Imagedojo.ai – Blind arena for Google, OpenAI, and xAI image generators

Name: Imagedojo.ai – Blind arena for Google, OpenAI, and xAI image generators
Availability: InStock
Author: vtail

by vtail·Mar 1, 2026·1 point·5 comments

Visit Project View on HN

AI Analysis

●●SolidCrowd PleaserEye Candy

LMSYS Arena for images, but the leaderboard lacks volume—359 images doesn't drive statistical confidence.

Strengths

•Blind voting removes model bias and surfaces genuine preference signals
•Fair comparison: all four models priced $0.02–$0.06/image, so no tier gaming
•ELO rating system is mathematically sound for pairwise comparisons

Weaknesses

•Only 537 active matchups is low volume; confidence intervals on rankings are likely wide
•No breakdown by prompt category, style, or use case—limits insight into where each model excels

Post Description

Hi HN,

I was curious which of the three major US AI labs generates images that people like more, so I built ImageDojo.ai.

It shows you two images side-by-side, both generated from the exact same prompt. You vote on which one you like more (you don't see the prompt or which model made each one).

Based on the votes, it calculates ELO ratings for the models — similar to LMSYS Arena for text.

The four models I selected (the original and the new Nano Banana, GPT-Image-1.5, and Grok-Imagine-Image) are all in the same rough price range ($0.02–$0.06 per image), so we're comparing fairly similar-class models. Please try it out and let me know what you think!