Instructions, scene text and visible results

The text you type, the scene arriving through a wire and the result shown on a node have different jobs. To understand a generation, inspect all three.

Follow the text through the graph

1. Your input is the brief or instruction you edit in a field or a Text (input) node.
2. Incoming context is supplied through connected text, scenes, Elements and applicable lore.
3. The node’s output is what the next node receives after execution. A scene can carry text, reference context and generated media together.
4. The visible result is the saved output of a particular execution. It can remain visible after you change the inputs.

For example, write “A blue paper boat crosses a quiet pond” in Text (input) (Текст (ввод)). Connect it to Prompt Scriptwriter (Промпт Сценарист). The writer’s individual scene outputs are the scenes to inspect and pass downstream; full text (полный текст) collects the written script on one output.

Keep the current scene as the source of truth

The complete saved scene text is the canonical content for the next step. A short card preview is a view of that content, not a replacement for it. Open the full result when reviewing long text.

A scene rewritten after the original script run may differ from the original script artifact. The old artifact records that earlier generation. It is useful history, but it does not establish what the edited scene now sends downstream.

Through MCP, read the current graph and use its returned scene_in or scene_out address with describe_scene. Compare the origin, rewrites and recorded deliveries. A report can explicitly mark text as shortened or omitted; consult the full node on the canvas instead of treating a partial report as the whole scene.

A text connection can replace scene text

In Image and Video, an unplaced text N input takes precedence over inherited scene text. This matters when connecting a style description: an extra style wire can become the main text instead of adding to the scene.

When the incoming text should appear at a particular point in your instruction, insert its text placeholder, such as {{текст 1}}: type {{ in the field and pick the input from the list, which names each connected input by its source and also offers the next free one. Typing @ does not insert texts: in Image and Video it addresses a character (element) by name. The placed input is inserted there and does not displace the scene through the unplaced text row. Check the incoming context preview and the prompt’s parts before running.

Your own field remains an editable direction alongside inherited context. Editing that field does not edit the upstream writer’s saved scene.

A result can be visible and stale

Changing a model, instruction, reference, upstream result or applicable lore can make an existing result stale. The saved image or clip remains useful to compare, but it was made with the earlier inputs. Read the freshness indication and current run plan before deciding whether to reuse or regenerate it.

See runs, cache and cost and inspect a project for the checks to make before changing or running anything.

Last checked: 14 October 2026. A scene report describes recorded information; missing history is not evidence that no generation took place.