Story Creator: Build Interactive AI Stories Step by Step
The Story Creator inside MyBabes.ai lets you build episodic, panel-based stories starring your AI characters - with generated pictures, video, voice and sound on every panel. Publish episodes to the community and earn Tokens when readers unlock the ones you put behind a paywall.
This manual walks the entire workflow start to finish, using one worked example: a romance called Test Drive, starring an existing character (George, an ex-racer turned vineyard owner) and a custom side character named Sofia that we will create - and photograph - from scratch. Every example image in this guide was generated inside the app, and the exact prompt is shown next to it.
Made inside the app: image generation is portrait-only, but Seedance accepts a 21:9 ratio - so this hero is a single frame pulled from a 5-second 1080p Seedance clip. One generation, one perfect widescreen still.
How stories are built
Everything in the Story Creator fits one simple hierarchy:
A story is the top level: a title, a description, a cover image and a cast. Readers discover and follow stories.
Episodes are the chapters. Each has its own title, description, cover, visibility and - optionally - a Token price.
Panels are the frames inside an episode, read one after another like a visual novel. Each panel can carry narration or dialogue, a picture or a video, a voice line and a sound effect.
The cast are who speaks in the panels: your main character, side characters, a narrator and the reader themselves.
Your story is private by default. Nothing is visible to anyone else until you publish an episode.
Step 1 - Start your story
Pick your main character. Every story stars one of your existing characters. Their look, voice and profile picture carry over, and you can give them a story-specific display name - our character card says George Car, but inside the story he is simply George.
Name the story and the first episode. For Test Drive we opened with an episode called Arrival: “You arrive at the vineyard just before sunset.” Title and description are optional at this point - you can rename everything later.
Meet your starting cast. The moment the story is created, three cast members exist automatically: your main character, a Narrator for scene-setting, and You - the voice of the reader. The first episode opens with two empty scene panels ready to fill.
Every new episode you add later also starts with two empty scene panels - that is the blank canvas, not content you have to keep.
Step 2 - Build your cast
Five kinds of cast members can appear in your panels:
Main character - the character you picked when creating the story. Brings their portrait and voice. One per story.
Narrator - created with every story, for scene-setting text with no face attached.
You - lines the reader says. This is what makes the story feel like theirs.
Imported characters - any other character you have created joins with their portrait and voice intact. Great for reusing an established face across stories.
Custom actors - story-only side characters. They need a display name, an avatar image and appearance tags.
Creating a custom actor, avatar-first
Custom actors are built avatar-first: generate their portrait before you add them to the cast. Describe the character, pick an art style, and the picture is ready in about a minute. This is the exact prompt we used for Sofia, our vineyard hostess, in the realistic art style:
Save the same description as the actor’s appearance tags - they are her visual identity. Whenever you generate story art featuring Sofia later, the tags keep her face and style consistent from panel to panel. Below on the left is the avatar from the prompt above; on the right, the same tags reused in a completely different scene - new outfit, new setting, same woman:
Avatar prompt - realistic style
italian woman, late twenties, long wavy brown hair, off-shoulder sundress, warm confident smile, golden hour light in a vineyard, glowing skin, portrait
The avatar: generated once from the prompt above, in about a minute.
Same appearance tags + “elegant red evening dress, candlelit rooftop dinner, playful look over her shoulder”. The tags carry her identity; the scene text changes everything else.
Write tags once, write them well: physical features first (age, hair, build), then signature wardrobe, then the vibe. Avoid scene words like “standing in a vineyard” inside tags - scenes belong to each panel’s own prompt.
Step 3 - Write your panels
A panel is one beat of the story. There are three types: scene (a regular story beat), to be continued (closes an episode on a cliffhanger) and end (the finale of the whole story). Here is the opening of Arrival, panel by panel, exactly as we scripted it:
Panel 1 - Narrator: “The gravel crunches under your tires as the vineyard opens up before you, rows of vines glowing gold in the late sun.”
Panel 2 - Picture: a generated shot of the estate at golden hour. No dialogue - let the image breathe.
Panel 3 - George: “You made it. Perfect timing - the light is just right. Come on, Sofia is setting up the tasting.”
Panel 4 - Sofia: “So you are the guest George would not stop talking about. I hope you are thirsty.”
Panel 5 - You: “Wouldn’t have missed it for the world.”
Panel 6 - To be continued. The cliffhanger that points readers to the next episode.
Beyond text, one panel can hold a picture with its prompt, a video clip with its prompt, a voice line in the speaking character’s voice (with auto-play and volume settings), and a sound-effect layer - rain, an engine, a crowd - with its own controls. Mix them: a picture panel with ambient sound and no words often lands harder than a paragraph.
Panels have a running order and you can reorder, insert and delete freely while editing; the episode saves as a whole, so readers never see a half-edited state. One rule: a panel can only speak as someone who is actually in the cast - add or import the actor first, then write their lines.
Step 4 - Generate the media
Story art comes from the same generators as the rest of MyBabes, in two families of art styles: realistic (like Sofia above) and anime, which suits fantasy and dark-romance material. Pick one family per story and stay in it - style consistency reads as production value. The panel below was generated for a gothic story in the dark-romance anime style:
Dark-romance anime style. One prompt, straight out of the generator - used as a full-bleed picture panel.
Panel scene prompt - anime style
seductive vampire countess, long silver hair, black lace evening gown, moonlit gothic balcony, smoldering gaze, roses and candles
One prompt, two worlds
The style family is not a filter - it is a different director. Below, the exact same prompt (“raven-haired woman in a black lace dress, candlelit wine cellar, holding a glass of red wine, sultry gaze”) rendered in both families. Same words, different story:
Realistic: photographic light, real fabric, cinema skin.
Dark-romance anime: inked shadows, painted candlelight, graphic drama.
For motion, video is generated from a picture: create the still first, then animate it with a short motion prompt (4 to 15 seconds). Describe the movement, not the scene - the scene is already in the image:
Motion prompt - image to video
she turns toward the camera with a slow smile, wind catches her hair, candle flames flicker, gentle push-in
Voice - dialogue can be read aloud in the speaking character’s own voice; imported characters and your main character bring their voice with them.
Sound effects - generated from a plain text description (“distant thunder and rain on glass”) and layered under a panel alongside the voice.
Media library - every story keeps its own library of all pictures used across its episodes plus the cast’s avatars, so you can reuse art without hunting through your gallery.
How WAN 2.7 and Seedance changed the game
Until recently, video in stories was a garnish. The first generation of motion - the WAN 2.2 era - produced three-to-five-second silent loops at 480p. Useful as living illustrations, but a panel was still fundamentally a picture: the words did the storytelling, the clip just shimmered behind them.
Two model families ended that:
WAN 2.7 stretched clips to 3-15 seconds at 720p and 1080p, and - the real shift - generates its own synced audio. A WAN 2.7 panel arrives with breath, rustle, and room tone built in. It also supports one-tap motion templates, so a strong still becomes a moving panel without writing a prompt at all.
Seedance 2.0 (in Fast, Mini and full variants) brought 4-15 second clips with audio and multishot - real camera cuts inside a single generation. Wide shot, cut to close-up, cut to detail: film grammar in one clip. The full model renders up to 4K for hero panels.
Seedance 2.5 pushed duration to 30 seconds of multishot footage - an entire mini-scene, blocked and edited, from one prompt.
Extend chains any clip into a longer continuous sequence, so a scene can keep running across the length of a whole dramatic beat.
What this means for your stories
Panels stop being illustrations and become playable beats. A story panel used to be text with a picture; now it can be a fifteen-second shot where your character walks toward the camera, speaks into ambient sound, and the cut lands exactly where you want the reader’s pulse to jump. Layer the speaking character’s voice line on top of the clip’s own audio and a panel plays like a scene from a show - the reader taps, watches, replays.
It changes structure, too. Multishot means one panel can carry an establishing shot and a close-up - what used to take three picture panels collapses into one moving beat, so episodes read faster and hit harder. And it changes monetization: ending your free preview on a moving cliffhanger - her hand on the door handle, the camera pushing in, sound rising - converts far better than any still. Motion plus audio right before the paywall is the strongest unlock trigger the platform has.
Multishot prompt - Seedance 2.5
wide shot: she walks the vineyard row at dusk, lanterns flickering - cut to close-up: she glances back over her shoulder and whispers - cut to detail: her fingers trail along the vine leaves, fireflies drifting
Which model when? A practical split:
WAN 2.7 - the NSFW workhorse. Explicit-friendly panels with synced audio and motion templates, at 37 Tokens per second (720p) or 56 (1080p).
Seedance 2.0 Fast and Mini - cheap drafts and pacing tests, from 25-40 Tokens per second. Draft here, then re-render only the winners.
Seedance 2.0 - hero panels: the episode cover moment, the cliffhanger. Up to 4K when a shot deserves it (100 Tokens per second at 1080p).
Seedance 2.5 - long multishot scenes up to 30 seconds, 45-70 Tokens per second. One generation can carry an entire panel sequence.
The workflow that wastes the fewest Tokens: block the scene as a still first, draft the motion on Seedance Fast at 480p, and only when the movement is right re-render at high resolution - or hand the still to WAN 2.7 with a template and let it do the directing.
WAN 2.7 prompt - audio on
she leans in the doorway in a silk robe, slow smile, hair falling loose over one shoulder, candlelight flickering, fireplace crackling softly
Step 5 - Set your paywall
Each episode can be partly free and partly paid. You choose a free preview length - the number of panels everyone can read - and a Token price for the rest. In Test Drive, Arrival runs ten panels: panels 1-4 are free, and the remaining six unlock for 50 Tokens.
Set the paywall position. The free panels are your hook - end them on a moment that demands resolution, never after a resolution.
Set the price in Tokens. Readers pay once to unlock the rest of that episode.
Change your mind anytime. Move the paywall, reprice, or remove the lock entirely and make the episode fully free.
Buyers are protected: a purchase is tied to the exact version of the episode the reader bought. If you edit an episode after someone has purchased it, your changes create a new version - buyers keep the version they paid for and see a notice that a newer one exists, rather than silently altered content.
The last free panel of Arrival. Readers unlock what is behind the door.
Cliffhanger panel prompt - realistic
woman in a silk slip dress pausing at a bedroom door, hand on the handle, looking back over her shoulder, dim warm light, suspense
Step 6 - Publish and grow
Visibility is controlled per episode: private (only you) or public (everyone). Publish episodes one at a time as you finish them - one public episode and three private drafts is perfectly normal.
Publishing surfaces your story in the community with your public username attached. Public content passes the platform’s standard media moderation.
Readers move through panels at their own pace; the story remembers where they left off in every episode. At a paywall they see exactly how much remains and unlock it in one step.
After reading, they can rate the episode from 1 to 5. Ratings are for readers only - you cannot rate your own work, so every star is earned.
Prompt cookbook
Copy, adapt, and make them yours. Structure beats magic words: subject, wardrobe, setting, light, mood - in that order.
Actor avatar - realistic
scandinavian woman, early thirties, platinum blonde bob, silk blouse, confident smirk, soft studio light, editorial portrait
Panel scene - realistic
brunette in a black cocktail dress leaning on a hotel bar, neon city lights through the window, rain outside, cinematic low light
Panel scene - anime
elven sorceress, long copper hair, emerald gown with gold embroidery, enchanted forest at dusk, fireflies, soft glow
Motion - image to video
she glances back over her shoulder and laughs, hair swaying, city lights bokeh shimmering, slow orbit around her
The realistic avatar prompt, exactly as generated. Ready to become a custom actor.
The realistic scene prompt: neon, rain, cinematic low light - one generation.
The anime scene prompt: copper hair, emerald gown, fireflies - dark-romance style.
Common mistakes to avoid
Mixing style families mid-story. A realistic Sofia in episode one and an anime Sofia in episode two reads as a different character. Choose the family before you generate the first avatar.
Scene words inside appearance tags. Tags describe the person - “standing in a vineyard at sunset” belongs to the panel prompt. If a scene detail lives in the tags, it follows the actor into every picture you generate.
Narration walls. Three Narrator panels in a row is a novel, not a story panel flow. Break narration up with dialogue, pictures and the reader’s own lines.
Paywalling after a resolution. If the free preview ends on a satisfied moment, there is nothing to unlock. End it mid-tension - the door handle, not the closed door.
Expecting buyers to see your edits. Purchases are pinned to the version bought. Big revisions are better shipped as a new episode than as edits to a sold one.
Animating a weak still. Video multiplies whatever the image already is. Get the still right first, then add motion - never the other way around.
Quick tips
Write appearance tags for every custom actor before generating story art - consistency is much harder to add later.
One art style family per story: mixing realistic and anime panels breaks immersion instantly.
End free previews on a cliffhanger panel, and price the unlock while the tension is highest.
Use the Narrator sparingly; dialogue and pictures carry pace better than long narration blocks.
Generate the picture first, then the video from it - motion prompts work best on a strong still.
Keep episodes focused: a tight 8-12 panel episode reads better than a sprawling one.