Character design lives inside Character & Scene, an Image-module feature you reach from the workbench, not as a standalone product. The feature has three separate entries, and this page covers the character one.
One Feature, Three Entries
The entry page always shows the current Project and offers three explicit doors: Create Character, Create Scene, and Multi-Image Compose. You pick the flow; the interface does not guess one from a mixed input box. An empty Project shows all three with short purpose notes, plus a link to Character & Scene Explore where you can browse and Remix other people's published characters, scenes, and composes. You do not need a character before you start: uploading a reference image and going straight to a compose is a valid first move.
Designing a Character
Pick Create Character and describe the character in natural language, optionally with reference images from your device or the account media library. Workflow Copilot can join the conversation to complete or refine the description, suggesting around persona and temperament, apparent age, body type, face and features, hair, palette, materials, art style, clothing, accessories, props, identity anchors, and forbidden deviations. Its suggestions stay editable and none of it becomes a required form.
The flow deliberately steers you toward one image first: a character sheet, a close-up portrait plus front, side, and back views. It is the recommended core material for everything that follows, and it is a recommendation, not a gate: nothing stops you from generating an expression or an outfit before the sheet exists.
Every generation shows its fixed price, an image is 1 credit, before anything executes, and one character context runs one paid task at a time; repeat clicks and refreshes cannot create duplicate charges.
How Character Work Is Organized
A character is a named collection of images organized into six categories: core, expression, action, outfit, accessory, and prop. The categories exist for browsing and choosing, not as deep object trees: no folders, no series, no user-visible batches. The workspace is a single main canvas with a collapsible asset tray showing the current category's thumbnails, or all images grouped by category when expanded.
Every image carries a user-visible name; defaults describe the content ("rain coat variant"), never bare numbers. Each collection and its images take up to 10 private tags for filtering, and the copilot can propose collection names, image names, and tags for you to accept or change.
Changing canvas ratio, size, or count is input for the next generation. It never crops or retro-edits an existing image; new settings produce new results and the tray keeps everything.
Combining Characters and Scenes
Characters and scenes are two separate collections with separate taxonomies (scenes organize by core, perspective, atmosphere, and element). Putting them together is the job of the third entry, Multi-Image Compose: a named task that combines references, including any named character or scene image from your Projects, and appends a new named result version per successful generation. The compose task keeps its own cover and result list, and what it produces is an ordinary generated image for you to judge, no special guarantee that a costume or prop survives the merge pixel-perfect.