The problem

Image generators often assume hidden expertise

Many image-generation interfaces expect people to already understand prompting, composition, lighting, style vocabulary, and model-specific constraints. That creates cognitive load and makes trial and error the default workflow.

Product goal

Turn prompt creation into a guided design process

The goal was to make visual intent easier to articulate through a structured sequence that teaches useful prompt patterns while producing something ready to reuse.

The solution

Six stages from subject to iteration

  • Subject and action: Define the focal content and what it is doing.
  • Setting and mood: Establish the environment and emotional tone.
  • Style and medium: Choose a visual language and add meaningful style details.
  • Composition and lighting: Describe framing, focus, depth, and illumination.
  • Quality and constraints: Set detail expectations and common problems to avoid.
  • Seed and iteration: Record stable references and what to change next.

Key features

Immediate feedback without model lock-in

  • A compiled prompt updates as source choices change.
  • Shortcuts cover blur, distortion, unwanted text, watermarks, anatomy errors, and visual clutter.
  • Seed references and iteration notes make experimentation easier to track.
  • The prompt can be copied into the image system appropriate to the project.

Accessibility

A workflow designed to reduce interaction and cognitive barriers

  • Semantic structure and visible labels identify every control.
  • All workflow actions support keyboard use and visible focus.
  • Status changes are announced without unexpectedly moving focus.
  • Readable typography, strong contrast, and named steps support scanning and orientation.

Implementation

A local, engine-independent prototype

The original prototype used vanilla HTML, Tailwind CSS, and JavaScript. The MyKMHub version modernizes the interaction with React and Spectrum 2 while keeping the prompt logic local and independent of a particular image-generation engine.

Impact

More intentional prompts and faster learning

  • People can build prompts without memorizing a specialized syntax.
  • The workflow exposes reusable visual-design patterns rather than hiding them.
  • Structured iteration reduces repeated guessing and makes results easier to compare.

Future concept

From prompt assembler to visual intent compiler

A future Visual Intent Compiler could identify conflicting choices, apply readability and composition rules, explain its recommendations, and translate stable intent into model-specific prompt formats.