How to use DesignPresso

The top bar holds the project switcher, the save indicator and version history, Run (▶), undo/redo, Share, a plan badge, your credit balance, notifications, and the account menu. The thin icon rail on the left opens and closes work panels, the middle is the node canvas with the Elements bar pinned beneath it, and the entire right column is the AI agent chat.

Enter the studio and make your first canvas

1. 1

   Sign in and onboard

   Sign in from a desktop browser and the studio loads. First-time users pick a nickname (there's a randomize button) and answer a short survey; accounts that skipped the signup form — social logins, for instance — confirm terms and age here too.

2. 2

   Claim your welcome credits

   Saving your onboarding answers triggers a 🎉 celebration that hands you your signup credits. Click through and you land straight in the workspace.

3. 3

   Create a canvas

   Click the current canvas name (chevron) in the top bar to open the list and choose 'New canvas' at the bottom, or open the Projects panel from the left rail and use 'New canvas' at its foot. The name is optional.

4. 4

   Choose the output ratio

   In the New Canvas dialog, pick 16:9 (wide, default), 9:16 (vertical), or 1:1 (square) under Output aspect. It becomes the canvas-wide default for generation, and any node can override it. Square tends to suit single product renders; the wide ratio suits comparison and spec sheets.

5. 5

   Start working

   The fastest way in is the chat on the right: describe what you want in plain language — "work up a concept for a folding camping lantern", "render this sketch". If you have a sketch or reference photo, drag the file onto the chat composer; on send it also drops onto the canvas as an input node.

The left icon rail

• ✨ Create — the input, generation, and edit tools you drop on the canvas; open by default.
• 📁 Projects — switch, search, rename, and delete canvases, sort them into folders, and move one into the team space.
• 🖼 Assets — your generated results, registered elements, and uploaded files.
• 🪄 Skills — skill nodes you drop on the canvas and run yourself.
• 📖 Knowledge — the design-methodology lenses the agent applies to your work.
• At the bottom: Team (only when you administer one), Theme, and Settings.

Handy to know

• Re-click an open rail icon to collapse the left panel and widen the canvas; the '<' at the left of a panel header does the same.
• Leave a new canvas unnamed and it auto-numbers to 'New canvas N'. Rename it anytime with the pencil icon in the top bar.
• The theme defaults to dark regardless of your OS setting; a change from the rail's Theme button is remembered for this browser only.
• Drag the thin divider between the canvas and the chat to resize the chat column; the width is remembered next time.

The core flow

1. 1

   Type your request

   Write what you want in the composer at the bottom of the panel and send with Enter or the up-arrow (↑) button; Shift+Enter makes a new line. While typing Korean, Enter only commits the character rather than sending. Pure conversation — questions, discussing direction — just gets a reply and generates nothing.

2. 2

   Pick Auto or Manual

   Use the small dropdown at the right of the row just above the composer to choose Auto or Manual. Auto generates straight away; Manual shows an estimated-cost plan card, one step at a time, before anything runs. Your choice is remembered per account.

3. 3

   Review and approve the plan

   In Manual mode, when a request would generate media, the agent posts a plan card: what it will make, how many nodes, and the estimated credits for the next single step. 'Yes, generate' builds exactly that step, 'Stop' cancels, and 'Modify' moves your cursor into the composer so you can say what to change and get a fresh proposal.

4. 4

   Attach references and elements

   Click the paperclip in the composer or drag files onto it to attach sketches, product photos, material swatches, or documents. On send, images also drop onto the canvas as input nodes while documents feed the project brief. Type '@handle' to pull in a registered element so the same product stays consistent across shots.

5. 5

   Watch it build and react

   From one message the agent can create several nodes, connect them, and run them in order, with a 'Thinking…' bubble and stage label while it works. When a generation finishes it takes another turn to comment on the result, continue, or explain a failure and offer a fix — it won't silently rebuild things that already succeeded. To move on, click the full-width 'Continue ▸' button above the composer.

Run mode: Auto vs Manual

• Auto — generates immediately without approval. It doesn't preview cost and only asks whether to keep going on long, multi-step builds.
• Manual — shows the estimated credits first and proceeds only after you approve with [Yes, generate / Stop]. Worth switching on before you ask for something expensive.

Buttons instead of typing

• Approve / Modify / Not now buttons under any reply that ends by asking whether to continue. 'Not now' drops the idea and tells the agent not to ask again.
• 'Yes, generate' / 'Stop' / 'Modify' on a plan card.
• When the agent proposes several options — concept directions, for instance — click the card you like. Your pick becomes the reference for everything downstream.
• The full-width 'Continue ▸' button that appears after a build finishes or the agent pauses itself.
• In Auto mode these gates are clicked for you, with an auto-approve notice in the thread.

Other controls

• Translation: messages written in a language other than your app language are auto-translated into yours, with a toggle under the message to see the original. In a shared project each person reads the same thread in their own language.
• New chat, Stop, queue: 'New chat' at the top-right clears the conversation and stops any in-flight run; while the agent works, the send button becomes a Stop (■). In a shared project everyone's messages are handled one at a time.
• The Run (▶) button in the top bar: it doesn't advance the conversation, but it re-runs any node left pending or failed in dependency order, which revives a build that stalled.

How it progresses

1. 1

   Brief — what, and for whom

   Purpose, who uses it, the direction of the form, and any hard constraints. If something's missing the agent asks one or two short questions rather than running a full interview.

2. 2

   Concept — get distinct options and pick one

   You get two or three concept cards, each with a genuinely different form language — three renders that differ only in color aren't options, and the agent is built to know that. Click the one you want and it proceeds with that.

3. 3

   Lock — pin the product to one sentence

   The chosen concept gets locked into one sentence covering its color, material, finish, and signature form. That exact sentence is reused verbatim in every later prompt, which is what keeps it looking like the same product from shot to shot. Register it as an element and you can summon it anywhere with its @handle.

4. 4

   Develop — angles, materials, sheets

   See the locked product from more angles, compare material candidates, and write the settled CMF onto a board. Through this stage only one variable changes at a time — if the angle shifts while you're comparing materials, the comparison stops meaning anything.

Sketch to render

If you have a hand sketch or line drawing, upload it in the chat and ask for a render. Don't re-describe the form in words — the sketch already supplies it, and saying it twice makes the two fight and blurs the result. Describe material, finish, lighting, and background instead. The agent reads heavy lines as silhouette, thin lines as part splits, and hatching as curvature, fills unfinished areas conservatively, and tells you what it assumed.

How closely to follow the sketch

• Loose — the sketch is a direction and the render tidies the form. Good early on, or when the sketch is rough. Saying "something like this" lands here.
• Strict — the silhouette and proportions are pinned to the sketch. Right when proportions are already decided or the drawing is near-technical. Saying "exactly this shape" lands here.

Re-shooting a render

To change only the lighting or angle of a design you already have, just ask for it to be re-shot. Only camera, lighting, and background change; the product's shape, color, and materials stay locked to the reference. Angles run from a three-quarter hero through front, side, top-down, and detail macro, and the material picks the lighting: soft three-point for matte plastic and wood, a large soft key for metal and gloss, backlight for glass and translucents, and a grazing light when you want brushed grain or leather pores to read.

Sheets that organize the work

• Material comparison sheet — the same product across several materials or colorways on one sheet, with the angle, lighting, and background all held fixed. Two to six panels works; beyond that they get too small to judge. This is for exploring while you're still deciding.
• CMF board — closer to a spec document, made once decisions are settled: a hero render plus per-part callouts naming the color (with its HEX), the material, and the finish. Every swatch sits under the same light so the finishes are actually comparable.
• Three-view sheet — front, side, and top on one canvas as straight orthographic projections, all at the same scale on a shared baseline with no perspective, so it reads as a drawing. If you know the dimensions, ask for them to be annotated.

Change one part

1. 1

   Pick the image

   Click the render node you want to edit so it's selected. With several results on the canvas that removes any ambiguity about which one you mean.

2. 2

   Name the part and the change

   Say one part and what it becomes: "change just the handle to oiled walnut". Write it in whatever language you use — the agent translates it into the part noun the masking step needs.

3. 3

   The mask and the edit chain up

   Two nodes appear on the canvas. The first produces the mask for that part; the second takes the original image together with that mask and regenerates only what's inside it. The original feeds both.

4. 4

   Check the result

   The result arrives as a new image node and the original stays where it was, so if you don't like it you can go back to the original and try again.

What works and what doesn't

• Name one concrete part. Handle, armrest, strap, lid, base, button — a single visible component gives a stable mask.
• Vague pointers like "the top bit" or "somewhere on the left", and names padded with adjectives, make the boundary wobble. Strip it back to the component name.
• Say how the material looks, not just what it's called. "Full-grain leather with visible pebble texture and softened creases" beats "leather" by a wide margin.
• Restyling the whole product into a different tone isn't part editing — ask for a plain full-image edit instead.
• If you want to see several material candidates at once, ask for a material comparison sheet rather than repeating part edits.

Tools in the Create panel

• Input — Text Prompt (a node that just holds wording, no model), Prompt (works your idea up), Describe / Ask (reads an image to caption it or answer a question about it), and Image Input (add reference images).
• Create — Image: make a new image from text.
• Edit — Compose (combine several images via a prompt), Remove background (cut the subject out onto transparency), and Upscale (raise the resolution).
• Click a card to drop its node on the canvas and open the inspector, or drag it to the exact spot you want. The search box at the top (⌘K) filters by name and description.

Working it by hand

1. 1

   Place a node

   Click a tool in the Create panel or drag it onto the canvas. Right-clicking empty canvas also opens a node menu.

2. 2

   Wire nodes together

   Drag from one node's output dot to another node's input dot. Colors mark the type — blue is text, green is image. Drag from an output dot onto empty canvas and release to get a menu filtered to compatible nodes, and the one you pick arrives already connected.

3. 3

   Adjust the settings

   Click a node and the right-hand inspector opens its inputs and parameters. Enter or change values there.

4. 4

   Generate

   Press the generate button near the bottom of the inspector to run that node. To process several at once, use Run (▶) in the top bar: it re-runs every pending or failed node in dependency order, and the badge on the button shows how many are left.

5. 5

   Use the result

   The finished image lives inside its node. If you ran the same node more than once, flip through past results with the 1/N control above it — and every result also collects in the Assets panel's library, where you can download it.

Tidy and organize

• Select two or more nodes to get a group button that wraps them in a named, colored box. Drag a node out of the box and it leaves the group automatically.
• The round toolbar on the left edge: the wand (Optimize layout, shortcut D) re-lays the graph by dependency, the frame button (Focus, F) fits the view to your selection or the whole graph, and the cursor button (C) toggles between select and hand/pan modes.
• Use the minimap and mouse-wheel zoom to move around a large graph quickly.

Elements are account-wide, so you register each one only once. Type its @handle in a chat message or a prompt and, at generation time, DesignPresso swaps in the real name, appends the description you saved, and — for models that accept image references — attaches the reference picture too. There are four categories (People, Places, Props, Audio); in product design the usual move is to register your locked product concept as a Prop.

Register and summon

1. 1

   Start from the Elements bar

   Click the dashed '+' at the right end of the thin Elements bar under the canvas and choose New element.

2. 2

   Fill in the form

   Pick a category, type a name, set the @call_id (auto-filled from the name and fully editable), and write a description. Add an image by upload or from the library, or save without one.

3. 3

   Summon it

   Write the @call_id in a chat message or a text/prompt node. You can mention several in one prompt.

4. 4

   Copy, preview, and manage from the chip

   On a chip in the Elements bar, click the @call_id text to copy the handle, or the thumbnail to open a large preview where you can change the image or copy the handle. Hover and click X to unpin it from this canvas — the element itself stays in your library.

Other ways to register

• Assets › Elements tab — your account-wide library. Elements registered here are not auto-pinned to a canvas, so pin one from the bar's '+' when you need it.
• Agent proposals — when something keeps coming up in the conversation, the agent offers a registration card; just accept or decline.
• From a result — choosing 'register as element' on a result tile opens the form pre-filled with that image as the reference and its prompt seeded into the description; you only set the name and handle.

Skill nodes

Click the wand icon in the rail to open the skills gallery. Cards are split into your own, your team's, and built-in, with a search box at the top to filter by name. The 'in N · out kind' line on each card tells you how many inputs it needs and what it produces. Built-ins include Multi-Angle, which re-renders the same product from a numerically specified viewpoint, and LightBox, which turns a layout of several products arranged at true relative scale into one studio product photograph.

Run a skill on the canvas

1. 1

   Add the node

   Click a card to drop its node on the canvas — a toast confirms — or drag the card and release it exactly where you want it.

2. 2

   Wire an input

   Drag a wire from an upstream node's output into the skill node's left-side input handle. Handles are labeled by role, and a * marks a required input.

3. 3

   Set the controls

   Set the on-node sliders, dropdowns, toggles, numbers, or text fields. A skill with no controls says so — it runs on its inputs alone.

4. 4

   Run it

   Press generate and the result appears as new output node(s) auto-wired to the right. Change a value and press again and the variants fan out sideways instead of overwriting the previous output.

Knowledge lenses

Knowledge lenses are bundles of methodology the agent consults while deciding. Two are on by default and shape how the product behaves: a design-director lens that holds the brief → concept → lock → develop order on new work, and a part-edit lens that routes "change only the …" requests into masked part editing rather than a full regeneration. Beyond those there are lenses for turning form and proportion into concrete visual facts, deciding colorways and CMF, describing how a material actually photographs, directing a render's lighting, background, and camera, judging grip and dimensions on things you hold, reading sketches and marked-up overlays, holding brand consistency across a series, and checking manufacturability.

The book icon labeled Knowledge in the rail opens the lens list. Each row has a name, a short note on when it applies, and a toggle on the right; switching one on applies it account-wide, not just to the current project. There's a cap on how many can be active at once and you'll be told when you hit it. System lenses are locked, so you can only turn them on or off, not edit them.

Make your own

• Custom skill: the '+' button at the top of the Skills panel lets you author one by editing a JSON template (a node manifest is required). It lands under your own skills, and malformed JSON is rejected with a format error. There's no visual builder.
• Custom lens: the New button in the Knowledge tab takes a name, a 'when to use' trigger, a category, and the methodology body. The trigger is what the agent reads to decide whether to apply it, so keep it to one or two clear sentences.
• Team share: if you're on a team, the share icon on your own skill or lens card pushes it to everyone's panel, with unshare and delete in the same spot. These controls appear only when you actually belong to a team.

The Projects panel

• Click a name in the list to switch canvases, and use the search box at the top to filter by name.
• Canvases you own can be renamed inline or deleted, with a confirmation step before deletion.
• Folders are a single flat level for tidying up: create one, drag a canvas onto its header to file it, and drag it to the 'no folder' area to take it back out.
• If you have a team, a canvas you own can be moved into the team space so every member can co-edit it — and moved back to personal just as easily.
• 'New canvas' at the foot opens the same dialog as the top bar.

The three Assets tabs

• Library — where generated results collect. A toggle at the top switches between this canvas and everything in your account, with a team scope too on team canvases. Download or delete straight from a tile.
• Elements — the account-wide list of registered @references. Filter with the category chips or find one with the meaning-based search.
• Uploads — everything you brought in through the chat or an input node.
• Library and Uploads each have a search box at the top, and the grid loads more as you scroll so a long history doesn't dump itself all at once.

The owner of a canvas can share it two ways from the Share button in the top bar: switch on a read-only public link anyone can open without an account, or invite specific people by email as Viewers (read-only) or Editors (full co-editing).

Share and co-edit

1. 1

   Open the Share dialog

   Click Share at the top right. It only appears if you're the owner, and its icon turns green while the canvas is public.

2. 2

   Turn on a public link

   Flip the public-link toggle at the top of the dialog and a copy button appears. Copy the link and pass it on yourself; whoever opens it gets a read-only view with no login. Flip it back off to unpublish.

3. 3

   Invite people by email

   Type the person's email in the invite row, pick Viewer or Editor, and click Invite. If they already have an account they join at once; if not, the invite waits as pending and they join automatically the moment they sign up or log in with that address.

4. 4

   Manage members

   Invited members are listed with their email. Use the role dropdown on their row to switch between Viewer and Editor, and the trash icon to remove them instantly.

5. 5

   Co-edit in real time

   When an invitee opens the same canvas you work on it together. Editors get the graph, generation, and the agent just like the owner; Viewers get a fully read-only studio and watch changes stream in live.

What each role can do

• Owner — the only one who can publish, invite, rename, and change roles. The Share button and the rename pencil show only to them.
• Editor — edits and generates like the owner, but generation credits come out of their own wallet. Renaming the canvas stays owner-only.
• Viewer — the whole studio is read-only. A read-only badge sits next to the canvas title, and editing, generating, and typing to the agent are all locked.

Presence and notifications

• Who's viewing — anyone else with the canvas open stacks up as a colored avatar in the top bar. Hover to expand the list; nothing shows when you're alone.
• Live cursors — while co-editing, each person's pointer shows on the canvas as a colored arrow with a name tag. There's nothing to switch on.
• Notification bell — the bell at the top right collects what happened while you were away, such as canvas invites and finished or failed generations. Click an item to mark it read and jump to that canvas.

Team workspace

If you own or administer a team workspace, a Team item appears at the bottom of the left rail. The team console has four tabs — Members, Credits, Billing, and Settings — covering members and seats, the shared credit pool, the plan, and team settings. Move a canvas you own into the team space from the Projects panel and everyone on the team can co-edit it.

Your balance sits in the credits card at the top of the page. You never set a per-job price — each generation deducts what it actually used — so you track spending by watching the balance and reading the recent-credit-history table at the bottom. Note that a one-time top-up gives 17–21 credits per $1 depending on the amount while spending happens at real cost, so the buy rate and the spend rate differ. All charges run through Stripe in US dollars.

Buy one-time credits

1. 1

   Pick an amount

   In the top-up section, tap a quick-pick chip or drag the slider to set an amount (minimum $50, maximum $600).

2. 2

   Check the live summary

   The panel live-updates the credits you'll receive, the per-dollar rate, any volume-discount badge, and a hint of how much more would get you a better rate. Bigger amounts give more credits per dollar.

3. 3

   Pay

   The pay button opens a confirmation showing the credits granted, the total, and the refund policy. Review it and continue to Stripe Checkout.

Subscription plans

• Use the annual/monthly toggle in the subscription section. Annual is preselected and carries a discount badge; picking it shows the effective monthly price (annual fee ÷ 12) on each card with the original monthly price struck through.
• Each card — Starter, Pro, Studio — shows monthly credits, price, a rough estimate of how much you can generate, and the concurrent-generation limit (4/8/12). Click subscribe, confirm the auto-renewal notice, and continue to payment. Going over the concurrency limit doesn't fail; extra requests simply queue.
• While subscribed, upgrading to a higher card applies instantly with a prorated charge and matching credits; downgrading takes effect at the next renewal with nothing charged now. Cancelling ends access at period end, and if a cancellation is already scheduled you can resume instantly.
• Manage invoices & payment method opens the Stripe customer portal to view invoices and update your card — available once you have a prior payment. Where organization plans are enabled, an individual/organization toggle switches to Education, Team, and Enterprise (Team needs a 3-seat minimum; Education requires institution approval).

More ways to top up and save

• If you've never topped up, the card shows a first-top-up badge and your first paid top-up automatically grants double the base credits — once per account.
• Under auto-refill, click Manage to set the refill amount, threshold, and monthly cap, tick the consent box, and save; when the balance drops below the threshold your saved card is charged automatically. You can switch it off anytime.
• In the coupon section, enter a code and click redeem or press Enter to add the credits and refresh your balance.