MIRAGE (Multi-model Interface for Reviewing and Auditing Generative Text-to-Image AI) is a tool for comparing multiple text-to-image AI models side by side to find patterns and detect inconsistencies.

How many AI models can I compare at once?

You can select and compare up to 4 different Text-To-Image AI Models from the available list.

Who developed MIRAGE?

MIRAGE was developed by Matheus Kunzler Maldaner at Carnegie Mellon University, in collaboration with Wesley Hanwen Deng and professors Motahhare Eslami, Ken Holstein, and Jason Hong.

MIRAGE - Instructions

About MIRAGE

MIRAGE was created by Matheus Kunzler Maldaner during a summer REU at Carnegie Mellon University working with Wesley Hanwen Deng, as well as professors Motahhare Eslami, Ken Holstein, and Jason Hong.

Find patterns and detect inconsistencies in AI generated images from multiple text-to-image models.

Select Models

Select up to 4 different Text-To-Image AI Models from the list.

Write Your Prompt

Write your prompt, click 'Generate' and wait for the images to appear.

Observe The Results

Select which models matched your original expectations.

Begin

If you want to do research with MIRAGE, contact mkunzlermaldaner@ufl.edu