Back to projects
Prathamesh Patil

Project 03 AI Tool · Personal

AI Illustration Creator

AI image tools generate fast and forget faster - ask twice, get two different styles. For product teams that need an illustration system, that's a dealbreaker. This tool makes style a persistent, enforceable thing instead of a lucky prompt.

Type
AI generation tool
Role
Everything - concept, design, build
Core idea
Style as a locked, reusable asset
Born from
A problem I hit in my own work

At a glance

What it is
An AI tool that generates illustration systems - not one-off images - by locking a style and enforcing it across every asset.
The problem
AI image generation drifts off-style between prompts, which makes it unusable for products that need a consistent illustration language.
My role
Everything - concept, UX, visual design, and the build itself.
Key idea
Style as a first-class, reusable asset: defined once with references and guided questions, tested in a playground, then locked.
What it shows
Applying design-system discipline to generative AI - constraints as the feature, not the limitation.
Status
Working build, born from a problem in my own workflow.

How it works

Four steps, one constraint: once a preset is saved, nothing downstream can drift from it - even the chat that keeps refining the output.

01 · Create collection Start a new illustration set 02 · Upload Brand assets and reference images 03 · Style presets References, AI prompt, and manual tuning 04 · Generate AI prompt follows the saved preset Saved AI chat iterates inside the preset style never drifts

The product, in three screens

The core journey - the three moments that make style consistency enforceable instead of aspirational.

Define Illustration Style screen: style name and brand fields, reference illustration uploads, a mood board, guided questions on tone, palette and composition, and a live-building AI style summary with a 92% consistency score
Fig. 01 - Upload brand assets and references into a collection (1); the AI turns them into a style summary and a consistency score in real time (2)
Style Library screen: a grid of named, saved illustration presets across brands and teams, each with a consistency score, collection count, and usage stats, plus AI recommendations for new styles
Fig. 02 - Once saved, a style preset becomes a shared library asset (1) - reusable across a whole team, not just one project (2)
Generate screen: a prompt field with suggested prompts and prompt chips, a saved-preset panel reading 'Cannot be edited', and a grid of eight generated illustrations that all share one consistent style
Fig. 03 - Generation reads the saved preset (1) and applies it to every output automatically (2) - no re-briefing, no drift

What makes it different

  • Collections keep work organized. Every project starts as its own collection - error states, empty states, loading, travel, dashboard - so brand assets and generated output never bleed across contexts.
  • Style presets are first-class objects. Build one from reference uploads, an AI prompt, and manual tuning, then save it - the result is a reusable preset, not a throwaway prompt.
  • Presets, once saved, can't drift. Generation always reads the selected preset; there's no way to accidentally generate off-style.
  • Iteration without drift. An AI chat refines any illustration, but output stays constrained to the saved preset - a user can push the details, never break the system.

Why it exists

The same reason as the Figma plugins: consistency shouldn't depend on vigilance. If a system can enforce the standard, the person is free to do the part machines can't - deciding what's worth illustrating in the first place.