Weekend hack: ref + sketch + prompt → single image (built base in AI Studio, finished with Claude Code, ~5h)

Weekend Project Spotlight: Integrating Reference, Sketch, and Prompt into a Single AI-Generated Image

This weekend, I embarked on a quick development project using AI tools to streamline the process of creating composite images from multiple inputs. The result is a simple yet effective tool built with minimal time investment, showcasing how AI can facilitate multi-input artistic workflows.

Introducing PixMoe: A Compact Multi-Input AI Image Generator

PixMoe is a lightweight, free application designed to merge reference images, text prompts, and rough sketches into a cohesive AI-generated artwork. Its intuitive interface allows users to upload images for identity, outfit, or color palette, input descriptive prompts, and sketch a rough layout—all within a streamlined process. Here’s the link to check it out: https://pixmoe.com/

The Motivation and Development Approach

The core idea was to achieve a smoother, multi-input workflow that leverages different forms of guidance:
– References: To anchor the subject’s appearance and color schemes.
– Text prompts: To specify style, mood, or additional details.
– Rough sketches: To communicate composition and layout.

To facilitate rapid development, I utilized Google AI Studio for setting up the initial project, creating prompts, scaffolding, and endpoints. The heavy lifting of wiring the UI, handling API calls, and managing retries was completed with Claude Code, a versatile coding environment that accelerates AI integrations. The entire process from start to finish took approximately five hours.

Current Functionality

• Upload reference images to define identity, outfit, or color palette.
• Input descriptive prompts to guide the AI in style and mood.
• Draw a rough layout sketch to guide composition.
• Generate images, compare results, and re-roll for variations while maintaining subject consistency.
• Export high-quality images quickly and cleanly—completely free of charge.

Sample Output and Performance

Below is an example of the system in action:

Despite my limited drawing skills, the AI successfully interpreted my rough sketch combined with the reference image, accurately capturing the composition, subject, and overall vibe of the concept