0-1
AI Agent
AIGC
Designing an AI Creative Agent for E-commerce Marketing Content
Overview
My Role
Product Designer
Duration
6 months
Ownership
AI generation workflow
Prompt optimization
Iterative editing experience
2 UX designer
1 Product Manager
3 Software Developers
Task accuracy
30%
less scrolling
88%
User satisfaction
01 / Problem
The "Prompt Anxiety" Problem
Many restaurant owners lack formal design training but still need marketing visuals.While they know what they want visually, expressing that intent through prompt syntax is difficult.This gap between visual intent and prompt language often leads to frustration, trial-and-error, and abandoned generations.
03 / Design Goal
How might we…
design AI image generation into a workflow that helps merchants express visual intent clearly, guide the generation process, and iteratively refine results.
Design Goals
Reduce prompt complexity
Enable iterative editing
Improve AI transparency
04/ Design Solution
Turning Prompt Writing into Visual Interaction
To reduce prompt complexity and make AI image generation more controllable for merchants, the interface restructures the generation process into three stages:
Old Version
Merchants struggled to interpret abstract style terms like Minimalist or Luxury by emoji, especially without design training.
Translate abstract style concepts into visual examples so users can recognize and select image intent quickly.
Visualize the abstract vision concept
We replaced abstract emoji labels with concrete visual examples generated, making each style clearly communicate the type of image it will produce. Users can now understand both the visual style and the expected output at a glance.
Streamline the Editing Workflow
To simplify the interface, format, style, and background selections were integrated into Edit Instruction, allowing users to edit images without navigating multiple controls.
2️⃣ Edit: Reference-Based Editing Loop
Enable reference-based editing so users can refine AI outputs using visual references instead of restarting the generation process.
3️⃣ Control: Guided Generation Controls for Stability
Introduce structured generation controls that guide users through adjustable parameters rather than open-ended prompts.
Model selection & Aspect ratio
for different generation capabilities
Reference Images
Replace abstract aesthetic prompts with visual style presets. Merchants select visual examples such as:
06 / Impact
Task accuracy
Increase in successful editing actions during image refinement.
Navigation path
Decrease in average steps required to reach a usable final visual
User satisfaction
Merchants reporting improved confidence when creating marketing visuals
Reflection
Designing AI means designing uncertainty
While AI models can generate impressive results, the experience can quickly break down if users feel they have no control over the outcome. As a designer, the challenge was not to expose more AI features, but to structure the interaction so users feel guided rather than overwhelmed.
When designing AI tools for non-technical users, the measure of success isn't sophistication—it's whether the user feels in control. Every abstraction must earn the user's trust through immediate, visible feedback.
AI could suggest styles or compositions based on product type and marketing context.
The system could better understand user intent across iterations and preserve key visual elements during refinement.
Over time the system could adapt to each merchant's brand style and visual preferences.