Skip to content

Repository files navigation

rpg-character-carousel — a personal brand built as an RPG character sheet

RPG Character Carousel

A LinkedIn document carousel that presents me as a classic 2D-RPG character sheet — a hero stats page, then one "item card" per hobby and tool, each graded by rarity like loot in an MMORPG. This build ships 8 slides, but the system scales to as many cards as you want: the hero locks the visual contract, and every extra item just reuses it.

It's a personal-branding piece, but the interesting part is underneath: it's a prompt-engineering project. The whole repo is a worked example of bending two generative-image models to a single, locked art direction and keeping them from drifting across an arbitrary number of separate renders.

Hero slide — character stats page


The carousel

DJ Controller — Legendary Coffee Dripper — Uncommon MTG Tome — Epic
Computer — Rare 3D Printer — Rare Controller — Epic
Watch — Epic

Each item is a real interest rendered as RPG loot: a DJ controller (LEGENDARY · WEAPON), a V60 coffee dripper (UNCOMMON · ARTEFACT), a Commander deck (EPIC · TOME), a 3D printer (RARE · CRAFTING STATION), and so on. Stats, passives, and flavor text carry the actual content.


How it was built

The carousel is the output of a two-stage generative pipeline. Each stage uses a different model for what it's best at, and the hard work is in the prompts, not the pixels.

Two-stage generative pipeline: CONTENT.md is the single source of truth for copy; Stage 1 generates pixel-art items in Gemini 3 Pro Image; Stage 2 composes finished slides in Claude Design; the result exports to a LinkedIn PDF carousel — with drift control as the connecting engineering.

Text version
CONTENT.md  ── single source of truth for copy (every word, proofread once)
     │           drift control: locked aesthetic lane · hard exclusions
     │           · reference images as locks
     ▼
Stage 1 · Item Art
  Gemini 3 Pro Image (Nano Banana Pro) · Google AI Studio
  → pixel-art items → assets/
     │
     ▼
Stage 2 · Slide Composition
  Claude Design · hero locks the visual contract
  · per-item template reused → slides/
     │
     ▼
LinkedIn Document
  export to PDF · post as a carousel

Stage 1 — Item art (Google Gemini)

Each pixel-art item is generated in Google AI Studio with Gemini 3 Pro Image (Nano Banana Pro). The prompts (GEMINI-PROMPTS.md, SLIDE-PROMPTS.md) lock a narrow aesthetic lane and feed two reference screenshots as "aesthetic locks" so every item belongs to the same world. Output lands in assets/.

Stage 2 — Slide composition (Claude Design)

Each item card is then composed into a finished, framed slide in Claude Design (claude.ai/design). CLAUDE-DESIGN-PROMPTS.md holds just two prompts: one long setup message that establishes the visual system and renders the hero, and a tiny per-slide template reused for every item card after it — run it as many times as you have items. Claude Design retains design context within a project, so the system is defined once and never restated. Finished slides land in slides/, then export to a PDF and post as a LinkedIn document.

The actual engineering: drift control

Generative image models love to wander. The bulk of this repo is the machinery that stops them:

  • A locked aesthetic lane — era, rendering language, palette, typography, and stat-bar style are all pinned, with reference images labelled as locks rather than suggestions.
  • Hard exclusions — earlier rounds returned an illuminated-manuscript look, a neon arcade look, and a modern card-UI look. Each failure mode is written back into the prompt as an explicit "do not return this" so it can't recur.
  • A single source of truth for copy — every word that appears on every slide lives in CONTENT.md, proofread once, because fixing a typo inside an image generator means re-rendering and burning quota. (CONTENT.md documents the full 10-card system; this build ships 8.)
  • Quota-aware sequencing — get the hero exactly right first, since the locked hero becomes the visual contract every later card inherits.

That's the transferable skill here: treating a creative generation task like a spec problem — constraints, anti-examples, a single source of truth, and a cheap-to-iterate path.


Repo structure

rpg-character-carousel/
├── CONTENT.md                 # Single source of truth — every word on every slide
├── GEMINI-PROMPTS.md          # Stage 1: item-art generation prompts (Gemini)
├── SLIDE-PROMPTS.md           # Stage 2: per-slide composition prompts (adding text to template)
├── CLAUDE-DESIGN-PROMPTS.md   # Stage 2: slide composition prompts (Claude Design)
├── assets/                    # Generated pixel-art items + the hero avatar
├── slides/                    # Finished, composed carousel slides
└── reference/                 # Aesthetic-lock notes (third-party art not redistributed)

Stack

  • Google Gemini 3 Pro Image (Nano Banana Pro) — pixel-art item generation
  • Claude Design — slide layout and composition
  • Exported as a LinkedIn document (PDF) carousel

A personal project by Oguz Oral. The slide content and artwork are mine; please don't reuse them as your own, but feel free to borrow the pipeline and drift-control approach.

About

A LinkedIn 'character sheet' carousel built with a two-stage generative-AI pipeline — Gemini 3 Pro Image for pixel-art items, Claude Design for slide composition.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors