Skip to main content
A Creative Brief brings a campaign’s scenes together. Use it to write prompts, select references and choose output settings. Press Generate on the brief card and every scene becomes its own node next to the brief, named after the scene, and runs.

​
When Studio chat makes a brief

Ask Studio chat for two or more images and it prepares a brief instead of generating straight away, even for “a few versions” of one idea. You review the scenes, then generate them from the canvas. A single image, a collage in one image, or one edit per selected canvas image generates directly. To skip the brief, say so: “just generate” or “no brief”. Studio then makes the images straight away. Scenes are independent: each one uses only the references you chose. Images build on each other, with each scene seeing the earlier results, only when you ask for a storyboard or tick Each image builds on the previous on the brief. Questions and creative options are saved with the task they belong to, including replies that contain only a card. Reopening chat restores those choices. If a choice belongs to an earlier task, follow the recovery message or ask for an updated choice; an old answer will not silently change the current brief. If a reply fails, point at the line under it, beside the copy button, and use Copy response link or Report this response to identify the exact turn and its surrounding history. Opening a report does not submit it, and a copied link does not grant access to private chat history.

​
Choose how to make the image

Name a model in your request or choose it in each scene’s output settings. Studio keeps that choice when you refine the brief. Model options depend on the attached references, their count and the requested format; a model with no image input cannot use a product photo or logo reference. Changing a model selects a supported output setting and updates the estimate. For a new direction, ask chat to compare suitable options and their costs. Text-heavy work, scoped editing and cinematic stills can benefit from different models. A recommendation is a starting point: inspect the actual product, lettering and logo in an output before accepting it. Reference support does not guarantee identical geometry. For a layout that must retain the supplied product and logo, explicitly ask for Exact assets. It uses the original image assets, a plain background, an optional accent and the kit’s font for the headline. It keeps the supplied view and lighting; it does not invent a new camera angle or relight the product. The output panel shows Exact assets · 0 image credits and an editable Rendered headline. Chat usage and storage limits still apply. Changing a creative prompt does not redraw these original assets; make an AI scene when you need a generative edit. Connected agents can continue to place original assets in presentations; Exact assets is offered in Studio and presentation chat.

​
Prepare your scenes

  1. Describe the campaign in Studio chat. Mention the brand kit and references you want to use.
  2. Open the brief from its canvas card or the Open brief link in chat.
  3. Choose a scene in the left list. Edit its prompt in the center and its output settings on the right.
  4. Add, reorder or remove scenes using the scene list, and rename the open scene from its title above the prompt. A brief can contain both images and videos.
  5. Choose Back to canvas, then Generate on the connected image or video generator.
If a connected generator was deleted, opening the brief creates the missing connection again. Reopening an already connected brief does not create another generator.

​
Write and refine prompts

Click anywhere in the prompt to edit. Repeated spaces, ordinary keyboard navigation, selection and undo work inside the editor. Changes are saved as you write, including the latest text when you switch scenes or close the brief. Write creative instructions such as “warm lighting” or “negative space” directly in your prompt. The brief has no separate Styles library or automatic attribute popovers. Older saved style selections are preserved as visible prompt text when the brief opens.

​
Choose scene references

Use + below the prompt to open the reference picker, or type @ to choose an available reference. The picker has four sources — Brand Kits (open one to pick its logos, products, and photos), Cinema Kits (open one to pick its cast, locations and props), Inspirations (with your collections as a row above the grid: pick one to see only what you filed there, All to widen it again), and Generations — each with its own search. Every tile shows the item’s name and kind (for example “Image · 16:9”). Tick any number across the tabs, then Add. Each scene’s thumbnail strip shows the references used for that scene. An inline reference is one locked chip. Its label cannot be edited inside the prompt. Use the asset’s own editor to rename the asset.
  • Remove a chip with its X or by selecting it and pressing Backspace or Delete.
  • If the same reference is mentioned twice, deleting one mention keeps the other.
  • Deleting its final mention detaches it from that scene.
  • Removing a thumbnail below the prompt also removes all its inline mentions in that scene.
  • Other scenes keep their own selected references. Removing a scene reference does not delete the original asset from its brand kit.
A color such as Stripe Purple (#533AFD) displays a small swatch. The exact color text remains in the prompt. Swatches do not become additional reference images.

​
Reference purposes

Each thumbnail has a purpose menu below it. Image scenes can mark a reference as Character, Location, Prop or product, Logo, Pose or staging or Style. The purpose is described to the model together with the image. A kit logo is marked Logo on its own, so the model reproduces the mark as given instead of treating it as a style to imitate. Video scenes can also mark a reference as the Start frame or End frame; a scene has one of each. A saved purpose stays with the reference when you reorder thumbnails. A video scene with no purposes uses its first reference as the start frame, and the strip labels that thumbnail Start · auto.

​
References not used in any scene

When you pick references in chat and the agent does not assign them to a scene, the brief lists them under Not used in any scene instead of adding them to every scene. Select one to add it to the current scene, or choose Add all to every scene. The brief card on the canvas shows how many references are still unassigned.

​
From a connected agent

A connected agent (see Connect your agent) can create a brief for a multi-scene ask, read the brief you have open, and edit scenes, references and purposes for you. Changes made while the project is open appear in the editor immediately and can be undone like your own edits. Generation still runs from the connected canvas generators, after you approve the estimate. The estimate comes with a signed receipt. The agent shows you the scenes that will run and their credits; when you approve, it runs exactly that set — a scene that finished in the meantime, or a brief that changed, needs a fresh estimate first, so you are never charged for something other than what you approved. The run answer lists every scene with its state, and the agent follows progress with check_brief, which reports each scene as pending, running, completed or failed with its reason. The agent reads the same per-scene states from the brief itself, and it can retry one failed scene on its own, with the same estimate-then-approve step, without touching the scenes that rendered.

​
Output settings

The Output card beside the prompt shows the scene’s model, aspect ratio, output size and estimated credits. Image size uses the model’s actual options, such as 1K, 2K or 4K; a fixed-size model shows that limitation. When a provider has a separate rendering-effort setting, it appears separately from size. Video scenes add resolution, duration and audio controls. Available options and estimates depend on the selected model. Each scene’s estimate also appears beside it in the Scenes list. Scene settings are what each scene node runs with. A scene that leaves a setting empty follows its node’s default, and the agent is told which fields defaulted. A shared character reference does not by itself make one scene depend on the previous scene’s output. A brief can carry a Brand Kit: when it does, the applicable kit rules guide each scene, including scenes with no kit reference attached. A product photo and a character scene can use different scoped rules from the same kit. The agent keeps that scene intent and its relevant collection scope through saved edits and follow-ups. A brief’s shared creative direction applies to every scene too; it is added to each scene prompt when the scene runs.

​
Generate from the brief card

Generate on the brief card lays every scene out as its own card to the right of the brief and runs them all at once. All of the brief’s cards sit in one group named after the brief. Each card is named after its scene, shows the scene text with an Edit in brief link and one line of settings, and carries its own copy of the prompt and settings, so it stands on its own. No wires are drawn between the brief and its cards. Every output is named after its scene, so it is easy to find again in the library and to reuse as a reference. Scenes that already have a card keep it; pressing Generate again re-runs every scene in place and only adds cards for new scenes, inside the same group. Credit limits and errors show on the card that hit them.

​
Reply yes in the chat

When the Assistant has just shown a brief in the chat, you can type a short yes instead of pressing the button: “yes”, “go”, “generate” or “looks good”. That runs the brief exactly as Generate does, with the same scenes, model and price shown on the brief card. The chat confirms how many scenes started and what they cost. A reply that asks for anything else, such as “yes, but make it 9:16”, goes to the Assistant, which changes the brief first. The canvas displays generation progress, outputs and errors. A failed generation leaves the brief available for correction. Credit limits and workspace permissions still apply.

​
Retry one scene

The brief card keeps score below its scene list, for example 3 of 4 rendered · 1 failed. The reason for a failure shows on the scene’s own node (Not enough credits, Generation failed, Scene has no prompt). Retry failed on the card runs only the failed scenes; the rendered ones keep their results and are not charged again. Generate does the same for every scene that has not rendered yet, so a brief is never generated twice by accident. To change a scene before retrying it, open the brief, edit the scene, then use Retry failed or run that scene’s node directly. Copy link at the top right of the brief editor links to the scene you have open; opening it brings up the brief at that scene. The preview of any image or video on the canvas has Copy link too; opening that link selects the item and zooms to it. Paste either link to a connected agent to work on exactly that scene or file.

​
Older saved briefs

Older chat brief cards now open the same expanded brief. If an older brief contains one document instead of structured scenes, Memo shows Your saved brief. Choose Create image scene or Create video scene to bring that document into one editable scene. Review it and add further scenes as needed. The import preserves the original document under Original saved brief. It does not use AI to reinterpret your saved work or start a generation.

​
Context efficiency

On brief-answer turns, Memo keeps the kit’s asset names and IDs, colors, fonts and rules available without repeatedly sending its images to the chat model. Recent brief details and your latest answer remain in context, including supported corrections from longer conversations. The brief editor groups continuous typing into useful canvas undo steps without postponing the saved text. These changes reduce unnecessary editing work; actual generation costs still depend on the chosen models and outputs.

​
Refine a saved brief

Name the scene and the setting you want to change, including an exact model name when it matters. Model, format, resolution, references and creative direction are separate scene settings. A refinement updates the saved brief in place; check the revised quote before generating. Changing a brief does not authorize generation. For copy or an inventory, say copywriting only or read-only lookup. The assistant answers in the conversation and uses reading tools as needed. It does not create a brief or run generation for that turn. Found references remain available when a second requested collection is missing. Generated artwork still needs review against the original product and logo. Exact geometry, hardware details and letterforms are not guaranteed by a text instruction. Reuse original assets in the editable design when those details must remain exact.