Mini Me Storyboard: build a character world from your photos

I started with a photo and a simple idea: what if smaller versions of me were living alongside the tools and people I work with? Twenty finished images later, the interesting part was no longer the miniature. It was the relationship between a real person, a visual cast, and a story that could teach something.

Here is the workflow we used, the corrections that mattered, and six prompts you can copy. For the broader image-production process, see from reference to campaign.

The full prompt vault contains the exact twenty generation prompts from the saved collections. Reference numbers and local paths belong to this production run; replace them with your own approved photos and character sheet.

For a repeatable assistant workflow, copy the portable Mini Me Storyboard skill with its prompt templates and review rubric.

The result

The first collection explored ten 4:5 editorial posts: music, open source, coding, research, automation, and three infographics. A second collection made ten vertical scenes, from a rainy archive and a vinyl-record world to a three-panel bug comic and an ensemble finale. These were generated images with captions, alt text, and provenance notes. They were not a controlled comparison of image models or a demonstration that the illustrated agents actually ran together.

Enlarge image

Echo is an original FrankX companion for the music story. It has no official connection to Suno.

Enlarge image

The process that made it work

My own likeness needed repeated correction. The hair is swept and a little wavy, sometimes with a loose fringe. The beard is lighter on the sides than at the moustache and chin. The head should not become an oversized toy. A black shirt with black buttons or a black suit can carry the elegant-leader frame; dark sunglasses are the only eyewear option when glasses appear. A good character sheet makes these constraints reusable across scenes.

Enlarge image

Six prompts to copy

Replace the square-bracket fields and attach only the references that serve the scene. A source photo is the identity authority; a prior illustration can be the style authority.

1. Audit the reference photos

I am making a recurring illustrated version of a real person. Examine the attached approved source photos as identity references. Describe stable traits, distinguishing details, hair silhouette, eye shape, beard density and side coverage, and clothing. Separate identity from artistic style. Make a compact character bible with a default outfit, two alternate outfits, three expressions, and explicit things not to change. Do not infer unsupported biographical details. Keep other people in the photos out of the reusable identity model.

2. Establish the character

Create one character sheet for [PERSON] using the approved identity photos and character bible. Show front, three-quarter, and side views plus two full-body actions. Refined stylized 3D: recognizable human proportions, expressive face, tactile materials, cinematic softness, no plastic toy sheen. Preserve [HAIR], [FACIAL HAIR], [EYES], [DISTINCTIVE FEATURES]. Show [DEFAULT OUTFIT] and [ALTERNATE OUTFIT]. Neutral background, coherent scale, no invented logos or extra characters. Use this sheet for continuity in later scenes.

3. Plan ten distinct stories

Plan ten distinct vertical posts for [AUDIENCE] about [THEME]. For each write: title of six words or fewer, one-sentence insight, human emotion, everyday setting, miniature physical action, supporting character and verified role, camera and light, aspect ratio, one proof or takeaway, and accessible caption. Use at least six different visual grammars, such as cinematic portrait, comic, still life, architectural cutaway, diagram, and environmental scene. Build toward a human invitation. Flag every brand-owned character and original invention. Do not generate images yet.

4. Generate one scene

Create ONE finished [4:5 or 9:16] vertical scene, not a contact sheet. The approved character sheet controls identity; the environment photo controls composition. When editing, keep [EXACT PHOTO REGIONS] unchanged and add the miniature as a separate plausible layer. Mini [PERSON] [ACTION] on [SPECIFIC SURFACE], while [SECOND CHARACTER] [COUNTER-ACTION]. The turning point is [SETUP -> CHANGE]. Match light direction, perspective, scale, cast and contact shadows, reflected color, and depth of field. One clean focal point, phone-safe space for a short headline, no invented paragraphs or interface.

5. Make an exact infographic

Create the illustration and character interactions for a [9:16] infographic about [ONE PROCESS]. Reserve clear areas for one heading and three short labels; do not render factual labels into the image. Show each stage through a physical action. Separately provide exact copy, reading order, alt text, and source notes. Set final typography and arrows in SVG, HTML/CSS, Figma, or another precise layout tool. Test the final at phone size and keep critical text in the central safe area.

Enlarge image

6. Review and revise

Review the final at phone size. Score 1-5 for likeness, silhouette, readable action, physical contact and shadow, compositional variety, factual accuracy, brand provenance, exact text, crop safety, and emotional clarity. Name the top three defects and their precise regions. Revise only those defects, preserving approved identity and successful composition. Export with prompt, references, version, caption, alt text, provenance note and change log. Do not claim a character is official without a current first-party source.

Label the character's origin

The project included interpretations of recognizable public characters such as GitHub Mona and Copilot, the Codex pet, Claude Code Clawd, and OpenClaw's lobster. Hermes was treated as a brand-avatar interpretation; the DeepSeek whale as a logo-inspired illustration. Echo for Suno and Kite for Kimi were invented for the FrankX story. Neither is an official mascot. Check current first-party rules and request any required brand approval before using a company's character in promotional work. A miniature agent in a picture does not establish an actual integration or a capability.

Keep the evidence with the image

For each scene, save the exact generation prompt, identity/style references, output, revision note, crop, caption, alt text and provenance. Our two saved collections preserve those pairings for twenty finished images. This article reconstructs the decision process from the saved artifacts and the creative direction; it is not a screenshot of the complete ChatGPT conversation. If you document your own run, capture screen recordings or screenshots during creation, with private photos and chat details removed before sharing.

The essential pattern is simple: preserve the human, give the tiny character an action, and use the scene to carry a real idea. The rest is art direction and review.