add_image
ChatGPTPuts images on the board on screen, to the right of what is there, and answers each one's name and box, for prompt notes. url: an https link to an image or a pin on Pinterest, Unsplash, Are.na, Dribbble, Behance, Wikimedia, Imgur, Pexels, Fonts In Use, It's Nice That, Cosmos, Savee or Flickr; any other site is refused, Studia's own links too; a picture attached in the chat goes as file. upload: every photo sent through an upload_link, together, as one step the user can undo. file: a picture the user attached in this chat (ChatGPT). image: an image this account already has, by its name from history (of this board or another): put back, or copied onto this board.
ask_user
ChatGPTShows the user one to three short questions as a card in the chat, each with two to five options to press: one pick each, or several where multiple is true. An option about one of the board's images can show it: give it as {label, image} with the image's name from look or boards. Their answers arrive as their next message. For the assistant's own questions: which board, what the project is for, whose site or current look to start from, what to do next. What an image should show or how it should look is Studia's to ask: prompt asks the user back itself when it needs to.
boards
ChatGPTLists the user's boards in the Studia tab, the one last worked on first (for "my latest project"), and the time now: id, name, which one is on screen, when each was created, last worked on and last opened, and what each is about (its goals, the user's first words on it, and what its images show). Names are often placeholders such as flowers; these lines say what each board is about, so the one that fits the user's words can be opened without opening the others.
get_image
ChatGPTOne image of the board on screen, by its name, larger than look shows it: px on its long side (128 to 1024, 768 by default), with what made it. For when the user asks to see or judge an image; prompt shows Studia the images itself.
look
ChatGPTLooks at the board on screen: every image by its name, with what is known of it (what it shows, how it looks, where it came from) and its box on the canvas, the user's own marks (MARKS), and one snapshot of the canvas. In its words #n is the user's n-th request on the board, unseen means Studia's own agent has not looked at that image yet (not the user), and ← last result marks the newest image. For questions about the board, or images not yet known by name. The user's (1), (2)… are under MARKS, and prompt carries them with the user's words.
marks
ChatGPTThe marks the user drew in Studia, which they name (1), (2)… in the chat: what each marks (which image, which part), and a picture of each. For when the user asks what their marks are; prompt carries the marks with the user's words.
open_board
ChatGPTPuts a board on screen, by its id from boards (whose lines say what each board is about, so finding one takes no opening), or a new empty one with {"new": true}; answers what look would for it. Refused while Studia is working on a request or the user is writing in the prompt bar.
open_studia
ChatGPTShows the real Studia in this chat: the user's canvas, signed in, beside the chat or full screen, where they see and change their images themselves. Pass a board id from boards to open that board; leave it out for the board Studia showed last. Once is enough in a conversation: it stays open, and the other tools act in it, so the user watches the work happen. Calling it again opens another Studia; open_board switches its board or starts a new one.
prompt
ChatGPTMakes, edits or varies images on the user's board in Studia, or finds references in the user's own library (never on the web), from a request in the user's words, as if typed into Studia's prompt bar. Studia's agent reads the board and the user's marks and writes the image prompts. Answers with what Studia made (each new image by name) or said, which may be a question for the user; images take 30 to 90 seconds, and in ChatGPT a longer one answers as a job for wait. One prompt is one request in Studia, as the user would type it: an image, a few directions, an edit, a question about the board, or references from the user's own library. Studia cannot search the web or read a site, and does not plan several requests: it knows a brand, a product or a place only from the words, the board and that library. A designer reads what already exists about the subject before designing (its site, its current look, the product and who it is for), for a rebrand too, whose new look keeps or breaks from the old on purpose; in a chat that reading is the assistant's, and what it found goes into the prompt words (palette, type, tone). A subject with nothing to find may still have a site or a screenshot the user can give. A task of several steps (references from the web, a study of them, directions for each) is the assistant's plan: its own web search, add_image for each reference it chose, then a prompt for each step, as many as the task takes. An image named in the words ("in the style of 3fa9c") lends its look; a note on an image is that image changed.
remove_image
ChatGPTTakes one image off the board on screen, by its name, as the user pressing Delete: they can undo it, and History keeps it (history, then add_image with its image puts it back).
show_images
ChatGPTMoves Studia's camera to images of the board on screen, by name, and selects them. Called by Studia's own button under new images in the chat ("Show on canvas"), not by the model.
status
ChatGPTHow Studia stands for this user: how many images they have left and when that resets, whether Studia is open in this chat, and whether a request is still running. Before promising many images, or when unsure why Studia is slow.
studio_code
ChatGPTA new one-time sign-in code for the Studia page shown in this chat. Called by that page only.
wait
ChatGPTWaits again for a command that answered {"job": …, "status": "running"}, and answers what that command would have. It waits up to about 50 seconds; while it still says running, call it again with the same job.