30 comparisons
Each set has three anonymous outputs shown in a newly randomized order.
Research study · approximately 20–30 minutes
Help us evaluate generated images.
You will compare three images created from the same text prompt. We are interested in your visual judgment—there are no right or wrong answers.
One simple score
Rate the high-resolution execution of each anonymous image from 0–10.
Your response
Your name identifies your submission, and a random study code keeps each response unique.
If you have any questions about the study, please contact kimias@cs.toronto.edu.
High-resolution image rating
Rate each image using any whole number from 0–10.
Hover over each image to magnify fine detail. Use Inspect at 100% to examine actual pixels before scoring.
Consider fine detail: does it resolve into real material, or go soft and smeared? Check structure for duplicated, malformed, or seamed objects. Does the detail hold up as you inspect it, or run out as if a smaller image were enlarged?
Ignore subject and style—judge execution, not taste.
The points below are reference guides—not the only choices. Use any score between them.
0 Unusable: broken or unresolved throughout5 Mixed: convincing areas, but clear failures elsewhere10 Exceptional: fully resolved throughout, with no visible structural defects
Full generation prompt