Skip to content

Make your first complete video

This tutorial produces one short clip without dialogue of a blue paper boat crossing a quiet pond. It teaches the complete path from an idea to a downloadable video, with a review before each paid generation.

You need a signed-in studio account, a balance for the estimates you accept, and permission to use any references you add. The text below is a fictional practice brief; no supplied media or private project is required.

If you want a free orientation first, open the prepared tutorial from Projects. That preview currently ends after showing a prepared scene branch at step 4. It does not include the complete production and export workflow below.

Create a new project. Add Text (input) (Текст (ввод)) and enter:

Make one five-second scene without dialogue. A blue paper boat drifts across a quiet pond at dawn. Keep the camera fixed and show one continuous action. No spoken dialogue.

Add Prompt Scriptwriter (Промпт Сценарист) and connect the Text (input)’s text (текст) output to the writer’s topic (тема) input. Set the project or writer format to the shape you want to export.

For this exercise, use the simple boat description. A recurring character belongs in a separate Element workflow.

Checkpoint: the writer’s incoming context shows your complete idea, and its chosen format is the one you intend.

Review the writer’s current estimate, then run it. This may charge. Open the full result and check every returned scene: one simple action, the correct subject, the requested duration and no dialogue. The model’s output still needs review, even when the request asked for exactly one scene.

If it wrote extra scenes or changed the idea, revise deliberately before generating images. Do not silently discard paid text to make it appear that the request succeeded. Read any critic findings and its actual check status.

Checkpoint: the written scene describes what you want to see, and the critic’s state is understood.

Use the writer’s scene expansion control. Expansion builds Image/Video branches and their connections without generating their media. Inspect the chosen video model and method.

For this continuous action, a first-frame route is a straightforward choice: a composed pond image establishes the opening, then video adds motion. If the expansion proposes a storyboard sheet or another method, read why before continuing. See frames, sheets and keyframes.

Checkpoint: the image has a clear job and the video mode accepts that job.

Review the Image node’s prompt, references, model, format and estimate. Run the image branch, then inspect the full image.

Look for one blue boat, an uncluttered pond, usable framing and a starting pose that leaves room for movement. Correct the instruction and generate another version only if the first result is unsuitable; regeneration can charge again.

Checkpoint: this is the image you want at the opening of the clip, not a sheet of several views or panels.

Review Video’s selected model, mode, duration, accepted inputs and current estimate. Run the Video node. Wait for the actual result, then watch all of it with sound.

Check that the boat moves as intended, the camera stays fixed, no unwanted speech appears and the clip has the expected duration. A camera instruction is not a guarantee: reject or revise a result that does not satisfy the exercise.

Checkpoint: the entire clip is suitable, including its opening, ending and audio.

Select the finished Video and use Download result (Скачать результат) on its toolbar. Open the downloaded file and verify it plays correctly.

If your accepted script contains several scenes, the film exists only after they are joined: connect every clip to one Montage (Монтаж), check their order and press Assemble (Собрать) — Montage joins the clips on the server and costs $0. Montage studio (Студия монтажа) under the card opens the timeline for order, trims and the song. Then download the film from Montage the same way. Assemble and finish a film covers music, trimming and captions.

Use MCP to discover the project and catalog, build the graph, quote/run the writer, inspect its scenes, call expand_scenes, then quote/run each chosen image and video branch. Use run a node for the sequence and cost boundaries.

Graph-aware Montage is not connected to run_node; finish there in the browser or use assemble_video with explicit approved clips. Verify the saved graph or returned material before calling the task complete. Publishing to social accounts is a separate browser flow.

Last checked: 14 October 2026. This is an executable workflow guide, not a claim that a new paid render was performed for this article. Estimates depend on the selected models and current inputs.