# Story → Video

> Turn a story prompt into a finished AI film. Story → Video writes the script, casts, plans shots, animates and edits, and stops for your approval.

Source: https://nodaro.ai/docs/nodes/video/story-to-video

The **Story → Video** node turns a short story prompt into a finished film. It runs an AI pipeline in eight stages: script, characters, props, locations, shot list, scene images, animation with audio, and the final edit. Between stages it stops so that you can approve, edit or reject what it made, and everything it makes appears on the canvas as ordinary nodes.

- Found in: Video › Story & Script
- Output: video
- API type: `generative-pipeline`

Story → Video runs on Nodaro Cloud only. It is not available on self-hosted installs.

## When to use it
- You have an idea for a trailer, a short film, a music video, a reel or a commercial, and want a complete first cut from one paragraph.
- You want the AI to cast characters and design locations, and keep them as reusable assets for later work.
- You want to direct at a high level, approve the important choices, and let the pipeline do the rest.

To build a film shot by shot yourself, start from [Generate Script](https://nodaro.ai/docs/nodes/video/generate-script) and [Generate Video](https://nodaro.ai/docs/nodes/video/generate-video) instead.

## Quick start
### Add the node

Press Tab on the canvas and choose **Video › Story & Script › Story → Video**.

### Describe the story

Select the node and type the story in **Story prompt** in the settings panel, up to 4,000 characters.

### Choose the format and the length

Choose a **Format** and set **Target duration (seconds)**. The length must fit the format; the panel tells you when it does not.

### Choose who approves

Choose a **Mode**. **Manual**, the default, stops after every stage for your approval.

### Run the pipeline

Click **Run pipeline**. The pipeline panel opens, and new nodes appear on the canvas as each stage runs. When a stage needs you, approve it in the panel to continue.

Workflow: Story → Video puts its cast and locations on the canvas as real asset nodes. Reuse them in other nodes, here in Generate Video for a new shot of the same character in the same place.

- Character Asset → Generate Video (assets)
- Location Asset → Generate Video (assets)

## Settings
Select the node to see its settings in the settings panel.

| Setting | What it does |
| --- | --- |
| **Story prompt** | The story, up to 4,000 characters. |
| **Target duration (seconds)** | The length of the film. It must fit the format. New nodes start at 35 seconds. |
| **Format** | **trailer**, **short film** (the default), **music video**, **reel** or **commercial**. The format shapes the script and sets the allowed length. |
| **Output resolution** | **480p**, **720p** (the default), **1080p** or **4K**. |
| **Mode** | **Manual** (the default), **Auto** or **Guided**. See [Modes](#modes). |
| **Video Critic Frames** | How many frames of each shot the video critic checks: **First + Last** (the default and cheapest), **First + Middle + Last**, or **5 Evenly Spaced**, which catches the most motion glitches. |
| **Image model** | Optional. The model for character, location and prop pictures and for the scene keyframes, for example Nano Banana Pro or GPT Image 2.5 Flare. **Auto** lets the director choose per shot. |
| **Video model** | Optional. The model for every shot, for example Seedance 2, Kling 3.0 or VEO 3.1 Fast. **Auto** lets the director choose per shot. A model that does not fit a shot's input falls back to Auto. |
| **Script LLM** | Optional. The AI model that writes the script and the scene plan: Claude Haiku 4.5, Claude Sonnet 4.6, Claude Opus 4.7 or Claude Opus 5. **Auto** uses Claude Sonnet 4.6. The critics always use the default. |

The **Run pipeline** button starts the run. After a run, it becomes **Re-run** and starts a new pipeline.

### Formats and lengths

| Format | Shortest | Longest |
| --- | ---: | ---: |
| **reel** | 7 s | 90 s |
| **commercial** | 10 s | 90 s |
| **trailer** | 30 s | 180 s |
| **short film** | 12 s | 600 s |
| **music video** | 30 s | 600 s |

## Modes

| Mode | What happens |
| --- | --- |
| **Manual** | Every stage stops for your approval. You can edit what the stage made, approve it, or reject it with feedback. |
| **Auto** | The pipeline runs every stage without stopping. AI critics check each stage. After three blocking verdicts in a row on one stage, the run fails and the credits not spent are refunded. |
| **Guided** | Like Manual, with a chat at the Script stage where you refine the plan in plain language. See [Refine the plan in Guided mode](#refine-the-plan-in-guided-mode). |

A running Auto or Guided pipeline has a **Switch to Manual** button. A pipeline that an upstream trigger starts automatically always runs in Auto mode, because nobody is there to approve.

## The eight stages

The pipeline panel lists the stages in order. Each one shows its status: waiting, running, awaiting approval or approved.

| Stage | What the pipeline does | What you approve |
| --- | --- | --- |
| **1. Script** | The Showrunner writes the plan: title, logline, cast, locations, props, the list of scenes and their lengths. Critics check the script and that every character is covered. | The plan. **Reject with feedback** makes the Showrunner try again, up to two times. |
| **2. Characters** | A portrait for each cast member and a matching voice from the ElevenLabs library. Then angle and expression variants for each character. | Each character's portrait and voice, then all variants together. |
| **3. Objects** | One reference picture for each prop. | All props together. |
| **4. Locations** | A main picture for each location, then variants such as time of day, weather, aftermath or another angle. | Each main picture, then all variants together. |
| **5. Shot List** | A Scene Director plans every scene: its shots, camera, motion and the video model for each shot. Each scene appears on the canvas as a [Scene](https://nodaro.ai/docs/nodes/video/scene) node. | Each scene. |
| **6. Scene Images** | A keyframe for every shot. | The keyframes. |
| **7. Animate & Audio** | Every shot is animated, dialogue is voiced and lip-synced, music is generated to the target length, and an AI editor chooses the cuts. | In Manual and Guided, a dialogue timing check and a silent preview before the music is made. |
| **8. Final Merge** | The finished film. | The film. |

### What the critics check

AI critics check the work at several points. In Manual and Guided mode their findings are shown to you; in Auto mode they decide on their own.

- **Pictures of characters and locations.** A vision critic compares each main picture with its description. A failed picture is made again with the critic's advice, up to two times. A score below 5 out of 10 always counts as a failure.
- **The storyboard.** After the keyframes are made, a critic reviews them as a sequence, for example for a character whose look drifts, a location that changes, an unexplained jump in time of day or a gap in the plot. It only warns. When the storyboard is incoherent, the banner offers **Branch from Shot List**.
- **Every shot.** After each shot is animated, the video critic checks that the action matches the prompt, that the first frame continues the previous shot, and that there are no broken bodies, garbled text or motion glitches. A failed shot is made once more with the critic's feedback. If it still fails, Manual and Guided show **Skip** and **Regenerate** for that shot; Auto fails the run and refunds the credits not spent.
- **Match cuts.** When a shot is planned as a match cut and the critic finds a break with the next shot, the stage waits until you choose **Accept break** or make the keyframe again.

### What Animate & Audio does

1. **Animate.** Each shot is animated from its keyframe with the chosen video model. Scenes that need continuity run one shot after another, each starting from the previous shot's last frame.
2. **Voice.** Dialogue lines are voiced, and lip-synced when the shot needs it. When the plan has a narration script, for example in a trailer, a narrator track is voiced once for the whole film.
3. **Check the timing.** In Manual and Guided, if the real dialogue lengths are far from the plan, the panel shows a **Dialogue duration recheck** with a proposed rebalance to approve or reject.
4. **Preview without music.** In Manual and Guided, the panel shows a **Silent preview** so you can check the timing before music is made.
5. **Score.** Music is generated for the film, trimmed to the target length with a short fade-out, and its beats are detected. Shot lengths shift slightly so that cuts land on the beat.
6. **Edit.** An AI editor chooses a hard cut, dissolve, match cut or overlap for each join, and avoids cutting during dialogue. The final film mixes the music with the narration, if any, with the music lowered under the voice.

## Work with the canvas while it runs

The pipeline writes nodes to the canvas as it works. The canvas lays them out automatically and follows the build while you are idle. Pan or zoom to take control; click **Follow build** to let it follow again.

The border of each pipeline node shows what you can do with it:

| Border | State | What you can do |
| --- | --- | --- |
| Gray, pulsing | Running | Move it or rename it. Its settings are locked. |
| Amber | Awaiting approval | Edit its settings, approve or reject. |
| Blue | Approved | Edit its settings. The editor warns that later stages may need to be made again. |
| None | Forked | Anything. The node is fully yours. |

A small orange **stale** mark means that something the node depends on has changed, for example a character used in a scene, and the node may need to be made again.

If you delete or disconnect pipeline nodes, the pipeline pauses at the next stage and shows a canvas drift banner. Put the canvas back in shape and approve again, or fork the pipeline.

## Control the run

- **Re-run from here.** Starts a new pipeline from an earlier, approved stage, for example to redo the shot list with a different direction. The original pipeline stays as it is.
- **Cancel run.** Stops the pipeline. The credits not spent are refunded.
- **Fork pipeline.** Takes the canvas off the pipeline's hands. The pipeline stops at the next checkpoint, the credits not spent are refunded, and every node becomes yours to edit. This cannot be undone; to continue with AI help, start a new pipeline from the forked canvas.

If the service restarts during a run, the pipeline resumes on its own, up to three times. After that the run fails and the credits not spent are refunded.

## Refine the plan in Guided mode

In **Guided** mode, a chat opens next to the pipeline panel at the Script stage.

- **Ask for a change.** Write what you want, up to 8,000 characters per message. The director answers in one sentence and can propose a change card.
- **Apply or branch.** A proposed edit to the plan, such as a new title, a scene's length or a cast member, has an **Apply** button. For a structural change, such as a new genre or a different main character, the director suggests re-running from an earlier stage instead.
- **References stay intact.** A change that would remove a character, location or prop still used in a scene is rejected, and the director explains what to fix.
- **20 messages per pipeline.** When the limit is reached, approve, branch or switch to Manual.

After the final merge, Guided mode opens a second chat to review the film, with 8 messages. It cannot edit the film in place. It suggests which stage to re-run, for example the script, the characters, the shot list, the keyframes or the animation.

## Credits
The cost of a film depends on its format and length, the models, and how often you regenerate. When the pipeline starts, it reserves its estimated cost; the panel shows the estimate and what has been spent so far. Guided mode reserves a little more for the chat.

When you cancel or fork the pipeline, or when the run fails, the credits not spent are refunded. The helper buttons of a [Scene](https://nodaro.ai/docs/nodes/video/scene) node cost a few credits each, reserved when you click and refunded if the helper fails. See [Credits](https://nodaro.ai/docs/concepts/credits).

## Tips
- **Write the story, not the shots.** Give the characters, the setting, the conflict and the ending. The Showrunner and the Scene Director plan the shots.
- **Fix problems early.** A wrong character portrait carries into every scene. Reject it with feedback at stage 2 rather than at the end.
- **Use Re-run from here to explore.** Branch from the script or the shot list to try another direction without losing the first one.
- **Pin the models for a consistent look.** Choose an **Image model** and a **Video model** when every shot should share one style.

## From the API
Pipelines have their own API on Nodaro Cloud: `POST /v1/pipelines` starts a run, and other routes list runs, stream events, and approve, reject or cancel stages. See the [Pipelines API](https://nodaro.ai/docs/developers/api/pipelines). AI assistants can direct a film through the [Film Director](https://nodaro.ai/docs/mcp/film-director) MCP tools.

## Frequently asked questions

### What does the Story → Video node make?

A finished film from one story prompt. It writes the script, generates the cast, props and locations, plans the shots, makes a keyframe for every shot, animates them with dialogue and music, and edits the result. Everything appears on the canvas as nodes you can change.

### Do I have to approve every stage?

Only in Manual mode, the default, and in Guided mode. In Auto mode the pipeline runs every stage without stopping, and AI critics check each stage instead of you.

### How long can a Story → Video film be?

It depends on the format. A reel is 7 to 90 seconds, a commercial 10 to 90, a trailer 30 to 180, a short film 12 to 600 and a music video 30 to 600 seconds.

### What happens to my credits if I cancel?

The pipeline reserves its estimated cost when it starts. When you cancel, fork the pipeline, or the run fails, the credits that were not spent are refunded.

### Can I use Story → Video on a self-hosted install?

No. Story → Video runs on Nodaro Cloud only.