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

Show HN: Imagedojo.ai – Blind arena for Google, OpenAI, and xAI image generators(imagedojo.ai)

1 points by vtail 78 days ago | 5 comments

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!

vunderba 78 days ago |

For reference, have you seen the Artificial Analysis Image Arena Leaderboard? They also show you two images from anonymized models (shows after you vote), and calculates crowdsourced ELO ratings.

https://huggingface.co/spaces/ArtificialAnalysis/Text-to-Ima...

vtail 78 days ago | |

Thanks - and no, I haven't seen this one. I like how they have the edit mode dashboard - show the original image + two edits; I was thinking about doing something like this.

I'm also a bit surprised they have gpt-image-1.5 so high above Nano Banana 2 - my limited testing shows that, at least for the visual styles, people like Nano Banana more.

vunderba 78 days ago | | |

Yeah I think that it's part of the issue with a single "squashed" comparative metric. Some users are going to grade higher based on the overall visual fidelity and others are going to value following the prompt.

For a point of reference, I run a pretty comprehensive image model comparison site heavily weighted in favor of prompt adherence.

https://genai-showdown.specr.net

EDIT: FWIW, I agree with your assessment. OpenAI's models have always been very strong in prompt adherence but visually weak (gpt-image-1 had the famous "piss filter" until they finally pushed out gpt-image-1.5)