How the studio works
A project is a graph of steps. Each node has a job: hold an idea, describe a character, write scenes, create an image, generate a clip or join finished clips.
Read a graph
Section titled “Read a graph”| Part | What it means |
|---|---|
| Node | One step, with its own settings and result. |
| Input port | Where the node receives text, a scene, an Element or media. |
| Output port | What the node passes to the next step. |
| Wire | A connection from an output to a compatible input. |
| Result | The text or media produced or held by a node. |
A typical scene workflow is Text (input) → Prompt Scriptwriter → Image → Video → Montage. An Element can provide the character or location information used along the way. This is a conceptual route: the actual inputs and connections depend on the chosen nodes and video model.
Russian interface names include Текст (ввод), Промпт Сценарист, Картинка, Видео, Монтаж and Элемент.
Build first, then run
Section titled “Build first, then run”Adding nodes, changing fields and connecting wires prepares the graph. It does not generate the finished media by itself.
Running a node can also run the steps it depends on. Unchanged, reusable results can come from the cache. Changing a prompt or an input can make downstream results need another run; use the current cost estimate before generating again.
A result marked done in a saved graph describes the stored result. It is not, by itself, a promise that running the node now will be free. Through MCP, price_quote evaluates the current run plan.
Inspect before changing
Section titled “Inspect before changing”In the studio, follow a wire to see what supplies a node. Through MCP, use describe_studio for the available node types and read_project_graph for the actual project. Use returned port names exactly when constructing connections, even when those names are in Russian.
Continue with Elements, scenes and media, then follow your first complete video. For exact ports, settings and execution support, use the node reference. An agent can start with the read-only MCP walkthrough.
Last checked: 14 October 2026. This page explains the shared graph model; MCP execution support varies by operation.