Edit the canvas and prompts

A project is a graph of nodes with typed inputs and outputs. The current flow view presents the result and the node’s action together, with inputs around the card. Changing the view does not create another copy of the project.

Find the right control

Use Studio menu → View to fit the graph, change zoom or arrange the nodes. Hand moves the canvas; Selection selects nodes. Edit contains selection, duplication and deletion actions, while project history and graph edit history are separate from generation history.

Select a node and inspect its details when a compact card hides a setting. Preferences → Parameters shown → all reveals the broader set of parameters; it does not enable unsupported model capabilities.

Connect an output to a compatible input. The input’s role matters: an Element reference, scene text, first frame and last frame cannot be treated as interchangeable wires. Follow the node reference for the selected type.

Edit a prompt as a reusable instruction

Open the editor with ✎ Editor under the field. It is offered for every text field you can type in, empty ones too: node prompts, the текст field of Text input, a character’s description, a video’s negative prompt, the style of a Prompt Writer, a subtitle context, a voiceover text and so on. Where the studio does not assemble the text (a character’s description, for example) the editor shows it as typed, without formula chips; where it rolls formulas before sending, they appear as chips. A field filled by a wire, a protected system instruction and read-only previews open a read-only ⤢ Expand view instead. The editor distinguishes editable text from protected material and represents supported formulas as chips. Open a chip to inspect or edit its alternatives. In View, a text placeholder such as {{текст 1}} on a node with a текст N input row is shown as a chip named after its source, for example текст 1 · Idea; click it to see the text that arrives over that wire, or the note that the source has not run yet. A placeholder whose input has no wire reads текст 5 · not connected; open it to see the same reason the greyed-out run button gives. Edit shows the placeholder exactly as typed. To insert one, type {{ in Edit, in the field on the card or in the inspector: a list of the node’s текст N inputs opens, named by their sources, and the pick replaces {{ with the placeholder. The list offers only texts. @ is for characters only: in an Image or Video prompt, typing @ opens a list of the characters (elements) connected to the node or arriving with its scene and inserts @name, and never offers texts; on Text input, typing @ opens no list. On the card, a chip for a text that arrives over a wire shows a text icon (a scene shows a clapperboard) and the source’s name without @, for example Idea; only characters are written with @. On Text input, the editor also shows Assembled — goes down the wire under the text: exactly what the node sends to the next one, with formulas rolled and wire texts highlighted and named by their source. A connected text without a placeholder is marked added at the end; a source that has not run yet, or has changed since its last run, is marked waits for a run of followed by its name, with the text the run will send shown when it is already known, as it is for another Text input. A placeholder with no wire stays as typed and is marked not connected — the run will not start.

For example:

A small ceramic cup on a wooden table at %rand(dawn, dusk)%.
A static camera; no camera movement.

The studio chooses one alternative and sends plain text to the model. The choice stays with the result; regenerating can choose a new alternative. An alternative containing a comma can be grouped:

%rand("soft, warm light", cool light)%

This is a limited formula notation, not a programming language. Unknown or invalid formulas are reported and can remain literal text. Correct the warning instead of assuming the expression was evaluated.

Distinguish the source and outgoing text

The editable source, connected text and the final instruction sent to a model can differ. Where available, expand Goes to the model and compare it with the intended composition. The prompt editor does not independently reroll the generated result.

When text is wired into another node, use that node’s documented input contract and mention controls. Do not assume a connected style instruction is automatically appended to a separate scene description. Read instructions and visible results before combining them.

A lore document uses the same editor shell with a different editing contract. It is saved as a document; formula chips are not a promise that lore is evaluated as a generation template. Confirm Save and resolve an unsaved-edit prompt before closing.

Example: change only the lighting

Keep the character references and camera instruction unchanged. Edit the lighting alternative, inspect the outgoing text and run only the node whose result you want to replace. Then check dependent outputs for freshness before assembling the clip.

The visible old result can remain while its instructions have changed. Inspect runs and cache rather than using the thumbnail alone as proof that the new prompt ran.