How donAI works, end to end.

1. Open the dashboard and type the campaign in the command bar.
2. Press send. The brief opens in Chat as your first message.
3. Pick a mode — Single for one image, Campaign for a set.
4. Watch the shots arrive; they land on the canvas as they finish.
5. Right-click any image to download it or run an automation against it.

The planner is a planner, not a mind reader. It reads four things out of what you write — the subject, the count, any style words, and an aspect ratio — then builds the shot list around them. Everything else you write still reaches the model as prompt text, so detail is never wasted.

A brief that lands well names four things:

The product

What it is, in a few words. “A matte-black insulated bottle” beats “our product”.

Who it's for

Age, style, context. It changes the casting, wardrobe and setting more than any other word.

The feeling

Editorial, cinematic, handheld, clinical. This is what the Style menu sets programmatically.

How many

The planner defaults to three; say “five shots” or “make me 4” and it uses your number.

One field, one action. Whatever you type is handed to Chat as an opening brief — it doesn't generate on the spot, it opens the composer pre-loaded so you can still change the mode and model before spending anything. The New campaign button takes you straight to an empty Chat.

A newest-first strip of everything you've generated, pulled from your projects rather than from a separate feed — so it includes images made on the canvas and from automations, not just chat. The count sits on the right of the heading.

Up to three of your Characters are shown as sticky notes, ready to @-mention without opening the page. Create more there and the strip rotates.

Signed out, the app still works as a demo: you can write briefs and generate. What you don't get is persistence. Chats are stored per account, so nothing is kept between sessions until you sign in. Characters, automations and the canvas are separate from that — see where everything lives.

The chat name sits at the top. Click it to rename in place — Enter commits, Escape cancels. To the right are Share, and the Switch chat menu, which lists every chat you have and opens chat history.

Every assistant message has three small buttons underneath it: Copy puts the text on your clipboard, and the thumb up / thumb down buttons mark the response so you can find the ones that worked.

Four controls sit in one row under the prompt field, and they never move when you change modes — the bar keeps its shape so you can't lose your place:

• Mode — the + button. Chat, Single, Campaign or Video, plus attachments.
• Model — which model draws it. The list swaps to video models when you switch to Video.
• Style — art direction layered on top of the model.
• Send — the round black button at the far right.

Enter sends, Shift + Enter starts a new line. While a mention list is open, the arrow keys move through it, Enter or Tab picks, and Escape closes it.

Four passes, in this order:

1. Lead noise — “can you create an image of” and similar wrappers are stripped, so the planner works on your subject rather than on the sentence around it.
2. The subject — what the shots are of. If the brief names no brand or product, this falls back to a neutral stand-in rather than inventing one.
3. Count — an explicit number in the text wins; otherwise it defaults to three.
4. Style and aspect — up to three style directives and a ratio, read from the words you used. The default ratio is 4:5.

Shots are drawn from a fixed set so a campaign has range instead of six variations of the same frame. Each one carries a human label and the camera language that goes into the prompt:

A campaign is capped at seven shots. The cap is not arbitrary: the image backend accepts at most four outputs per request, so anything past that has to run as sequential calls, and going wider costs more than the extra frames are worth.

Video swaps the camera vocabulary for movement. One shot, one continuous take — no cuts — drawn from:

Push-in · Orbit · Tracking · Tilt-up · Hero static · Pull-back

The list is filtered to what actually works. Mentioning an app you haven't connected would be a dead end, so it doesn't show up. Characters always appear — they live on your device and are always ready.

A chip carries the full name and the app's logo, so a long character name is never truncated to something you then have to remember. Chips are removed by deleting the text around them, not with a separate cross.

For a character, the profile is injected into the brief — age, build, personality, usual setting — so the same person appears across every shot. For a connected app, it tells the run where the result is going. More on publishing.

Characters live in this browser on this device, so they appear instantly and are visible in the full list. The page shows them as simple cards — photo on top, name and description below.

Right-click any image on the canvas and the menu lists every automation you've saved. The source image is passed straight into the prompt, which is the whole point: the recipe stays fixed and the product changes. The result comes back to the canvas as a new node.

Five kinds of node live on the canvas:

• Images — every generation, from chat, automations or uploads.
• Campaign plan — the shot list donAI wrote, kept on the canvas so you can read what it decided.
• Characters — persona notes, so the cast list sits next to the work.
• Automation runs — what an automation produced, with the automation that made it.
• Text — anything you want next to the work. Double-click to edit in place; Enter commits.

Drag a node to move it. Every node has resize handles on its edges — text nodes scale their type as they grow, so a wider note stays readable rather than truncating.

Each node has a handle on both sides. Drag from one to another and donAI draws the link between them — the canvas is a flowchart you build by hand, not a layout it imposes.

Delete is the only destructive item and it turns red on hover, so a mis-click is visible before you commit to it.

The canvas takes freehand drawing over the top, in the colour you pick. Useful for marking up a frame or sketching a layout before generating it.

Saves, runs and errors report themselves in a small pill that follows your cursor, so a long operation never leaves you guessing whether it worked.

Claude and ChatGPT are listed but not yet wired. Opening one lets you store an API key against it so the connection is ready when it ships — the key is kept on your device and nothing is sent anywhere until the integration goes live.

Each provider card shows its state — connected, or ready to set up. Clicking one opens a detail panel with the connection button; most providers hand you off to their own authorization page and return you here when you approve. Connections are per-account: nobody else's account is ever connected to yours.

A connected provider can be disconnected from the same panel, which removes the stored connection and nothing else.

It only lists connected apps plus your own characters. Connect something under Integrations first, or create a character under Characters.

Chats are stored per account. You need to be signed in — a guest session can write but nothing is kept when it closes.

Characters live in this browser. A different browser, a different device, or a cleared site cache starts with an empty set.

Some briefs read like a conversation rather than a generation — “what should I post?” — and the planner answers instead of generating. Say what you want made, or pick Single or Campaign explicitly.

The list follows the mode you're in. Switch to Video and the model menu swaps to the video models; switch back and your image choice is still there.

Shots run sequentially and a run can be interrupted. Everything that finished is already on the canvas and in the Library — send the brief again and say how many you want this time.

Connections are per account. Make sure you're signed into the account you expect — the top bar shows the email you're on right above Sign out.