AI image generator shoot-out: I tested ChatGPT vs Gemini vs Meta AI to crown a winner

[admin]

I performed a side-by-side comparison to evaluate the image generation capabilities of three leading AI models.

AI image generator shoot-out: I tested ChatGPT vs Gemini vs Meta AI to crown a winner : Read more

 

[cam.macleod]

The funny thing is this totally personal preference, because I honestly thought Gemini easily pulled first in every scenario, except the interior design. I was honestly very impressed by Meta though. ChatGPT's always feels extremely artificial somehow.

 

[jansyren]

I feel that Gemini followed your prompt better while the others embellished the picture more. While they might given you more what you hoped for, I think it is better that it doesn't embellish and you have to refine the prompt to get what you envisioned.

Like the view over the ocean, that's very vague, change the prompt to be "a view overlooking a rocky beach with the ocean beyond fading into the horizon." This might have given you a better picture. Because I think it is better that the ai takes the prompt very literal.

 

[jansyren]

The funny thing is this totally personal preference, because I honestly thought Gemini easily pulled first in every scenario, except the interior design. I was honestly very impressed by Meta though. ChatGPT's always feels extremely artificial somehow.

I agree, gemeni followed the prompt closest without embellishing. I tried the following prompt on the living room: "Create an image of a minimalistic living room with a large window overlooking a rocky beach with the ocean beyond fading into the distance" and that's exactly what I got.

 

[AI_Nerd]

Imagen is one of my favorites for generating photos with minimal prompting. It's got good prompt adherence and is fairly effective at generating some good images. I thought they were the best overall. I don't know what Meta uses but it works pretty well.

Dall-E was impressive maybe a few years ago, but I hardly ever use it anymore. There are simply better generators for nearly every kind of image I would want to create.

For the most part I use SDXL and various fine tunes/variants for SDXL, like Dreamshaper XL lightning, when those fail I also use Imagen 3.0 and it's fast variant, or Flux and Flux Schnell.

Really I'm kinda disappointed an open source image generator like Stable Diffusion XL wasn't included. Especially when Dall-E was. I'm not sure which version is free, but honestly I don't get great results from 3, or 2, and I haven't used the first in a long time.

 

[AI_Nerd]

I agree, gemeni followed the prompt closest without embellishing. I tried the following prompt on the living room: "Create an image of a minimalistic living room with a large window overlooking a rocky beach with the ocean beyond fading into the distance" and that's exactly what I got.

Imagen (Alphabet/Googles image generator) is really very good for photos. It tends to have stronger prompt adherence than some other generators, so what you type is largely what you get.

 

[AI_Nerd]

The funny thing is this totally personal preference, because I honestly thought Gemini easily pulled first in every scenario, except the interior design. I was honestly very impressed by Meta though. ChatGPT's always feels extremely artificial somehow.

AGREE, I hardly ever use Dall-E anymore. It's very expensive and even for creating cartoon or "artificial images" there are simply better, and cheaper generators. I think OpenAI has easily one of the best LLMs out there, but they are not even worth mentioning with image Gen. I don't know much about Dall-E 3, whether it's a CNN, GAN, a VAE/transformer like Dall-E or a Diffusion Model like Dall-E 2, but I've never cared much for the images any of them have generated, other than when I was new to image generation and they were the first one I tried.