Prompt System

I created a modular prompt library tailored to different visual goals, such as tone, color palette, composition, subject, or era. Prompts are structured around visual anchors and stylistic parameters, and tracked in batches to test precision and repeatability.

Example prompt:
“Generate a minimalist interior with soft natural lighting, raw textures, and midcentury silhouettes. Use a muted palette, neutral shadows, and strong negative space.”

Model Testing

Image generations were evaluated across five categories:
Stylistic Consistency, Realism, Compositional Control, Prompt Responsiveness, and Editability

Each platform (Midjourney, DALL·E, SDXL, etc.) was tested for strengths, limitations, and use cases. This informed how I structured prompts, where I handed off assets for editing, and which models I used for what phase of the process.

Refinement Loop

Once generated, visuals were passed through a multi-step refinement phase including:

  1. Upscaling & cleanup (e.g. Gigapixel, RemBG, Photoshop AI)

  2. Inpainting or masking to fix details or swap elements

  3. Visual annotation to mark direction for reuse or iteration

  4. Figma-based mockups to place images in real use cases

The goal was to go beyond “decent generations” and build visual assets that were polished, purposeful, and ready to use.

Model Comparison

Insights from testing across leading AI image platforms

To understand the strengths, weaknesses, and ideal use cases of today’s top image generation tools, I tested each across a set of creative prompts and refinement goals. My evaluation was based on output quality, prompt responsiveness, consistency, and ease of iteration.

These are my personal findings based on hands-on testing and visual experiments throughout the workflow:

AI Image Generation + Refinement Workflow

Designing faster, smarter, and more intentional visual outputs

As generative image tools become part of early-stage design work, I wanted to explore what it takes to create usable, expressive visuals consistently. This project focuses on building a flexible system for prompt-based image generation and iterative refinement, helping turn raw AI output into directionally aligned creative assets. The workflow supports everything from moodboards to brand styling to speculative storytelling.

Role & Scope

Prompt development, model testing, visual tagging, refinement loop design, and workflow systems.

Tools: Midjourney v6, DALL·E 3, Stable Diffusion XL, Leonardo, Runway, RemBG, Photoshop AI, Notion, Figma

The Workflow

A seven-stage process for generating, refining, and curating AI-generated visuals.

INPUT, TRANSLATION, GENERATION, EVALUATION, REFINEMENT, RE-GENERATION, FINALIZATION

Prompt 1: Base Concept

“A minimalist reading nook with linen cushions and low shelves, soft natural light, neutral palette, architectural photography”

Goal: Establish composition, lighting, and materials
What to evaluate: Object placement, lighting realism, mood

Case Study: A reading nook in a minimalist, sunlit interior (using Midjourney v6)

Prompt 2: Refine Composition

“A minimalist reading nook with cushions on the floor, shot from a low angle, filtered sunlight casting soft shadows, clean lines, architectural magazine style”

Change: Refines camera angle and light behavior
Why: To test spatial depth and how the light interacts with surfaces

Prompt 3: Stylize Further

“A reading nook with natural textiles and pale wood tones, in the style of a Kinfolk interior, soft focus, high-res editorial lighting”

Change: Adds cultural and stylistic reference (Kinfolk aesthetic)
Why: To steer toward a softer, lifestyle-photography feel

Prompt 4: Detail Pass

“Close-up of a linen cushion on a pale wood bench, reading glasses and a ceramic mug nearby, soft morning light, warm tones, shallow depth of field”

Change: Focuses on a micro-interaction scene
Why: To build supporting assets or secondary shots for visual narrative

Prompt 5: Color & Contrast Shift

“Same reading nook in moody evening light, cooler tones, contrast between natural textures and shadow, cinematic interior”

Change: Time of day + emotional tone
Why: To explore tonal flexibility and contrast direction

Why This Matters

As generative tools grow more powerful, the challenge is both creativity AND control. This workflow is about designing better collaboration with AI by setting clearer parameters, tracking results, and refining with intention. Whether for brand development, visual storytelling, or concept design, this system helps produce images that support clarity and creativity, on top of volume.

[scroll down for full cheat sheet]