AI Character Studio · Pipeline Guide

From a Folder of Art to a Finished Game Character

The complete flow for creating a production-ready character — new design, full turnaround sheets, and targeted fixes — running entirely on a local GPU. Every step below is a tab in the Studio, and every claim was earned by experiment.

TrainBuilderCharactersKrea EditorErase
01

Teach the game's language — Train

Point the Train tab at a folder of 15–40 existing characters from one game. A local VLM writes one caption per image using a closed trait vocabulary — the same words the Builder's dropdowns use (hero / villain, tiny / tall…), fixed before captioning, because sliders only work if the training words match them. One overnight run (~26s/step on an RTX 3060) produces a style LoRA that appears in the Builder automatically.

Proof the pipeline trains: same seed, without (left) and with (right) a home-trained LoRA — the character was never in the dataset; only the rendering shifts.
Proof the pipeline trains: same seed, without (left) and with (right) a home-trained LoRA — the character was never in the dataset; only the rendering shifts.
Rule learned: a closed loop validates consistency, not truth. Captions and declarations must be derived from the images — never written from memory.
02

Design a new character — Builder

Pick traits from structured axes (gender, age, build, alignment, mind, role), add a free-text line, and anchor the style: a single reference character at low reference-pull (copies the drawing language without cloning the character), the trained LoRA, or both. Generate 2–4 candidates and pick.

Style anchor on the left; a brand-new character on the right — tall · villain · clever · mage, plus “autumn-leaf armour, thorny wings”. Same world, new individual.
Style anchor on the left; a brand-new character on the right — tall · villain · clever · mage, plus “autumn-leaf armour, thorny wings”. Same world, new individual.
Rule learned: reference-pull is one dial with two jobs — high pins identity and pose, low frees rotation but loosens identity. Choose per task, not per habit.
03

Declare the character — Characters

The winner becomes a reference. Its declaration is a table, not a prompt: each feature says which views it is visible from, and — critically — what is there instead where it is hidden. The model cannot subtract; absence must be stated (“the backs of the trousers are plain — no stripe”). Style is declared explicitly too, or a glossy 3D reference comes back as flat 2D.

One reference in. The declaration is drafted from it, feature by feature — colours as seen, sides named, per-view visibility.
One reference in. The declaration is drafted from it, feature by feature — colours as seen, sides named, per-view visibility.
04

Generate the sheets — one view per run

Each view renders separately (~80s) and passes gates: geometry in plain code (shared height, ground line, stance), and content checks by a local VLM judge — calibrated on known answers before it earned a veto. Failed views re-roll automatically; the designer accepts or rejects; layouts are assembled by code.

The presentation sheet: front, three-quarter, side, back — one identity, colour-normalised at assembly.
The presentation sheet: front, three-quarter, side, back — one identity, colour-normalised at assembly.
The rigging sheet: a true T-pose, front and back — historically the hardest ask, solved with words by an edit model.
The rigging sheet: a true T-pose, front and back — historically the hardest ask, solved with words by an edit model.
The face close-up panel — full-resolution identity for modellers.
The face close-up panel — full-resolution identity for modellers.
Rule learned: the unit of generation is the view, not the sheet. Sheets are assembly queries — presentation, rigging, and detail layouts from the same accepted views.
05

Fix, don't re-roll — Krea Editor & Erase

Local defects get local fixes: instruction edit (change one thing, keep the rest), head swap between characters, outpaint to grow a canvas, and identity-aware Character fix — paint a region, describe what belongs there, the character stays intact.

A clean head swap: the second character's head on the first one's body, outfit and lighting preserved.
A clean head swap: the second character's head on the first one's body, outfit and lighting preserved.
Rule learned: the designer's note fixes the declaration, not the image — so every correction is capital that transfers to the next character.

What makes it hold together

LayerWho does itWhy
Geometry, colour, layoutplain codeexact and free — never ask a model to count pixels
Where is a featuregrounding VLM (boxes)only after its phrase is validated on the reference
Counts & verdictslocal VLM judgechosen by calibration on known answers (10/10)
Taste & truththe designerthe loop's corrections become permanent system rules

Hardware: one RTX 3060 12GB. Generation ~80s per view; LoRA training overnight via WSL2. No cloud, no per-image cost, no licence fees.