The Story Generator: One Idea In, a Narrated Video Out

Most of the platform hands you one thing at a time: an image, a clip, a line of speech. The story generator is the one place you give it a premise and it plans the whole piece for you.

Which is either the most useful thing here or the most expensive way to be disappointed, depending entirely on whether you review the plan before generating it.

---

What actually happens when you press go

Six agents run in sequence, each handing its output to the next. Worth knowing, because it tells you exactly where to intervene:

Story Planner, writes a complete summary of the story before anything is broken into scenes. Its job is making sure the whole thing hangs together.

Story Director, breaks that summary into a scene-by-scene blueprint, and it works to a clock. Each scene is roughly 5 or 8 seconds depending on your video model, so "Mochi sees a butterfly" gets rejected as too vague to fill 8 seconds.

Script Writer, writes the narration, with character targets sized to how long each scene runs. This is why the narration actually fits rather than running past the picture.

Visual Blueprint Director, decides the look, and holds it steady across scenes.

Visual Prompt Composer, turns each scene into an actual image prompt.

Motion Director, decides what moves in each scene and how the camera behaves.

The whole orchestration costs 30 credits. That is remarkably cheap for what it is, six model calls to plan a video, and it's deliberately separate from the cost of generating the scenes.

---

What you choose up front

Length: 3 to 15 scenes. At roughly 5–8 seconds each, that's 15 seconds to two minutes. Start at 3. A 3-scene story tells you whether the tool understands your premise for a fraction of a 15-scene bill.

Visual style, photorealistic, cute anime, cyberpunk and others.

Image model and video model, chosen separately. This is the single biggest lever on cost: 15 scenes on an expensive video model is a very different bill from 3 scenes on a cheap one. See choosing a video model.

Narration, on by default, with a voice and a language. It runs through the same text-to-speech engine as everything else.

Characters, name, type (human, animal, fantasy, robot, alien, monster or custom) and appearance. Define them properly. A character with a described identity stays recognisable across scenes; one you left vague will drift.

---

Where to step in

Between the plan and the generation. This is the whole game.

After the agents run, you have scene descriptions, narration and prompts, all still text. Editing text is free. Editing it after 15 scenes have been generated means paying again for every scene you change.

So read the blueprint before you generate. Check that:

• the scenes actually tell the story you meant
• the narration says what you want, in words you'd use
• each scene has one thing happening: that's what fits in 5 to 8 seconds
• your characters are described the same way in every scene

Five minutes here saves more credits than any other habit in this guide.

---

What it costs

Three separate charges, and it's worth seeing them apart:

Stage	Cost
Orchestration (all six agents)	30
Scene generation	per image and per video, at your chosen models' rates
Final export	5

The middle row is everything. Orchestration and export together are 35 credits: the planning is nearly free. The bill is entirely decided by how many scenes you asked for and which models you pointed them at.

So: plan expensive, generate cheap, then generate again properly. Run the orchestration, read it, fix the text, and only then commit to the models you actually want.

---

What it's good at

• Short narrative pieces with a clear beginning, middle and end
• Children's stories: it's visibly tuned for them, down to the scene-pacing warnings
• Explainers where a sequence of scenes carries an argument
• Storyboarding, even if you never generate the video. Thirty credits for a six-agent scene breakdown is cheap as a planning tool on its own

What it struggles with

• Complex plots. Three to fifteen scenes is not a lot of room. Subplots don't survive.
• Precise dialogue. It writes narration, not a screenplay.
• Exact character likeness across many scenes, describe them tightly, and expect some drift over a long story.
• Anything needing a specific real product to appear. Use Canvas or image-to-video where a reference image holds the thing steady.

---

Try it

Open the story generator, write a one-sentence premise, set it to 3 scenes on a cheap video model, and stop after the agents have planned it. Thirty credits, and you'll have a full scene breakdown to judge before spending anything on pictures.

Then: the timeline editor for assembling what comes out · text to speech for the narration