Tabletop players and indie designers usually need the same person across a dozen poses: traveler cloak, winter armor, injured version, portrait token. Random text-to-image rolls produce twelve cousins, not one hero. This workflow locks identity early, spends credits on controlled edits, and keeps published game IP out of your prompts. Use it with prompt craft, image-to-image, and commercial copyright notes.
That phrase lets the model pick hair, age, armor era, and camera distance independently every run. Tabletop art needs invariants: age band, silhouette, hair length, a signature prop, and a medium (painterly illustration, ink wash, soft cel). Write those invariants once on paper before you open the homepage.
Aspect ratio is part of the character sheet. Use 3:4 or 2:3 for full-body cards, 1:1 for tokens, 16:9 only for party wide shots. Generating a 16:9 hero and cropping to a token later cuts off weapons and boots you paid for.
Prompt skeleton: “[age] [gender presentation] [species if not human], [hair], [face marks], wearing simple [base outfit], holding [signature prop], standing three-quarter view, plain studio backdrop, [medium], soft side light, no armor yet.” The plain backdrop matters. Busy taverns bake into the identity and fight every later outfit change.
Run quantity 1 on one model. Judge only face, hair, and prop. If hands are wrong but the face is right, keep it—hands are cheaper to fix later than identity. If three attempts miss the face, rewrite invariants; do not hop models yet. Failed and policy-blocked jobs do not take credits; pretty failures do.
Download the Pass A winner. Optionally crop a clean bust. That file is now your reference, not a mood. Switch to image-to-image. Prompt only the change you want: “keep face, hair, and prop identical; replace outfit with travel cloak and leather boots; same plain backdrop.” Restating the whole biography invites a new person.
Strength / creativity settings (when exposed) should stay conservative for identity locks. If the face drifts, lower the edit strength or shorten the prompt to wardrobe nouns only.
Build a short variant list before spending: travel, ceremonial, injured (bandage, torn sleeve—not gore), winter cloak, hooded stealth. One variant per job, quantity 1. Name files clearly (hero_travel.png, hero_winter.png). Do not ask for “six outfits in one grid”—grids waste composition and often merge faces.
For combat poses, keep the same reference and change camera verbs: “same character, lunging forward, prop raised, still plain backdrop.” Add environment only after the pose works: “place on stone bridge at dusk, keep character identical.”
That is roughly under a dozen successful jobs if the brief is stable. Raising quantity to 4 on an unfinished Pass A multiplies cost without teaching you anything. Details on meters: models and credits. Starter balances are small on purpose; treat them as a lab budget, not unlimited free art.
Say “cel shaded illustration with soft outlines” or “oil-painted character portrait with visible brush texture” instead of dumping “anime, realistic, cinematic, masterpiece.” Mixed medium words fight each other and scramble silhouette. For a party that must match one art bible, lock the medium clause and never change it between variants. See also anime vs realistic.
Private home games: avoid pasting exact names of copyrighted heroes into prompts; describe class fantasy instead (“scarred human ranger with green hood”). Published Kickstarters and store modules: same rule, stricter—do not ship lookalikes of famous franchise characters. Nanomelon policy filters catch some brand prompts; near-misses that still look like known IP remain your legal risk under the Terms.
Leave Public Images off until you are happy for strangers to read the prompt. Staff approval applies to gallery cards; it is not a copyright clearance. Read Community Guidelines.
Hypothesis: identity locks better when Pass A uses a plain backdrop and Pass B changes only wardrobe. Method on Nanomelon: three Pass A tries with a busy tavern vs three with a plain studio; then two img2img cloak swaps each. Expected pattern: tavern winners drift more when the cloak prompt restates “tavern hero.” Plain-backdrop winners keep hair and prop more often. Record which prompt won in a text note next to the file—future you will not remember the lucky clause.
We do not publish fake precision percentages here. Your account, model list, and day-to-day capacity vary. The value is the method: one variable per run, quantity 1, write down the winner.
Generate at the ratio you will print. Upscale packages, when offered, cost extra credits or a plan—check pricing before promising a client a poster-size sheet. For VTT tokens, a sharp 1:1 bust from img2img beats a tiny crop of a muddy full-body. For printable standees, keep feet inside the frame; models love to crop boots.
On the homepage, lock Pass A at quantity 1, then switch to image-to-image for outfits.