Planning assets — setting up characters, locations and props

This is the step where you give every generated image and video the same reference for your characters and locations. Without this setup, faces and spaces can change a lot from one result to the next, so image and video generation in Production only runs once the preparation below is done.

Bring assets into other projects from My library

A project's asset screen shows characters, locations and props as separate groups, top to bottom. Within each group, cards sit in a dense horizontal row, so you can compare many assets at a glance without mixing types.

• Open “My library” under “My works” in the account menu to see all the characters, locations and props you've saved.
• Click “Import from library” in the top right of a project's asset screen to choose what to bring into the current project.

Switch between characters, locations and props with the tabs at the top.

• To keep an asset from the current project in your account, click the bookmark button in the top right of the card image

(“Save to my characters”, “Save to my locations”, “Save to my props”). The saved copy is never silently overwritten when you change the original project.

• To create a new asset, choose character, location or prop under “Create”, then fill it in on its planning screen.

If you opened the library from a project, add characters with “Cast in this project” and locations and props with “Import into this project”. The name and snapshot saved in the library stay as they are, and you can give the asset a different name inside the project. Within one project the same name means the same character, so if the project already has a different character with that name, casting stops and tells you — bring it in under a different name.

For props, only the master image and fixed appearance — shape, material, size — are copied. Owner, condition, damage and approval belong to the story context of a project, so they are cleared on import and set again in the new project. If a prop with the chosen name already exists, the two are never merged silently. You either copy it under a different name or choose to overwrite the existing prop yourself.

You get a warning before importing into a project with a different art style. When you “Save to my characters”, the project's art style (live-action/illustrated, styles to avoid) is saved with the character, and the two styles are compared when you cast it in another project. If they diverge — for example, bringing a character from a live-action project into an anime project — you're asked to confirm, with the reason, and “Import anyway” proceeds as is. After importing, we recommend regenerating the master in that project's look — this is not done automatically. Characters saved before this feature have no art style record, so no warning appears; save them again and the style is recorded then.

Aspect ratio — set it once and everything follows

The aspect ratio of images and videos is chosen per result in the spec bar on the Production screen.

The aspect ratio (landscape 16:9 / portrait 9:16) applies to production images, video clips and sequence sheet panels. For vertical video such as short dramas, keep it at 9:16 — no more portrait images paired with landscape videos. If you leave it empty, the project type's default is used (portrait for short dramas, landscape for everything else). Changing it leaves images you've already made as they are; it applies from the next generation onward.

A sequence sheet is for reviewing several shots on one page, so the sheet itself always comes out landscape; only the panel frames inside it follow the aspect ratio above.

Character sheet layout — portrait sheets for portrait projects

Choose Landscape / Portrait / Match project format under “Sheet layout” in the ⋯ menu at the top right of the character detail screen. The default is Match project format, so in a portrait (9:16) project the master, look and derived sheets are made in portrait layout (panels stacked vertically).

If you attach a landscape sheet as a reference to a portrait shot, the sheet gets shrunk and the face blurs — matching the layout puts the face in at a larger size. Changing the layout leaves existing sheets as they are; it applies to sheets you make afterward. The number of panels is set by the image model, regardless of layout.

Lists are card grids, details have their own screen

The character, location and prop tabs list assets as a card grid. Each card shows the master image, name (with tier for characters), a one-line description and a status badge (Master confirmed, Pick a candidate, Needs definition, etc.). Click a card to open that asset's detail screen. Use the bookmark icon on the image to save it to My library (works even without an image).

Characters that don't appear in any episode's script get "Not in script" next to their tier. Removing lines from the script doesn't delete the character automatically (images and settings may be attached), so review and clean them up yourself.

• The detail screen starts with a stage showing the master image large, and below it a grid of sheet cards (two rows for characters, “Looks” and “Sheets”;

night, dawn, rain and snow variants for locations). Pick a card and the stage switches to it; the selected card has a green border. Each card has a name and a status badge (Confirmed · Ready · Check passed · N warnings), and for characters you create a new sheet in the dashed “Create” slot at the end of the “Sheets” row (locations and props don't have this slot).

• Use the button row under the stage (desktop) to “Create”, “Regenerate” or “Upload”. If the project look rules

haven't been approved yet, a “The style isn't approved yet” confirmation appears once before generating — you can continue with “Create anyway”; approval happens under “Project look rules” at the top of the asset main screen.

• Choose the model for this generation in the “Model” box to the left of “Create”. The default is the project model set by the shared

“Image model” picker above the character, location and prop tabs. Changing it in the box applies to this generation only and is recorded on that result (the project default stays the same). For expression, motion and prop sheets and location variants, the box is locked to the master image's model — layout and tone are tied to the model. The list shows only models published by the operators; models without a key or that are suspended are locked.

• Generated images arrive as “Candidates” first — not just character masters, but looks, expression/motion/prop sheets, location masters,

location variants and prop master images alike. The stage turns into a comparison view with the current image and the candidates side by side; pick one and click “Use as master” (locations, props) · “Replace look with this” · “Replace sheet with this” · “Replace variant with this” and it becomes the image in that slot. Candidates you don't pick drop off the list, and their images stay in the archive. Until you pick, the current image and approval stay as they are. You can set the number of candidates from 1 to 4 (default 1) with −·+ next to the estimate, and if you want more to compare, add them with “Generate N more candidates” in the comparison view. Master image candidates for characters, locations and props come out as distinct alternatives from the second one on — they keep the fixed traits you wrote and interpret what the description leaves open differently (face shape and hair texture for characters, layout and materials for locations, silhouette and finish for props). Look, sheet and variant candidates must stay the same subject, so they aren't spread out this way. Nothing picks a winner for you automatically.

• Use the “Edit with a prompt” button next to “Regenerate” (also in the ⋯ menu) to revise the current image as a base — master, looks, expression/motion/

prop sheets, location masters, location variants and prop master images. Write at least one line about what to change (up to 600 characters), and attach reference images if needed (e.g. a photo of clothes to put on the character — this reference is used only for this request and isn't saved to the profile). The base image is always attached as reference #1, and you can attach only as many references as the model's limit allows after subtracting the base image and the references that are always attached (identity master, art style, etc.) — a one-line note in the dialog shows what's attached. You can change the model in the dialog; the default is the model that made that image (changing it here doesn't change the project default). Results arrive as candidates. The goal is to change only what you wrote, but depending on the model the base image can change a lot — review the candidates and pick.

• When editing a master, first choose where the result goes — “As a master candidate” (when adjusting only face, composition or expression;

picking it creates a new master version) or “As a new look” (when dressing the character or changing their physical condition; you enter a name). If you choose a new look, the master stays as is, and the clothing photo you uploaded is attached under that look's Outfit images, with results arriving as candidates for that look — e.g. upload a clothing photo and write "put this coat on" to get a new look wearing the coat. The master is the character's identity reference and must not have an outfit baked in, so this isn't detected automatically; the writer chooses.

• Definitions such as introduction, appearance, story, voice and performance are written below the stage. You don't write them in the list.
• The “Appearances” card at the bottom of the detail screen lists the episodes and scenes where this character, location or prop appears. Click one to go to

that episode. For props, the scenes you linked when registering the prop by dragging its name in the script are the basis — this holds even if you rename it. If it isn't linked to any scene yet, scenes where the prop's name appears are shown instead, along with a note that “Appearance count is by name”.

• Within one project, the same name means the same character. Even across several episodes, dialogue speakers with the same name

attach to the same character (even before it's registered, they're linked to the character used in other episodes), and creating a character under “Create” or on the planning tab with a name that already exists opens that character instead of making a new one. However, if several characters with the same name were created earlier, they aren't merged automatically — one of them is opened with a notice, and if they're the same person you can merge them in the character's Prep tab. Until then the “Appearances” card notes that the count is by name.

• Changing the tier (lead, supporting, extra) and deleting items you added yourself are in the row's ⋯ menu. Items extracted

from the script can't be deleted from the list.

• You also reach the same detail screen from the asset hub (“Assets”), the asset list on the Production screen, and generation-complete notifications.

The right panel is guidance

The right panel of the character, location and prop tabs (wide screens only) is guidance, not editing.

• “Still to do” shows how many profiles are empty, how many have no master image, and how many have a confirmed master,

and below it “Fill all at once” opens a window that fills empty fields with AI drafts (it runs only when you click).

• There are the four steps and tips of “How to keep characters consistent”, plus links to the setup and shooting guides. “Don't show again”

collapses the steps guide.

• The right panel on a detail screen is guidance for that one asset. Clicking an unfinished item (no master image, empty

appearance fields, voice not set, etc.) takes you to that field, with why it matters and tips. For characters it uses the same wording as the screen — Base look (master image) confirmed · 2 sheets left to check · `1 look from the story is missing` (a look change found in the script with no matching look · if it has never been checked it shows Not checked yet and is optional).

• The character detail screen shows one “Next step” above that first. Where the checklist says "everything

not done", this says the one thing to do now — if there's no master image, “Generate master candidates” (1 by default — to compare, “Make 2 more candidates”); once candidates arrive, “Pick from N candidates”; before confirmation, “Confirm this image as the master”; then “Check the image comparison” → “Generate an expression sheet” (leads) → “Check appearance changes in the story” (if never checked and the character appears in 3+ scenes · 1 credit per episode batch) → “Choose an everyday look” → “Create the “injured” look (3 scenes)” → “Regenerate N outdated sheets”, in that order. When everything's done it changes to “All set” and the button disappears. Clicking it takes you to that spot — it doesn't make images for you (the writer always clicks Create). Only “Confirm as master” (free) and “Look change check” (1 credit per episode batch) run directly. While candidates are being generated, it only shows progress and has no button.

• Turning on focus mode (⌘ . / Ctrl .) hides the panel too (Esc also exits).

There's no Save button — everything saves as you type

Planning screens have no Save button. About 0.5 seconds after you stop typing, your work is saved automatically, and “Saving…” → “Saved” briefly appears at the bottom of the screen and then fades.

• Nothing is saved while Hangul is still being composed. So that "ㄱ→가→각" isn't saved as separate steps,

it saves once after the character is complete.

• It's fine to move to another screen before it saves. If you move to an episode or the project home while your last keystroke

hasn't been saved yet, it's saved first the moment you leave.

• You're told if saving fails. If the internet drops or the server can't be reached, “Couldn't save” appears at the bottom of the screen

and stays there. What you wrote is still on the screen, so click “Save again” once you're back online. While this notice is showing, don't refresh or close the tab — your work isn't saved yet.

The Planning tab also has an outline — it follows the on-screen order exactly (unlike the character and location outlines, it isn't reordered — the outline should be a map of the screen). The outline header shows how many fields are still empty (“7 left”), and the right side of the “Project info” row shows how many fields you've filled (“1/8 fields”). The eight fields counted are project title, author name, logline, genre, target audience, synopsis, planning intent and differentiators; the project title is already filled when you create the project, so a new project starts at 1/8.

Episode loglines — how many episodes, and what happens in each

"Episode loglines" below Project info is where you plan episodes one line at a time. It's the same plan that the Co-writer conversation's "Full episode plan" card fills, so edits on either side show up in both. Creating or editing it costs no credits.

• Start — enter a target episode count (5–100) and click [Start · Free].
• Episode rows — each episode shows a one-line logline and its status (Planned, Drafting, Written and so on). Click the episode

number to expand it and write the next-episode hook, event summary and ending; click [Save] to store them (the one-line logline saves when you leave the field).

• Change the target count — edit the number in the target field and we first show how many episodes will be added or excluded;

it changes only when you click [Apply].

• [Import existing episodes] turns episodes you've already written into plan rows, and [Add episode] adds an empty row.
• ⋯ menu — go to or start this episode's script, "Exclude from plan" (stays dimmed; bring it back with "Restore to plan"),

and "Delete" (removes the row after you confirm — this can't be undone, and the episode's script isn't deleted).

• On mobile it's read-only.

Group locations into sets — the same house in the same tone

For locations with several spaces, like a house (living room, bedroom, kitchen), group them into a set. The flow:

1. Click “Use this location as a set” on the Locations tab, or, when names start with the same prefix like "Minsu's house living room" and "Minsu's house bedroom",

a "Group into set" suggestion appears automatically (they're grouped only if you accept — nothing is confirmed automatically).

1. Create the set master first — a concept sheet holding the shared reference: architectural style, materials, window frames, lighting character

(it isn't used for shots of individual rooms).

1. When you create each space's master, it's generated inheriting the set master — that's why the living room and bedroom

look like "the same house". A scene's shots and video reference that scene's space master.

1. When you change the set master, a “Set changed” marker appears on the space's detail screen and list row, so you can see which spaces need

to be regenerated.

Single-room locations like a café or a street don't need grouping; just use one master as before. Areas smaller than a room, like a dining table, shouldn't be separate spaces; handle them as details on that room's (the kitchen's) sheet. Times of day and weather such as night or rain aren't separate locations; make them as variant sheets. You can change or remove a space's set under “Location group (set)” on the detail screen, and a set that still has spaces in it can't be deleted (unlink them first).

Write character and location descriptions in their own tabs

The character and location description sections that used to be on the Planning tab are gone. They meant the same character was managed twice — description only on the Planning tab, images on the Characters tab.

Now you write the one-line introduction and profile on the character or location detail screen. Those values

• appear in the planning document PDF (see below), and
• go into the sheet and image generation prompts too.

All of one subject's information lives in one place.

Fill empty character and location profiles all at once

If the script picked up many characters and locations, click “Fill all at once” in the right panel of the Characters or Locations tab. This prompt appears when there are empty profiles, and also while drafts are being filled or failed items remain (it then changes to “View progress” · “Try again”). No AI is called until you click the button, and it never runs automatically just because you saved the script or opened the tab. If you come in through “Fill all at once” on the asset screen, the window opens only once, and if there are no empty fields to fill and no images to make, you get a one-line note instead of the window (it doesn't pop up again when you come back to the planning screen).

“Regenerate” makes a different picture every time — the first generation uses a fixed starting point (seed) per subject, but regenerating when an image exists uses a new starting point. A character profile's “Introduction” is still included in the prompt after the appearance DNA is filled, so editing the introduction and regenerating reflects it (though the canonical source for appearance is the DNA fields — edit those for face, hair and outfit).

1. In the selection window, check which characters and locations to fill. Items from the script, items you created in planning,

and even extras are all selected at first. Just deselect the ones you don't need.

1. Click “Fill” and characters' appearance and voice directions, and locations' space and shooting fields, are added to the

draft queue in turn. Each item shows a Waiting, In progress, Done or Failed status, and only failed items can be retried. The queue keeps running even if you close the window or visit another tab, and when you reopen it, progress and failures are still shown. “Stop remaining N” stops items that haven't started yet and keeps what's already filled. “Try again” resends failed items with the same input, and you can select items not yet queued and fill them too. With many items you may hit the server's per-minute request limit; in that case they aren't marked as failed but pause in “Waiting” and resume automatically.

1. AI drafts fill only empty fields. The introduction, appearance, voice and location settings the writer already wrote are never

overwritten. Items where a draft was applied show “AI draft” in the source label on the detail screen.

1. When filling is done, you see the number of characters and locations without images and the total credits needed.

Click “Generate images” and master images are queued the same way as a single generation on the detail screen. Items that already have an image are skipped; to regenerate, run it one at a time from that detail screen. If your balance is short, you see the credits needed instead of the generate button. If all profiles are filled but items without images remain, the notice at the top of the tab changes to “N characters · M locations without a master image”, and “Generate all images” reopens this step directly. The server generates at most 6 images at the same time per person, so if you queue more, the rest wait in “Waiting” and start in turn as earlier ones finish. If you're also generating in another tab or device and there's no free slot, it doesn't fail — it waits until a slot opens (up to about 10 minutes).

Filling a profile costs 1 credit per item (the button shows “Fill N drafts” with the total). Image credits are deducted at the usual rate each time an image succeeds, and failed ones follow the usual refund rules.

Continue lead and supporting prep

At the bottom of the same window is the “Continue preparing leads and supporting roles” table. The steps to make a character ready for production follow an order (definition draft → master candidates → pick a candidate → confirm master → expression sheet → usual look → check look changes), and the table shows, one row per character, which step they're on now · what happens next · whether it's a decision for the writer. Extras need only a text definition, so they aren't in the table. Click a name to go to that character's screen.

• Click “Continue” and the next automatic step for each character is added to the generation queue. The steps that can run

automatically are definition draft (1 credit) · master candidates · expression sheet · check look changes (1 credit per episode batch).

• Picking a candidate, confirming the master and setting the usual look stop the flow. The writer picks the reference

image — there's no feature that picks a winner automatically. When that character's automatic steps finish, a notification such as “Candidates ready” appears with a “Go pick” link.

• A paused character's next step doesn't continue automatically. After you pick or confirm, reopen this window

and click “Continue” once more (so credits aren't spent without you knowing).

• Below the table you see how many images and credits it will take to finish prep (1 candidate × characters + expression

sheet × leads). If your balance is below that total, you see the credits needed instead of the button.

• Expression sheets are made only for leads. Supporting characters need only a master sheet.
• Checking look changes is recommended only when the character appears in three or more scenes (1 credit per episode batch).

Lead characters appear with images in the planning document PDF

The planning document PDF (Project home → [Export] → Planning document only / Planning document + script) includes only characters set as leads, with their main image (2026-09-07). Supporting characters, extras and locations aren't included — a planning document is for readers to grasp the characters quickly, so it focuses on the leads.

• Set a character's tier on the character screen. Unassigned counts as supporting, so promote a character to lead to include them in the planning document.
• Character image: the approved master sheet if there is one, otherwise the main image.
• If there are no leads, the character page is empty, and the export dialog tells you in advance.

Minimum prep required for generation

When you set the tier (lead/supporting/extra) on the character detail screen (or in the ⋯ menu of a list row), the readiness badge tells you the next step as "Needs: ○○". Even with many characters, as in a feature-length work, there's no need to make sheets for extras.

Extras go in as extras on their own

When characters are extracted from the script, entries that are roles, numbers or sounds rather than names — a voice on a walkie-talkie · Gang member 1 · Police officer — go in as extras from the start. Video can be made without a face sheet, and no one-line introduction is required.

• In the character list, they're grouped in the “Extras” group at the bottom. You can change the tier from the row's ⋯ menu.
• If the call is wrong (it looked like an extra but the character has a real role), promote them to supporting or lead from the ⋯ menu or the detail screen.

You can also demote a character picked up as supporting to extra.

• This also applies to existing projects. However, characters that already have sheets or an introduction

are left alone — characters you've touched stay as they are.

The same character doesn't appear twice

@조직원 1 and @조직원 1 (O.S.) (e.g. @GANG MEMBER 1 (O.S.)) count as the same character even with the parenthetical. Standard extensions such as (O.S.) (V.O.) (CONT'D) and Korean script extensions such as (소리) (목소리) (필터) (전화) (sound, voice, filter, phone) are stripped from the name.

Unknown parentheticals aren't stripped — 민수 (젊은) and 민수 (늙은) (Minsu (young) / Minsu (old)) stay as separate characters. Two different characters merging into one is harder to undo.

Project look and production rules — open them right where you make assets

The project-wide art style and color, materials and exclusions, and video sound policy are edited in [Project look & rules]. The button is in two places — the top right of the character, location and prop tabs (next to the image model), and the spec bar on the Production screen. Either one edits the same single set of values. New screenplay projects come with defaults you can edit. When generating character sheets and location and prop images, the same project look and the production rules that apply to images are included automatically.

Character sheets default to a live-action style. They're drawn in another style only when you write words like “Animation”, “Illustration”, “Webtoon” or “Comic” in the art style field; if the field is empty or the wording doesn't name a style, like “Muted colors”, they're made live-action. If your project will be illustrated, say so in the art style field — the character sheet format itself leans toward illustration, so if you don't specify, a live-action instruction is added to hold it in place.

Approval is a record marking the current version as the production reference. The values you enter are used for generation even before approval. For the full list of fields and how the mood board works, see the Production guide.

Project style — set with images, not words

The Project style tab in [Project look & rules] runs top to bottom: approval bar → project style → style images → style settings. The Production rules tab also has its approval bar at the top (Approve this style / Approve these rules ↔ Remove approval once approved).

• Project style — the same card grid as when you create a project: presets for the project's format (short drama: Modern

K-drama · Modern cinematic / short anime: TV anime · Feature anime · K-webtoon · Watercolor), your own styles, and Create from images at the end. Picking a card copies its style images and settings into this project. If character or location sheets already exist, you're asked once before switching. Edit or delete your own style from the card's … menu (only the library changes — this project stays as it is).

• Which card shows as selected — when the project's values match the style you picked, that card is selected. If you

change the images or settings after copying, the selection clears and Save as my style appears. If the style this project came from is deleted from your library, an Original deleted label and Save as my style again appear. Saving turns the project's current images and settings into a new style as they are (the default name is "original name copy").

• Style images — up to four. Under each one, choose character, background or color (JPG or PNG, up to 10MB each, 1:1

recommended). Projects that already had up to six keep showing them; you can add more once you're below four. Drawings of people are expected, so you're asked once whether an image shows a real person (photos of real people aren't accepted).

• Style settings — rendering/tone, color, proportions and background tone; expand for line art, shading/lighting,

camera/lens feel and things to avoid. Fill empty fields with suggestions fills only empty fields from the chosen preset (or the format default) and uses no credits.

• Each preset also has production rule defaults (light color temperature, key light direction, contrast, materials to emphasize

or suppress, camera moves) that are laid down in the Production rules tab when you create a project. Switching cards in the project look doesn't change the production rules.

Art style images are attached when character, location and prop sheets are made (in anime projects, right after the master reference and before outfit and other reference images). They aren't attached again to images and videos on the production screen — the sheets already carry the style, and attaching it every time lets faces from the style images leak into your characters. So after changing the art style, remake the sheets for them to follow it.

If you write a studio, artist or title name in the art style field (such as "in the style of …"), you get a warning, and anime projects drop that name when generating. Describe only what you can see (line weight, shading steps, colors).

In anime projects, the image check that runs when you approve a character master also reads a 3-tone color script (base, shadow and highlight for hair, skin, eyes and outfit). There's no field to fill in. The colors are added as text when that character's other sheets and the production screen's images and videos are made, so shadow colors drift less from scene to scene. They're read again when you remake and approve the master. The color swatch panel on the master sheet is also drawn as 3-tone swatches (only on models that draw 12 panels).

Voice — how the character sounds when speaking

In Voice (video) on the character detail screen, write the timbre, speech habits and intonation, speaking pace and temperament. These four are bundled as the character's voice constants and included in video dialogue as voice acting directions. In temperament you can also note conditions where the voice changes, such as under pressure. If you leave them empty, the same voice reference can't be passed along.

This is different from the speech style above. Speech style is the written style of the dialogue (casual speech, short sentences); voice is how the model sounds when it delivers that dialogue. This isn't a post-production dubbing feature.

The voice sample is one audio clip in the same section. When you generate with Seedance 2.5 it is sent along automatically as this character's timbre reference, which stops the voice changing from shot to shot (only timbre and intonation are referenced; the prompt decides what is said). There are two ways to add one.

• Upload audio — WAV or MP3, 2–30 seconds, up to 15 MB. A real person's voice requires their consent.
• Design voice with AI · 10 credits — make one voice candidate from a description (age, gender, timbre and manner in one or

two sentences) and listen to it. The description is pre-filled from the identity, apparent age and voice constants, and you can edit it before sending. Candidates are made one at a time, take about 30 seconds, and each one is charged. Click “Use this” on the candidate you like to link it as the sample. Every candidate you made stays in the archive. Regional accents aren't reflected well yet. Descriptions that refer to minors or real people are rejected by the AI provider's safety policy (e.g. “17-year-old high-school girl”) — no credits are charged then; drop the age or make the character an adult and try again.

Video models other than Seedance 2.5 don't take a voice sample — only the voice constants are sent then.

Performance habits — the same person across scenes

In the Performance habits (video) field on the character detail screen, write physical habits — habits and the moments they show up, gait and posture, the face they show others, and what leaks out when they break down.

Unlike voice, these values aren't copied straight into shot prompts. The AI reads them when building the video plan and rewrites them for the scene — someone who twists their ring when nervous has that habit translated into how it shows up in an interrogation scene. If you leave it empty, the plan imagines it anew each time.

Character setup steps

1. Enter a description (Characters tab) — a one-line introduction and a detailed profile. Characters who appear in the script are

added to the list automatically.

1. Fill in the character definition (appearance) (Characters tab) — ⭐ the step that affects result quality the most.

Write identity (age, gender, role), face shape, eyes, hair, build and default outfit. If this is empty, the AI imagines those parts anew each time you generate, and you get a different person every time.

• The detail screen shows a Core N/6 badge, and if fields are empty, a warning tells you what's missing.
• Click Fill empty fields with AI and it reads the introduction and profile together with the project's genre, logline, synopsis and look

to draft a character who belongs in that world — it fills only empty fields and leaves values you wrote as they are. Review, edit and confirm the draft. For live-action projects it's as specific as an actor you could actually cast; for anthropomorphic or fantasy projects, the non-human parts go separately in the “Non-human features” field below.

• Non-human features (optional): if you mix the "non-human parts" of an anthropomorphic or fantasy character into free text in face or hair,

the whole face tends to become an animal or the feature gets drawn as a prop like a headband. In this field, for each item write body part · position (what it replaces) · count · shape (e.g. ears — in place of human ears, count 2, upright lion-like, short fur matching the hair color) and when the sheet is made it's pinned down as "body parts growing from the body, exact count, human everywhere else", and automatic checks gain a “Declared traits” item whose badge tells you whether it was drawn as specified (errors like a headband or four ears). Leave it empty for human characters — if it's empty, generation works exactly as before.

• If you've attached reference images, it reads the pictures too (reference images in the Detailed appearance section,

up to 3). Appearance read from the pictures takes precedence over values inferred from text. It costs no extra credits, and if the pictures can't be read, the text-based draft still comes out.

• Once set, all later sheets, images and videos are generated from this definition.
• The “One-line introduction” is for the story; the “Base appearance” is for the pictures. The introduction is text read by the script and the co-writer;

the canonical source images and videos follow is the base appearance. Once the base appearance is filled, editing the introduction doesn't change the pictures — they follow the base appearance (the introduction is included only as a supporting sentence).

• When you confirm a master image, the picture and the text are compared automatically (no credits). A line such as

3 empty fields can be filled from the image · 1 field differs from the image appears above the base appearance, and “View by field” shows Image / Current text side by side for each field.

• Each field has “Use image value”, and the line has “Fill empty fields only”. “Use image values for all”, which overwrites

values you wrote all at once, is in the ⋯ menu and asks for confirmation once more.

• The text is never edited automatically — the confirmed picture is the reference, so it only suggests changes toward the picture,

and the writer is the one who clicks. Fields you apply drop off the line.

• Confirming a new master image runs the comparison again. If results judged against the previous image remain, they're

marked “Previous master image”, and you can check again with “Compare again (free)” in the ⋯ menu.

• It's an AI judgment and can be wrong. If the comparison fails, the confirmation still stands and no line appears.
• Props are handled in the Detailed appearance section. Project props registered on the Props tab are toggled on and off under “Linked project props”,

and anything written comma-separated in “Other text props” can be turned into a project prop with the box-shaped button next to each item — an item is created on the Props tab, linked to this character, and removed from the text list. If a project prop with the same name already exists, a new one isn't created; it's just linked, and you're told so. There's no button to move everything at once — if names that end up on every character, like “Phone”, all became project props, the list would be unusable, so you pick them one by one.

1. Generate the master sheet (Characters tab) — the identity reference image (10 credits per character).

If you have a fictional character image made elsewhere, you can attach it as a reference image to give an identity anchor (output from other AI tools, commissioned illustrations, your own work). Photos of real people aren't supported yet — you're asked when attaching, and if you answer that it's a real person, the file isn't saved and the upload is canceled. “Create” and “Regenerate” make candidates (1 by default, up to 4), and the stage switches to a side-by-side comparison view — if there's a current master, it sits in the “Current” slot on the far left. Each candidate carries a check badge, model and seed; pick one and click “Use as master” and it becomes the new master version (candidates you don't pick drop off the list, and their images stay in the archive). After picking, the “Confirm as master” → approval sequence is the same as before, and the approved master becomes the reference for all later generation.

• Change the number of candidates between 1 and 4 with −·+ next to the estimate. The estimate

defaults to 1; when you raise the count it shows the total, like Create 3 × 10 credits = 30 credits, and if your balance is below the total, a top-up prompt appears. Each candidate uses a different seed, so 3 candidates are 3 different pictures.

• While generating, the master item in the filmstrip shows progress like Candidate 1/3, and each time one finishes,

only that slot in the comparison view fills in. It keeps generating while you do other work.

• If there's nothing worth picking, clear them with “Discard all candidates” in the ⋯ menu (images stay in the archive).

Nothing picks a winner for you automatically — the writer picks the master image.

• If you regenerate while an approved master exists, the approval stays in place until you pick and confirm a candidate

(candidates aren't the reference yet).

• If there are two or more versions, a version list appears below the button row. Click an entry to open

the same comparison view with “Compare with current”, view them side by side, then choose “Use this version”.

• Below the “Create” / “Regenerate” row, a line like References attached: master · reference 1 shows **the reference images

actually attached to this generation**. The order is the priority (master first, then reference images in order). Each image model accepts a different number of references at once, so anything past the limit is noted, e.g. (reference 2 excluded — this model's limit is 3) — this is so references never drop out silently and leave you wondering "why is the face different". If the model doesn't accept references, it says only the written definition is applied. The location screen works the same way (the master inherits from the set; a variant has that location's master first).

• The master sheet is drawn in the order of angles video needs — front full body → 3/4 full body → side

full body → back → front face → 3/4 face → the rest. With image models that fit fewer panels on one page (4 panels), only the four angles go in and the face close-ups are dropped. Video often uses back and 3/4 angles, and without them the model imagines those angles.

• The moment you approve, a face reference image is created. It's a single square cropped from the master's front face panel (or, if there isn't one, the head

of the front full-body panel), shown as “Face reference” next to the master in the filmstrip. It's view-only — there's no Create, Regenerate or Delete, and it's recreated when you re-approve the master. It costs no credits (no model is called; the server just crops). If it fails, the approval still stands and the item stays as “None”. If you roll back a version, that version's face reference comes back with it. It's not used in production (shots, video) yet — production references are still one sheet.

• When you generate or upload an image, an automatic check runs once and adds a badge (master, derived

sheets and looks alike). It checks five things — does the panel count match the spec, is every panel the same person, is the outfit the same in every panel (for a look, is it that look's outfit), is there embedded text such as Hangul or Chinese characters or a fake screen, and does the art style match the project look (live-action/illustrated). It shows as Passed or N warnings in the filmstrip and on the stage, and you can read the reasons by hovering over the badge next to “Confirm this image as master” or in the line below the button row. They're also gathered in the right guidance panel as “N check warnings”. It costs no credits (free). And it doesn't block approval — which image to use as the reference is the writer's call. The check is an AI judgment and can be wrong, so treat the reasons as guidance (especially "same person" judgments). To check again, click “Check again” in the ⋯ menu at the top right of the stage. If a check couldn't run, the image is left without a badge and remains valid. Uploaded images weren't made to our spec, so the panel count isn't judged.

• You can still make changes after approving. Regenerating removes the approval and creates a new version, and

previous versions don't disappear — click one in the version row below the sheet to roll back and re-approve. When you change the master, looks and derived sheets already made for that character get an Old appearance marker, and the top of the detail screen names what's out of date. Regenerate only what you need (nothing is regenerated automatically — it costs credits).

• If anything is out of date, a line appears below the sheet row —

Master changed — 3 to regenerate · est. 6 credits — order: usual look → expression sheet → other looks. Click [Choose what to regenerate] and a list appears with everything checked by default. Uncheck an item to leave it out, and the rest are made in order from the top (with the total credits shown). The order matters: the Everyday look is the default for every scene with no outfit written, so it's used most widely, and the Expression sheet goes in as the face for every shot — even if credits run out partway, the ones with the biggest impact are updated first.

• If the video model in the operating settings requires pre-registering character assets, registration starts

the moment you approve, and the detail screen shows Video asset ready. If it shows Rejected, that image is hard to use for video, so regenerate the master and approve it.

• This marker also appears on the look and sheet items in the filmstrip. Video ready

or Video rejected appears at the top right of the thumbnail, and hovering over a rejection shows the reason. It's there so you don't find out about a rejection only when submitting the video (unregistered items show no marker).

1. Sheets — sheets that keep the master's identity and art style and change just one axis.

The axes are expression · motion · props, and you can make several sheets on the same axis (e.g. “Motion: Fight” and “Motion: Café routine”). Character images are made on two axes: “Look × Layout” — the look is "how this character appears right now" (base look, hospital gown, injured…), and the layout is "which sheet shows it" (master-style, expression, motion, props). So there's one place to make them:

• The filmstrip has two rows. The top row, “Looks” (base look (master) · face reference · your registered

looks), and the bottom row, “Sheets” (sheets made from the look selected in the top row). Pick a look and the bottom row switches to that look's sheets, and the stage shows the selected item as is.

• There's one add button: [Create]. Click it and choose, in order, ① which look (base look / a registered look /

new look) ② which layout (master-style look sheet / expression / motion / props) ③ what to include, and an estimate like 1 image · 8 panels 4×2 layout · 2 references · 10 credits appears below. If you chose a new look, enter its name, outfit, physical condition and outfit image right there, and when you click [Create], the look is saved first and then generation starts.

• Layouts you can't choose are locked with a one-line reason — the master-style sheet of the base look

is the master itself (made on the stage), and axis sheets can't be made from a new look or a look without a sheet (that look's sheet is the reference). It's generated on the server with the same model as the master. You can do other work while it generates. Results arrive as candidates, and you have to click “Replace sheet with this candidate” for one to become the sheet (versions don't stack up like the master's — candidates you don't pick stay only in the archive). If it's worth keeping, make it a new sheet. Outfit variations are handled not by sheets but by looks (6 below). While the first candidate for a new sheet is being made, an empty slot showing Candidate 0/1 appears in the “Sheets” row first.

• You can choose which look to base it on. The default is the master (base look);

choose a look and the sheet is made referencing that look's sheet — that's how look × axis combinations like "an expression sheet in the injured state" come out. Outfit, hair and physical condition wording come from that look, and face and build come from the appearance definition (the layout is set by the axis).

• If the look you want as the base has no sheet yet, you can't make it — you're told to make that look's sheet

first (otherwise it would be made from the master by mistake).

• The default title includes the base look (Expression: Sobbing · Hospital gown). A Look · Hospital gown

marker also appears next to the sheet name.

• If you regenerate the base look's sheet, sheets made on top of it get an Old base marker.
• The “Expression sheet” required by character tiers and public candidate review counts only sheets based on the base look (master).

An expression sheet based on the hospital gown is used only in that look.

• Next to each sheet name is a badge like "8 panels 4×2" — the spec fixed in advance for how many panels that sheet

draws in what grid (hover over it to see the panel list). The panel limit is how much the image model published by the operators can fit on one page, and within that it's set to the number that fills a landscape sheet with no empty slots — so it can be one or two panels below the limit. Per-model limits depend on the operating settings; check the badge on screen. A larger character for the same credits is more useful for production than drawing more panels and leaving empty space. The panel list is in order of importance, so only the ones at the end are dropped.

1. Look changes in the story — above the “Looks” card on the character detail screen. If it's never been checked,

only [Check now (free)] is shown. Click it and it reads scenes where this character appears and that have action lines, split by episode, finds passages where outfit, hair or physical condition change, and records them.

• It costs no credits (it's a judgment that reads only the script's action lines). With many episodes it's split per episode and

runs in turn, and it continues even if you leave this screen (progress shows in the generation indicator at the bottom right of the screen). Results are saved to the project and visible on other devices too.

• When the check finishes, each item comes with its source scene and a state description. If a matching look already exists it shows

Look: handled by Hospital gown; otherwise [Create this look] appears. Click it and a new look is added with its name, outfit wording and physical condition chips prefilled, and you're taken to the editor — the writer clicks to generate the image right there.

• After editing the script, use [Check again (free)] in the ⋯ menu at the top right of the card. Checking again

replaces that episode's old results with new ones, and scene states you wrote yourself stay as they are.

• If no change is found, it stays as “No changes in outfit or physical condition” — meaning one usual look

is enough. You can register looks directly under “Looks” below even without checking.

1. Editing looks — register looks that appear in the story ("Hospital gown", "Episode 1 suit", "After escaping the hospital") and

make sheets for them, and in production, when a scene's outfit matches the name, that look's sheet is automatically referenced. It's the key tool for projects with many outfits (especially feature-length ones).

• A look sheet is made to the same spec as the master sheet (same panel count, grid and per-panel poses) and

changes only outfit and physical condition. Every panel must wear the new outfit, and the default outfit sentence from the appearance definition is removed from the look prompt (and hair too, if the look sets the hair). It must be made with the same image model that made the master for the panel counts to match.

• You can describe the outfit in text or upload a photo of the clothes. Clothes uploaded as the “Outfit image”

are put on as is (the person in the photo isn't copied). If you provide both text and a photo, both are applied. You can also upload just a photo and leave the text empty.

• Each image model accepts a different number of references at once. The master takes the first slot,

so on models with a limit under 2 the outfit image can't be attached — in that case you're asked before generating, "This model can't reference the outfit image" (if you generate anyway, only the written outfit is applied). Change the model in the project's production settings.

• Past the limit, items are dropped from the end. The order is master → outfit image → character reference

images, and the line next to [Create] shows what's actually attached and what's dropped.

• You can also choose a physical condition — injury · bandages · blood · wet · dirty · aging · disguise ·

pregnancy · weight change · scars (multiple choices allowed) + a one-line detail ("bandage on left arm"). You can make images of an injured or wet character, and this wording is also included in production prompts.

• Conditions marked · face (injury, blood, aging, disguise, scars) affect the face. Then the prompt pins down "layer it

on top of the same face" — face shape, eye spacing and apparent age stay the same, and only wounds and bandages are added (only aging changes apparent age too).

• The image model may refuse depictions of blood or injury. In that case, tone down the wording or try

a different model.

• The one place to add a new look is the [Create] button at the top. The “Looks” card is for editing existing

looks (name · outfit details · hair · physical condition · outfit image · set as usual look · delete).

• Set one look as the usual look. Scenes with no outfit written use that look —

if you don't set one, the model imagines the clothes and they change from shot to shot.

• The outfit is decided in this order: shot setting → that scene's setting → carried over from the previous scene → look match → usual

look → the default outfit from the appearance definition. A “Look match” applies even when the name doesn't match — if a physical condition written in the scene ("bandage on left arm") or outfit wording overlaps with a look's condition or details, that look is attached. That's how character state changes found in the script carry through to the images.

• In the character list on the Production screen, looks are shown with their condition, like “Look: Hospital gown (bandages)”.

Only one image is attached per character, so if a look matches the scene's state, that look is attached before expression and motion sheets.

• You can make derived sheets (4) based on this look. In production, when a scene's state matches this look

and the shot calls for an expression or action, the axis sheet made from that look is attached first (if there isn't one, the look sheet, then the base look's axis sheet, then the master — one per character).

When a character is decided in a co-writer conversation, you receive a suggestion with the name and introduction plus an appearance draft (identity, hair, outfit, signature color, key prop) — click [Apply all] and steps 1–2 start mostly filled in. The AI doesn't overwrite values you wrote yourself.

When you change a character's definition, existing sheets show an "Update needed" marker — update them with each sheet's “Regenerate”.

Using external images directly (upload)

If you already have images made elsewhere (output from other AI tools, commissioned illustrations, photographs), you can upload them and use them as is instead of generating — for characters (master, looks) and locations (master image, time-of-day variants), with no credits. Uploads come in as new versions just like generated images and go through approval, marked with an "Uploaded" badge (they have no seed or prompt, so they can't be remixed). You're responsible for copyright and portrait rights of the images you upload (you're told once, on your first upload).

When you upload to a slot that can contain a person, you're asked once, "Is there a real person in this image?" Fictional characters made elsewhere (AI-generated, illustrated, your own work) are used as is; photos of real people aren't supported yet, so the upload is canceled (the file isn't saved).

Why is this blocked: video models block images with real faces by policy, and identity verification and rights management processes are still being prepared. The official path for using contracted actors will open once preparation is done — details are in the Real person (actor) shooting guide.

Choose video specs in Production

In the spec bar on the Production screen, choose model, aspect ratio, length, quality and sound for each result. When you change the model, only the values that model actually accepts remain, and the estimated credits update to match the selected specs.

Higher quality costs more credits. The exact difference depends on operating rates and the model, so check the estimated credits on the Production screen. Starting drafts at the default quality is the safer choice.

Location setup steps

1. Enter a description (Locations tab) — locations from the script's scene headings are added to the list automatically.
2. Generate the master image (Locations tab) — a reference sheet with an establishing shot, reverse angle and details

(10 credits per location). You can attach reference photos. Results arrive as candidates, and you have to click “Set this candidate as master image” for one to become the master image (the same flow as a character master — see “Lists are card grids…” above).

1. Video elements — fill in day/night, lighting, weather, color and spatial layout, and they're reflected in the prompt.

• Like the character detail screen, there's Fill empty fields with AI — it reads the location name and notes to

draft them. It never overwrites values you wrote and fills only empty fields (free).

• If day/night and lighting in particular are empty, the same location looks different in every shot — start here.

1. Time-of-day and weather variants — generate night/dawn/rain/snow versions referencing the master image, and

production reads the scene heading ("INT. 카페 - 밤", i.e. INT. CAFE - NIGHT) and automatically references the matching version. Variants also arrive as candidates and are confirmed with “Replace variant with this candidate”. “Generate N missing variants” makes 1 candidate per slot.

A scene's location is linked automatically based on the script heading — you only need to create the location profile.

Prop master images

On the Props tab, write the shape, material, color and the condition at first appearance, and make the master image on the detail screen's stage. One image with front, back, side and a material close-up in four panels is that prop's reference (you don't make several images per condition — damage and changes of owner are written per scene in production). Results arrive as candidates, and you have to click “Set this candidate as master image” for one to become the master image.

• Like the character and location detail screens, the “Ready” tab on the right has “Fill empty fields with an AI draft”

(shown only while a field is empty). It reads the prop's name and the fields you already wrote, the script scenes where the prop appears, and the project's genre, logline and look, then proposes the one-line definition, shape/material/ color, shape/size, on-screen text, wear and damage. Values you wrote are never overwritten — only empty fields are filled. On-screen text is filled only when the script or your input actually contains letters, and the owner is not drafted (you pick the character yourself). 1 credit for non-members, free with a membership.

If you confirm the art style in [Project look & rules], prop master images follow that style too — mood board images uploaded to the project look are included as reference images. What's attached, and what's dropped because of the model's limit, is shown in a line below the [Create] button (wide screens only).

Per-episode plans are made in co-writer conversations

How many episodes in total, and what happens in each, is decided in conversation with the co-writer. Choose “Full episode structure” as the writing mode and it asks for a target episode count; apply the structure card that follows and that plan is saved to the project (episodes already decided aren't overwritten). When you finish an episode, the “To next episode” card picks up the plan and starts the next one. You can still write directly without a plan. There's no screen for editing this plan on the planning screens.

Next: Production — attach assets to make images and videos