused first-run state with an animated canvas illustration and a New Project action. Search appears after the first project exists, when it has something to act on. From the Studio tab on your dashboard you can:

• Create new projects
• Open a project’s assets (card menu → Assets) — everything generated in it, filed into folders
• Rename projects (card menu, or click the title in the canvas header)
• Search across all your projects
• Delete projects — the project moves to the Trash with its pictures and chat, and Undo brings it straight back

Use the List / Grid switch beside Studio search to choose landscape previews or compact rows. Each row shows the project name and last-edited date, with the same Assets, Rename, and Delete menu actions available for your role. Memo remembers Studio’s layout independently of other libraries. Search works in both views. Project cards show a preview of the project’s latest generations (up to four, as a mosaic), with the project name and last-edited date below. Select anywhere on the card to open the project; its menu holds Assets, Rename, and Delete. Projects are paginated for large collections.

​

Project assets

Project Assets opens in List by default. Use List / Grid beside search to choose your layout when available; an existing saved choice is remembered. List uses the same rows as Collections and Brand Kits: open a folder to browse its files, or open a file to expand it. Each file row keeps to its thumbnail, name, type and date, with Move to folder, Download and a menu holding Rename and Delete beside it. The layout remains selected as you enter folders, return to the project’s assets and reload the page. Every project has an assets page — the folder icon in the canvas header, or Assets in the card menu. Its back arrow returns you to the canvas (inside a folder, to the project’s files). It works like a drive:

• Folders. A project starts with no folders. Add your own with New folder (for example “Posters” or “Scene 3 — Kitchen”); rename or delete them from the folder’s menu. A folder can only be deleted once it is empty, so files are never lost with it. A project holds up to 100 folders; past that, Memo asks you to reuse or tidy one before adding another.
• Say what belongs in a folder. Open Details from a folder’s menu and write a line about what goes in it and what to keep consistent (for example “Final A3 posters for the autumn run, plum and cream palette”). Your connected agent reads this line whenever it looks at the project, so it files new work in the right place and takes references from the right folder. The line also shows under the folder’s name in the list.
• Files. Everything you generate in the project lands here, unfiled, the moment it completes. Drag a file onto a folder to file it, or hover it and choose Move to folder; a file lives in one folder at a time. The Unfiled section only shows while something is in it; once every file has a folder, the page is just your folders. Search looks across every folder in the project.
• On the canvas. The project root also shows uploaded pictures, Asset references, outputs referenced from other projects, media in Lists, cutouts and rendered Scene Builder videos. Repeated copies of the same file appear once. Search and the status filter include these assets in both Grid and List. Open a card to preview and download it; references keep their source ownership and are managed on the canvas. Scene videos marked saved in this browser need that browser’s local file: if it is missing, render the shot again in Studio. A scene with no rendered video yet has no video file to show.
• File several at once. Tick the box that appears on a file to select it; shift-click selects a run, Cmd/Ctrl-click adds one, Cmd/Ctrl+A selects everything loaded. Dragging any selected file carries the whole selection onto a folder (the drag shows how many), or use Move to folder in the bar that appears. Press Escape or Clear to deselect.
• Every move can be undone. Files leave the view the moment you drop them, and the confirmation offers Undo for a few seconds, which sends each file back where it came from.
• Inside a folder, the other folders stay in reach. They show as a row above the files, so a file can go straight from one folder to another, and the project name in the heading takes files out of any folder.
• Tidy up. Memo knows what each picture shows. When unfiled files look like they belong in one of your folders (a “poster” in Posters, a “character” in Characters, a “kitchen” in Scene 3 — Kitchen), a line above the files says so, and File them files the lot in one go, with Undo.
• Moving between projects. Files moved to another project appear in Unfiled unless you choose a folder in the destination. Moving within the same project keeps their folder. Files carrying an old folder that no longer exists also appear in Unfiled.
• Edits survive background work. Filing or approving a file while Memo finishes saving or describing it keeps those changes. An automatic title fills an empty name; it does not replace a name you have already entered.
• Views. Switch between the grid and a list with the toggle next to the search box.
• From the canvas. Select a group frame and choose File into folder from its menu: a folder named after the group appears in the project’s assets with the group’s images inside (an existing folder with that name is reused). Files stay unfiled until you do this.
• Review status travels with the file. The status you set on a canvas image or group (Approved, Pending, Rejected, Needs Review, In Progress) shows as a coloured dot beside the file here, and the Status menu next to the view toggle narrows the page to one status. It is a tag, not a gate: files without a status stay fully usable.
• How it was made. Open any file and it expands into the same view a card or a clip opens on the canvas: the prompt as you would write it, Inputs with the pictures it was made from, and a Details card with the model, aspect ratio and quality it was generated at — 1K / 2K / 4K for images, 720p / 1080p for videos — plus the clip length for videos. The row keeps the date, and the name is edited from the row’s menu, so the expanded view reads the same here as in Studio.
• Versions. A corrected file can replace an earlier one instead of sitting beside it. From a file’s menu in the list, choose Replace with… and pick the corrected image or clip: it lands in the same project and folder with the original’s name, and the original leaves the list, marked Replaced by the new file. Your connected agent does the same when it files a new version with the original named as the one it replaces. Show earlier versions next to the view toggle brings replaced files back.
• Final. Choose Mark as final from a file’s menu and it shows a Final badge in the list and in its expanded view, so the final set is simply every file marked Final; Unmark final takes it back. A version chain has one final: marking a file clears the badge on the version it replaced and on the one that replaced it. The approved newest version of a chain also reads as Final without marking. The agent marks with update_generation and final: true.
• What the picture shows. When an image finishes, Memo looks at the result and writes down what is actually in it — the kind of image, the subjects and a one-line description. Search matches it, and unfiled files get a hint such as “Looks like a character” to help you file them.
• Conflicts are visible. If a file or folder changed since you read it, Memo asks you to review the latest version. Review badges change after the save succeeds. Reopening a canvas uses the latest saved file verdicts and authored names, including files outside the recent-generation feed. Saving canvas layout uses its version, so an older snapshot cannot replace a newer one.
• Characters keep their identity. When you move a file into a folder named Characters, Memo asks which character it is: pick an existing Character to add the image to its sheet, or create a new one. Memo shows progress while saving a permanent sheet. A failed copy can be retried; a changed Character link is shown before you choose again. The sheet keeps its source record and survives deleting the original generation. The agent can ask for the Character by name in any project.
• Everything else you could do in the old Generations tab is still here: open, download, delete and Add to collection.

Your connected agent reads the board in one call with get_project: every folder with its description, counts, what its pictures mostly show, its newest files and when it last changed, plus the Creative Briefs on the board and a page of the newest files. It reads one folder’s files, or only what changed since its last visit, with list_project_files; searches by meaning inside a folder or asks for “more like this” with search_generations; creates folders (with a description) with create_project_folder; and files outputs with move_generations and a folder_id. It can edit file details through update_generation; see Connect Your Agent for revisions, full details and run records.

​

Sending inspirations to Studio

From your library, right-click any inspiration and select Send to Studio (or use the action button in the detail modal). Memo creates a new project pre-loaded with that inspiration on the canvas — ready for the agent to use as a reference.

​

Filing generations into collections

In a project’s assets page, hover any completed image or video and choose Add to collection. Memo copies the result into your library and files it in the collection you pick, so the reference stays available even if you later delete the project. Each generation maps to one library item — no duplicates when you file it in several places. See Collections.

​

Cost

Image generation, video generation, background removal and agent chat all consume credits. The ring around your account avatar shows available credits. Open the account menu for your balance, usage and credits held by pending generations. When credits are low, pre-action warnings appear. When exhausted, an upgrade prompt. The system tracks cost per project, so you can see exactly what each campaign is costing.

​

Mobile

Studio requires a desktop-sized screen. On mobile or tablets (under 900px or coarse pointer), you’ll see an interstitial with a link back to the home grid. If the project is deleted or is no longer available in your current workspace, image and video generation stop before holding credits or submitting the request. Reopen a project in that workspace and try again. If Memo cannot check the project, wait and retry; it will not guess a different destination.

​

Recovering a chat turn

Question and option cards are saved with the task they belong to, including replies that contain only a card. Reopening the chat restores the choices. If a choice belongs to an earlier task or a different Brand Kit, the response explains how to continue; use the latest choice or send a new message. For a temporary connection or service failure, Retry checks the same message with its original references and settings. It keeps one copy of your message in the conversation. If that turn already started, Memo reports its saved or uncertain outcome without running it again. Check the chat and canvas, then send a new message when the recovery guidance asks you to. A response that could not be saved is reported explicitly. Tool cards distinguish matching references, empty searches and failed searches. References without a visual preview show a placeholder instead of a broken image.

​

Share the exact chat for diagnosis

Choose Copy chat link in the chat header, or point at the line under a finished reply, beside its copy button, and choose Copy response link. A response link identifies the exact turn, including older turns outside the latest history page. Share that link with your connected Memo agent to inspect the surrounding history and safe diagnostic facts: which tools were offered, request sizes, usage when reported, and whether the response was saved. The link grants no access; the reader must already be authorized in the workspace, and private history remains visible only to its owner. Report this response opens the existing Support form with the exact chat address attached. Add what went wrong and submit when ready. Copying a link or opening the form sends no report automatically. Design chat keeps a separate private history for each canvas source. Assistant runs remain single requests: their response link identifies a receipt, and a completed model response alone does not confirm that the result was saved on the canvas. Interrupted text remains visible so you can inspect the result before starting another request.

​

Scene Builder (Beta)

Open a Studio project and choose Add node → Scene Builder. The builder is a native canvas node: it moves, saves and reloads with the project and participates in canvas Undo and Duplicate. Drag any empty area in its header, above the viewport or along the footer to move it. Buttons, the timeline, camera controls and 3D viewport keep their own interactions. If this browser has an older draft for that project, the action imports a copy once and keeps the original browser draft untouched. Repeating the action focuses that imported builder. Add another builder creates a separate empty sequence; multiple builders can coexist. A new builder starts with an empty 3D stage. New scene clears only that builder. Street template and Doorway study load optional examples with Undo available. Expand editor gives the controls more room. Each shot automatically gets its own connected Video output node. Moving an output, renaming it or reordering shots preserves its identity and connections. Removing its shot leaves the output labelled Source removed; deleting that output cleans it up, and Undo restores deleted work. Build with chat. Selecting a builder adds a compact Scene Builder · N shots chip to the composer. Click it to focus the builder, or remove it to return to ordinary Studio chat without deleting the scene. Choose a shot thumbnail or This shot to target that shot; Sequence targets the whole scene. Describe the environment, subjects, action and shots, review the proposal, then choose Apply scene edit. Proposals are bound to the selected builder and revision: a late reply cannot modify a different builder or overwrite manual changes. Connect a Text node to the builder’s direction input to include its text in the next chat request. Image inputs retain reference metadata; automatic image reconstruction and image inspection are unavailable in this local increment, so describe the reference’s appearance in chat. Connecting or changing references alone never regenerates geometry. Local diagnostics. In a local Studio project, Save diagnostics in the builder footer captures the current scene, playhead and view for local inspection. It stops playback, leaves geometry unchanged and uses no AI credits. It does not share the scene publicly or sync the browser draft to another device. The latest capture replaces the previous diagnostic snapshot for that project. A valid project and editor permission are required; the standalone preview disables this action. Complete sequences. An explicit numbered shot/scene brief retains its requested count and order through validation and repair. A four-beat story cannot be accepted as only its introduction. For supported numbered action beats, checks sample actual object movement and requested camera travel during that shot; naming a shot “Follow the key” does not make a static camera pass. If a complete proposal still fails after two corrections, the stage stays unchanged and the failure is reported. These checks prevent incomplete proposals from being applied; they do not guarantee that AI can solve every requested set or action. Complex staging can still be refused. Supported walking beats are checked for excessive speed, stated stops must hold the subject, and locked shots must hold both camera position and aim. Scene planning works across environments. AI chooses the set geometry, subjects, required props, lenses, shot durations and moves. The builder calculates camera distances from the actual geometry, including moving subjects, and searches for clear viewpoints without changing the set or lens. Streets, landscapes and product sets use this same planner rather than environment-specific templates. A tracking shot follows both the subject and its composed aim; a locked shot keeps its camera and aim fixed. Explicitly fixed viewpoints remain fixed, and an impossible view is reported for revision. Ground and surface placement constraints seat actors and props using their actual bounds, including stacked objects such as a product on a plinth. A static actor’s hand placement can establish an initial held-prop pose outside the hand. Its release path remains editable, with the initial waypoints retargeted to the held pose and the final destination kept in world space. Flat, static supports are supported; moving platforms, moving-actor attachments, articulated grasp/release, curved-terrain contact and physics are not. For writing-room stories, a parametric interior additionally calculates the doorway, desk placement and action cues for a seated writer, letter and candle. Both paths return ordinary editable geometry for Apply without loading an example template. The prototype uses rough shapes and recipes for buildings, trees, people, cars, cyclists and sheep, plus a hinged door, seated writer, candle and small ring-and-shaft key. The door can open on a timed hinge, the writer has a simple writing gesture, and the candle can fade on with a warm light. A person following a horizontal path has a basic distance-driven arm and leg walk cycle, settling during holds. These are rough action recipes. Objects have editable positions, dimensions, colors, orientation and optional timed paths with several positions. For new sequences, chat can assign an object’s destination and intermediate positions to each shot; the builder calculates sequence timing and holds the preceding pose between actions. They can hold before release, move through later beats, then stop or drift. An open drain recipe supplies four shaft walls with no cap, and its mouth stays at ground level when seated. Water can rise as an opaque surface proxy; this is not fluid simulation. Ask chat to add or change an object while preserving the rest. New scenes support up to eight shots, 100 objects and two minutes of sequence time. This is previsualization geometry; imported models and photorealistic scene generation are not supported. Custom sets can extend underground, and camera height can go below the world origin. Shafts and holes require actual openings between floor/wall pieces; the default ground plane is solid and cannot have a hole. Clearance checks still apply to solid geometry. Water surfaces allow a water-level lens while continuing to block opaque views through the water. The original street template retains its above-ground camera limits.

• Rig shows the set, filming camera, camera path, selectable gold keyframe markers, and a live lens preview. Drag to orbit, scroll to zoom, and right-drag to pan.
• Lens shows the filming camera in a 16:9 frame. Change its focal length, position, target, keys and easing. Follow cameras can track an animated subject.
• Play runs the sequence or selected shot. Scrub to any time and add camera keys. The original street template also has cyclist speed/lane controls.
• Camera moves provides Locked off, Pan, Tilt, Dolly in/out, Truck, Crane up, Track subject and Orbit. These replace the selected shot’s motion with editable keys.
• Undo, Redo, Download scene, Load scene, and Save frame support local iteration. Scene files contain editable JSON; saved frames are PNG images.

Cinematography and framing. Scene chat automatically uses a directing guide for shot purpose, subject, physical scale, framing and movement. Named moves start from the current camera; a dolly retains its lens and an already authored approach path. Smooth paths carry motion through intermediate keys. Camera checks samples each shot for camera clearance, subject visibility, safe framing, cropping and movement pace. A tiny subject cannot pass as a close-up or detail shot; distant background landmarks should not force a close-up to become a wide view. Shot purpose and reveal timing appear beside the results. New shot plans must identify a subject. Invalid plans receive up to two bounded corrections, each shown the reason the last draft was refused; a still-invalid plan leaves your scene unchanged and explains the obstruction. Review notes also appear in the chat proposal. Fit subject derives a camera distance from the subject bounds and lens without moving the set; Undo restores the previous view. If that angle is blocked, ask for another angle. These are sampled checks, not a guarantee of cinematic quality or continuous collision detection. They check horizontal and vertical framing, clipping range, the primary subject and declared required props. Required props can specify a visible anchor or their complete bounds; a whole group member or product must fit when full coverage is requested. Distant landscape landmarks can use an anchor without forcing the entire terrain into a close frame. The interior composer requires the writer’s hand, letter and candle in its medium writing shot. Undeclared props are not guaranteed visible. Tracking deliberately keeps the subject steady in frame. If no static landmark enters the view, a motion-readability note recommends reviewing a path or nearby scenery so the travel can be seen. This note is advisory; an empty background can be intentional. Review the full shot in Lens, and inspect the path in Rig. Intentional reveals can defer subject visibility checks, but never camera-clearance checks. Mirrors, focus pulls and lifelike character acting are not implemented. For a targeted revision, select a shot and ask, for example, “Crane up over six seconds, keep 35 mm” or “Add two trees on the far bank, leave the sheep and other shots unchanged.” Review the proposed changes and apply them. A stale proposal cannot overwrite newer manual scene edits. Clear returns the composer to ordinary Studio chat; Focus scene brings the editor beside the chat. Scene documents save in the normal Studio canvas; pending proposals remain browser-local and specific to a builder. New chat messages and replies save to their Studio conversation. Download scene JSON for an independent editable copy. Find or restore output cards. Use the action above the builder: Create video outputs creates one card per shot when none exist; Restore video outputs fills in deleted cards; Show video outputs brings existing cards into view. It preserves surviving cards, their positions and connections. With an empty scene, add shots first. Creating cards is undoable and does not render video or spend credits. Local shot videos. Choose Render shot on an output, or Render all shots beside its builder. Rendering evaluates the Lens camera at fixed global sequence times and produces a silent 1280 × 720, 30 fps H.264 MP4 without AI credits. A four-second shot has 120 frames. Rendering has progress and Cancel; unsupported encoders and browser-storage failures report an error. Each output distinguishes Not rendered, Rendering, Ready, Needs update, Failed and Source removed. Editing during a render cannot publish the older revision as current. A failed render retains the last good preview, labelled as a previous version. Shot outputs use compact, named poster cards with an exact 16:9 frame. Drag the poster or name to move a card; its video socket connects to a Video Generator video input. The play button opens Preview, the standard Studio video viewer with Download video. Bytes live in this browser’s storage; graph data stores artifact IDs and metadata, with no temporary blob URL. Reload restores the preview in the same browser. Clearing browser storage removes those bytes; render again if a clip is missing. The files do not sync to another device. A clip can connect to a video input supported by the destination model, and cannot connect to its image inputs. The generator’s Video references or Reference video row shows the connected shot’s thumbnail, name and local render status. Disconnect it with the row’s remove control; separate reference images stay attached. An unrendered shot remains labelled as such. When you press Generate, a current rendered clip uploads through the normal workspace upload flow before the video request. The image references stay separate and inline tags bind to the matching image or video. Missing or outdated clips must be rendered again; permission, quota and network failures stop generation with an explanation. An upload is reused for the same immutable render in the same workspace during the current session. Connecting nodes and rendering never starts an AI video generation. This is a local development feature. MCP control, imported 3D assets, and cross-device rendered media are pending. Studio requires a desktop-size viewport. In local development, the monthly chat-budget ceiling is disabled for testing. Authentication, workspace permission checks, credit availability, rate limits, usage accounting and ordinary model charges still apply. An invalid model draft may receive up to two corrections; every text-model call is metered. Customer production budgets are unchanged. Validation errors now identify the rejected field; no image/video request is sent when local validation fails. Incomplete, failed, and unconfirmed work stays expanded so you can see what needs attention. You can close it yourself. Copy appears after the response has finished displaying. Sending a new message returns you to the latest reply, even if you were reading older messages. When you send a message, Memo acknowledges it immediately in the conversation. A moving stroke around the Memo icon indicates an active response; the same reply updates as work progresses. The canvas launcher uses the larger face icon. Canvas edits save automatically, including when leaving the canvas. If a save has already completed with identical content, Memo can resume from that saved version. When the saved canvas contains different edits, saving pauses and Reload latest loads that version. This can also happen after an agent edits the saved project; it does not necessarily mean you opened another browser tab.

​

Sources for the assistant

In Studio the composer chooses kits only: the Kits control selects the Brand Kits and CinemaKits the assistant works with. To build from another project, a presentation or a library collection, use the chat in Presentations, whose composer has Kits & projects. In Studio you can still name a project or paste a Memo link in your message; the assistant then reads small relevant sections of it rather than loading the whole library, within your workspace access. Private chats must be yours, and using their content in shared outputs requires the composer permission. Activity details open only when you click them and retain your choice during the browser session. The monthly chat limit draws from the same workspace credit balance.