RUPA, the image studio
How RUPA renders images to a preset with your brand attached, when a render waits for your approval, what it costs, and where the finished plates are kept.
Browse documentation
RUPA makes the pictures, and your brand travels with every one
RUPA is the image teammate in your Content studio. You describe the image you want, pick a format if you need one, and she renders a sheet of plates with your workspace's brand guidelines attached. Then she reads your brief back against the brand book, so you can see what she was told before you use what she made.
The studio is labelled Beta in its header: "New, and rough in places. The Edit and Write modes are not built yet." Everything below works today unless it says otherwise.
RUPA's page has five tabs: Today, Review, Studio, Library and Settings. Open the studio from the Studio tab, or go straight to nagent.ai/admin/rupa/studio. You need permission to write content to render. The finished images live in the Library tab beside it, the Asset Library. In the Content team's room, the studio opens inside the room itself; the Asset Library still opens as its own page.
The studio, left to right
The studio has a rail of icons on the left (presets, skills, references, brand kit), the thread and composer in the middle, and the output panel on the right: the preset, size, plates, quality and autonomy, and what the brand kit sends.
A new studio says "make something worth keeping." and "Describe an image, or start from a format. The cost is shown before anything renders." That is the whole first step.

1. Pick a preset, or leave it on No preset
Presets are set formats. There are 22, in seven groups: Freeform, Paid social, Organic social, Profile and brand, Product, Life-cycle and Editorial. A preset fixes the size and how many plates you get, so a Meta ad or a LinkedIn post comes out at the right shape without you working it out.
The default is No preset: any size, one to four plates. Three sizes exist today: square, wide and tall. There is no 9:16 or custom crop yet.
Your workspace can make presets of its own, shown first under Your presets with a Beta tag. See "Your own presets" below.
Where Nagent has curated a sample for a preset, the tile shows it, so you can see the kind of image the format is for.
2. Write the brief
Type in the composer and press Send. Enter sends; Shift and Enter starts a new line. A brief can run to 4,000 characters.
It is the same composer the team room uses. A thin strip above it holds the mode, the preset, a one line summary of the settings and the plate count. Make renders. Ask answers from what is already loaded in the studio, with no render and no cost.
Click the summary to open Settings. On No preset it holds the shape and the number of plates, one to four; every preset holds quality (Standard or HD), Brand Lock and a style note under Advanced. Use for this one changes this brief only. Remember for the preset keeps the settings for everyone in the workspace, after you approve a confirm card.
3. Add a reference, if you want a look borrowed
References are images whose look a render borrows. Press the paperclip in the composer to open Pick from the library. Its tabs are All, References, Knowledge Hub, Renders and Uploads, so one sheet holds the Knowledge Hub's images, RUPA's renders, your uploads and the references. The box above the grid, Search what is in the images, searches what each image shows, not just its name: RUPA has described every image in the library, and the search reads those descriptions (see "What RUPA knows about each image" below). Hover a tile to read its description. Pick a reference and, above the grid, say what to take from it. There are five choices, and each one shows what it Keeps and what it Drops:
- Match the style
- Reproduce the layout
- Change the headline
- Localise it
- Remix it
You can add a Style note in plain words as well.
4. Attach what must appear
Attachments are images the render must include, such as your product shot or your logo. Pick one in the same library with As an input image. A render takes up to ten input images.
5. Send a product as it is
If your workspace has active offerings with product images, Settings holds a Product field: No product, or one of your products. Its images go with the render as references, the primary one first, with the instruction to keep the product exactly as shown. They count towards the ten input images. An offering that has no images yet is listed with "No images yet; add one to the offering to use it here." Product images are added on the offering itself; see products and services.
In a team room, RUPA takes a product the same way: ask her for a picture of one of your products and she looks it up among your active offerings and sends its images as references. A product with no images is refused in words, and a product cannot be combined with a skill such as Cut out.
Your own presets
A preset of your own adds words, references and default settings to one of the built in presets, whose size and price it keeps. Open the presets, choose New preset, and give it a name, the base preset, the instructions every brief made with it should follow, a line on when to use it, and up to eight references. Save shows a confirm card, which you can approve yourself; nothing is saved before that. The person who made a preset, or an admin, can change or retire it. RUPA uses your presets in team rooms too.
Make an infographic
An infographic is drawn, not painted. Type /infographic in the composer,
then what it should show, for example /infographic our 2025 results, and
press Send. The command is also in the rail under Skills. The
infographic is labelled Beta, like the studio.
RUPA reads your knowledge base and your brief, picks out the numbers, and writes down where each one came from. The studio then lays them out in your brand colours and fonts as one tall 1080 by 1350 image, in one of four layouts: big numbers, bars, a timeline or a side by side comparison.
- No image model draws a number. Every figure on the infographic is text set by the studio, exactly as RUPA wrote it, so it cannot come back misspelled or changed.
- A number without a source is flagged, not invented. It carries a Source needed mark on the image, and the card above it lists every number that has no source. Fix those, or leave them out, before it goes anywhere. Numbers with a source carry a small reference, and the sources are listed at the foot of the image.
- Export SVG downloads the vector file. Export PNG draws it in your
browser at twice the size. Save to library files the SVG in the Asset
Library, tagged
infographic.
It costs one short text call to RUPA's model, counted against RUPA's budget like her renders. No image is rendered, so nothing waits for an approval. If RUPA's budget is used up, the studio says so and nothing is drawn.
Today the studio draws only what RUPA writes: you cannot edit a number on the card. Change the brief, or the knowledge it reads, and send again.
Plan a product shoot
Type /shotlist and what the shoot is for, for example "/shotlist autumn launch for the leather tote". Pick the product in Settings first, so its own photos go with every shot and it comes out exactly as it is.
RUPA answers with a shot list: up to eight shots, each a packshot, lifestyle, flat lay, scene or mood image, with the scene, the camera angle and the props. Planning is one text call against RUPA's budget. Nothing is rendered yet.
Each shot has its own Render. It renders one plate on the shot's own preset (Packshot on sweep, Lifestyle scene, Flat lay, Styled scene or Mood image), with the product as the reference. Brand Lock is always on for a shot, so the image style in your brand book is the shoot's style; the card says which style that is, or that none is on file. Each render shows its own plan card and cost as any render does, and a shot you have sent is marked so it cannot go twice.
Every shot files in the Asset Library with the tags product-shoot and the shoot's own id. A team room can ask RUPA to plan a shoot too: that files one approval card for the plan, and each shot she renders afterwards waits for its own approval. The Content room's Product shoot workflow does both steps.
Make a batch of variants
A batch is one image template run over many rows: a launch across forty
products, a sale across twelve cities. It is Beta. Type /variants,
then on the first line the prompt with a {{column}} for each thing that
changes, then your rows as CSV with a header line. The first column is
each row's key, which every output keeps.
/variants A {{product}} on a sunlit oak table, {{colour}} accents
sku,product,colour
A1,ceramic mug,teal
A2,linen napkin,ochre
Nothing renders when you send it. RUPA checks the paste, makes a draft batch and shows its card with the quote: one render a row at RUPA's own price. Press Preview three rows to render the first three, so you see real output before the rest. Then Run the batch. Running a batch waits for a person under Actions unless CREA is at L5 Lead, and the approval card carries the quote.
A run works through the rows a minute at a time and saves its place after every row, so a restart loses at most the row it was on and never renders a row twice. It pauses on its own when RUPA's budget cap is reached, and it stops for a person once spend passes 110% of the quote: carrying on past the quote is always a person's decision. Stop ends it at the next row. Every plate files in the Asset Library tagged with the batch and its row, and each row exports as SVG or PNG from the card. Batches in the rail lists every batch with its progress.
You can also start a batch from a team room. Paste the same prompt and rows in a message to CREA and she files a card with the price. Approving it saves the batch and renders the first three rows, and CREA replies with them and the batch id. Ask her to run that batch and a second card carries the quote for the rest. Nothing is saved or rendered before a card is approved.
A row with a missing value, or an image link that is not https, fails on its own with the reason; the rest carry on. A batch takes up to 1000 rows.
Our logic: a batch is the one place a single click could spend a lot, so you see three real rows and the price of the rest before anything else renders, and the runner can never spend past what you approved without asking again.
Annotate a screenshot
Type /annotate to open the annotate card, then Pick a screenshot from your computer. The screenshot stays in your browser: nothing is uploaded, no model reads it and nothing is spent. Pick a mark and drag on the screenshot to place it: a Numbered callout, an Arrow, a Box or a Blur. Undo takes back the last mark.
A blur shows hatched on the card. Export redacted PNG pixelates each blur into the image itself, in blocks coarse enough that small text cannot be read back, so what was under it is not in the file. Use it for the screenshots in a product guide or a demo, and blur any email address, name, number or key that is not yours to show.
Brand Lock
Brand Lock is on by default. While it is on, "This workspace's brand guidelines go with every render." The guidelines come from your brand book, which you edit from Open the brand book in the rail, or at nagent.ai/admin/brand.

You can switch it off for one brief with Turn off for this brief. When it is off, no brand material goes with the brief at all, including your logo.
On the bench, Send the brand mark with this render decides whether your logo travels with the brief. When no logo travels, the render is told to draw no logo of its own, except on the Logo mark preset, whose whole job is the mark. What goes with every render shows the exact text that is sent.
Our logic: an image model knows nothing about your brand unless it is told, every time. Brand Lock makes that the default rather than something you have to remember, and the bench shows you what was said so nothing is hidden.
When a render waits for you
Whether RUPA renders straight away depends on her autonomy level, which you set in her page in the agent workbench. The bench shows the level under Autonomy.
| Level | What happens when you send |
|---|---|
| L1 or L2 | RUPA writes a plan and stops. She does not render. The plan says to raise her to L3 Operate or above |
| L3 | A Plan card appears: "Nothing renders until you approve it." It shows the estimated spend. Click Approve and run to render |
| L4 or L5 | She renders at once: "Ran on standing approval." Every render is written to the audit log |
The approval at L3 happens on the page, in the plan card. It does not go to your Inbox or the Actions page. A RUPA added with a content or marketing team starts at L4. If the bench reads Not read, RUPA is not set up in this workspace and every render is refused.
Our logic: a render spends money. At L3 the person who clicks Approve and run sees the cost before it is spent. Higher levels trade that click for speed, and the audit log keeps the record either way.
What it costs
The plan card at L3 is the one place a cost appears: "Spend about $X on this render." The figure grows with HD quality, the number of plates and each input image.
At L4 and above no plan card appears, so the studio shows no cost before or after a render. The spend still counts against RUPA's budgets, which are checked before anything is sent:
- RUPA's daily and monthly budget, set in the agent workbench.
- A per-action cap for one batch.
- Your workspace's own daily and monthly limits.
If a batch would go over, RUPA refuses and says which limit and by how much, for example that the batch costs more than her per-action cap, with what to change: fewer plates, standard quality, or a higher cap.
The plates
Rendered plates appear as a sheet in the thread. Each plate offers Vary, Export and Copy link, and RUPA's skills:
- Ad set builds the ad set from a plate.
- Restyle restyles it to a reference.
- Relight relights the subject.
- Cut out cuts the subject out.
A Brand check card follows the sheet. It lists banned words found in your brief and the rules that need your eye. It reads the brief, not the picture, so it does not judge the image itself. Look at the plates yourself before you use them.
If a plate shows "Not saved to library. Download to keep it.", saving failed. Download what you want; skills are not offered on an unsaved plate.
The Asset Library
Every saved render, and every file you upload, lands in the Asset Library at nagent.ai/admin/rupa/library. The same library is the Assets tab of the Knowledge Hub.
Filter by source (AI or Uploaded), search prompts or pick a category. Product shoots shows every plate rendered from a shot list, and Whole shoot on a plate shows the rest of that shoot. Hover an asset to rename, retag, download, copy its link or delete it.

Admins see the whole workspace's library. Other people see only what they made or uploaded, and only the person who added an asset can edit or delete it.
What RUPA knows about each image
Every image that arrives in your workspace is described by the model shortly after it lands: an upload, a Knowledge Hub image, a logo in the brand kit, a RUPA render, a preset reference, or an image an agent draws in a team room. The description, with tags and any words read off the image, is kept with the image and indexed, which is what the library's search reads and what an agent reads when it looks through your assets.
Describing an image is a model call, so it costs. It is charged to your workspace's credits and to the budget of the agent the image arrived through: RUPA for anything that lands in the library, the agent that drew the picture for a render made in a room. Both are checked before the call is made, and the charge is about a cent an image. If the budget or the credit allowance is used up, the image is kept all the same; it is simply not described, so a search will not find it until it has been.
Our logic: a library you cannot search by what is in the pictures is a folder. One small call per image makes every asset findable by a person and by every agent, and the budget guard keeps that call from ever running unchecked.
In a team room
RUPA's skills and presets work in a team room as well as in the studio. A teammate can ask her for a render on a built in preset or one of your own, for a product shown as it is, or for a plate edited with Ad set, Restyle, Relight or Cut out. The render follows the same autonomy level and the same budgets as the studio. Images an agent draws in a room are checked against that agent's budget and the workspace's credits before the image model is asked, and they land in the Asset Library and are described like any other image.
Limits today
- Edit and Write modes are not built. Adjust the brief on a sent plan is not available yet: send a new brief instead.
- Cut out and the Logo mark preset ask for a PNG with a transparent background. If the image model set for RUPA cannot draw one, the render goes to a model that can, so the job no longer fails on the model choice. If the provider still refuses the job, the render stops with a clear error rather than a plate.
- The brand check reads text, not pixels.
- Three sizes only, no 9:16.
- An infographic cannot be edited on its card, and the PNG uses the fonts your browser has.
- A shot renders one plate, but its plan card prices the preset's usual four, so the figure shown is higher than the spend.
- A shot list is not saved. Once you leave the studio, the plan is gone; its plates stay in the library under the shoot.
- A batch takes rows pasted as CSV only, renders square plates from /variants, and cannot be edited once made: make a new one instead. Its PNG export needs the plate's host to allow the browser to fetch it.
Where to go next
- The Knowledge Hub: where your brand book and assets live.
- HOOK: the ad creative team lead, who can hand a brief to RUPA.
- MOXA: social posts that carry a RUPA image.