avatarCeltic Chameleon

Summary

The article compares six AI image generators—Midjourney, DALL-E, Stable Diffusion, Fotor, Craiyon, and Dezgo—assessing their quality, cost, and flexibility in creating images based on text prompts.

Abstract

The author of the article has conducted a comprehensive comparison of six AI image generators by using the same prompts across all platforms. The generators are evaluated on their ability to produce engaging and useful images, with a particular focus on how they handle adult content and photorealism. The article includes visual examples and discusses the strengths and weaknesses of each tool, noting that while some like Midjourney and DALL-E have restrictions on adult content, others like Dezgo are more lenient. The author also highlights the use of Arc face restoration software to improve the quality of faces in generated images. A subjective ranking of the image generators is provided, with Midjourney being the preferred choice due to its image quality and cost-effectiveness.

Opinions

  • The author favors Midjourney for its high-quality images and reasonable cost, considering it the best overall.
  • DALL-E is criticized for its deteriorating image quality and strict content policies, which can lead to account suspension.
  • Stable Diffusion and Dezgo are noted for occasionally producing good free images, though they can have strange ideas about human anatomy.
  • Fotor is seen as a last resort due to its uninspired and sometimes unusable images.
  • Craiyon is heavily criticized, with its images being compared to those from the "seventh circle," indicating a very low opinion of its output quality.
  • The author appreciates the Arc face restoration software for its ability to enhance the facial features in AI-generated images.
  • There is a hint of disappointment with DALL-E's current performance compared to its past capabilities.
  • The author expresses a liking for some of the abstract and artistic images produced by Dezgo and Craiyon, despite their flaws.

Cage Match — Six AI Image Generators Go Head to Head

Who Will Triumph on Quality, Cost and Flexibility?

Created by author using midjourney, all rights reserved

Update — since time of writing Midjourney have revoked their free trial period.

Dazzling and engaging images attract attention — and if there’s one thing writers and content creators thrive on, it’s an audience.

There’s a huge buzz around AI image generators. For some of us they’re the perfect solution to the problem of copyrighted images — but they’re also great fun, a creative outlet and a means to find images that really work for us individually.

But how do you get started? Which generators create useful images, and how much do they cost?

Below, I’ve provided direct comparisons using precisely the same prompt each time but with six different image generators.

I think you’ll agree, the results are startlingly different.

And if you scroll to the end, I’ve provided a comparison table and links to each generator.

Midjourney, DALL-E, Stable Diffusion, Fotor, Craiyon and Dezgo Go Head to Head

I’ve been trying out six AI image generators, three are free, three give you some freebies.

This is a direct comparison.

Every image generator received exactly the same text.

Arc Face Restoration Software

I’d be remiss if I didn’t mention this free software.

Arc Face restoration is exceptionally helpful at tidying up peculiarities on faces produced by image generators.

Before Arc upload and after Arc upload.

Simply upload the image and download the results.

Four Prompts Using Six Image Generators — You Be The Judge

Prompt One: alluring woman high definition taking a bath, red lips, smiling welcomingly, full body angle.

Midjourney

The first is the original Midjourney image before I ran it through the Arc restoration app. On the right is the improved image.

created by author using midjourney all rights reserved

Like DALL-E, Midjourney is not keen on “adult” prompts.

And if you use prompts it does not agree with — such as bikini — you can receive a warning . This can make it a bit awkward when you need a more adult image. For example, on one occasion I wanted an elephant in a bikini. But even elephants can’t wear bikinis in the world of Midjourney.

The other images Midjourney created on this occasion were somewhat peculiar

DALL-E

DALL-E wouldn’t have a bar of the prompt, so I can offer no comparison.

I lost my first account on DALL-E because I was trying to create a picture of a sparkling black bra with some cleavage in it.

And they do not respond to emails.

Perhaps the coders are nuns.

Screenshot under fair use

Stable Diffusion

Alluring apparently means flashing your breasts.

This looks like something an alien might draw if it had never seen a human and a child tried to describe one.

Fotor

Fotor was also fairly enthusiastic in its approach to toplessness — and that sinister extra hand is an interesting touch. The image might have been usable if not for the slithering arm and the fact that her face looks like a melted mannequin.

And wouldn’t you love to know what the Not approved image was? I sure would. It seems odd that it’s possible to create an image that your own site does not approve — but who am I to judge the coders?

Craiyon

Coded by Yog Sothoth, Craiyon generously offers us 9 images designed by the Lurker at the Threshold.

If using Craiyon, when you click on an individual image to download, ensure you click the X on the top right to leave and return to the image array.

In this case, losing the image array was a merciful release.

Dezgo

Accepts adult/mature prompts without hesitation, but has some strange ideas about human anatomy.

I was able to clean up a couple of the images using Arc face restoration software — originals are on the left. but both images are still decidedly odd.

Prompt Two: view from behind, three little scottish girls walking to school in their school uniforms, one girl with red ringlets, one with bright blonde hair, photorealistic, photographic

Midjourney

All four have some merit. I chose top right initially, but have now opted for top left after noticing the girl in the middle top right has an odd growth on her neck.

The bottom left and bottom right hair styles have a clonish feel about them. And I found the weird not quite a plait on the girl in the bottom right really unpleasant for some reason. But it nailed Alison Smith’s blonde hair in that bottom right picture.

DALL-E

Almost but not quite.

One of the girls in the first image has had some very distressing surgery, or was born with an unfortunately shaped head.

The next image features Midwich Cuckoos.

All look a little odd, shoes don’t quite fit, there are weird little glitches everywhere — what’s that pink blob on the cylindrically shaped girl in the bottom right image?

Some of their legs are a little peculiar, and all have hair designed by Minecraft.

I wouldn’t be happy with any of these images on my articles.

Stable Diffusion

These aren’t bad. In images one, two and three the hands are a bit off, and some of the details aren’t quite right. Only one actually has three girls.

But still, for a free image generator, they’re reasonable.

Fotor

These girls look like they’ve been photoshopped or photographed in front of a green screen. I rather like the sky and the oppressive lighting in the first image, though one girl is more or less missing a leg or has her hips on sideways.

But I’m not sure what the background has to do with walking to school in Scotland. Perhaps Fotor thinks everyone in Scotland lives in a field.

The girls’ hair in the second image is fairly believable, but their hands are off. Very nearly usable.

Craiyon:

Oompa Loompa versions of Rod Stewart who’ve endured at least one trip through The Fly’s telepod.

I think not.

Dezgo:

These are pretty good for a free generator, particularly if you don’t look too closely at the arms or hands.

Prompt Three: divine madness a woman standing with raised arms on a dark road with the moonlight shining, hyperreal hypercolour, detailed, realistic, photorealistic

Midjourney:

I used the bottom left image. They’re all interesting, without extra limbs or strange protrusions. The last image has alarmingly long arms however, I can’t help thinking of Every Which Way But Loose.

DALL-E:

They’re not all terrible, but I’d be unhappy to have paid money for these.

The hands are weird, the arms are too long, the hair is blocky, and the face on the bottom right is blurry and strange. Kudos to the Queen tribute in the third image though.

Stable Diffusion

There’s a definite cartoon vibe to several of these, and bottom left (third image) looks like she’s hailing a taxi, but top right (second image) is actually quite pleasing, till you look at the legs closely and realise human ankles don’t do that.

Fotor:

The first image is to say the least uninspired, and looks badly photoshopped, but is probably usable if you don’t mind looking like an amateur. The second image is ruined by the spindly little arms and legs — disturbingly reminiscent of a woman metamorphosing into an insect.

Craiyon:

John Carpenter presents Return of The Ring Into The Mouth of Madness.

I quite like some of these, actually. They might be fun if I’m writing a horror story.

Dezgo

These have a certain simplistic, somewhat artistic charm.

And Finally, Prompt Four: a photograph of bluebells on mars bright colours digital art art station

Midjourney

I love every one of these, even though three of them aren’t images of bluebells. They’re visually appealing, and remind me of Omni magazines of the 70s and 80s.

-lost

DALL-E

Disappointing. Fuzzy blubells in some sort of modern art style was not the prompt.

Stable Diffusion

Uninspired, but serviceable, meets the elements of the prompt.

Fotor:

Not sure where Mars went, but these are pretty enough, in a basic sort of way. I’d use them if I was talking about a walk through a bluebell wood, and didn’t have anything better.

Craiyon:

Interesting, not much of the “Mars” is evident, but there’s a touch of the abstract expressionism about these.

Dezgo:

I rather like these. The one on the right doesn’t say “Mars” but it’s oddly pleasing.

Comparisons

My favourites for image quality, ranked.

  1. Midjourney — the clear winner for me. Best images, won’t break the bank if you do decide to pay, allows you to trial for free.
  2. DALL-E — In my opinion, quality of images has deteriorated. It still produces some nice images and is generous with the free trial.
  3. Stable Diffusion/Dezgo — Both sometimes produce good free images.
  4. Fotor — If you’re really stuck you could try it.
  5. Craiyon — Images from the seventh circle.

Table of comparisons. The quality comparisons are subjective, obviously, and I encourage you to compare and contrast for yourself.

If the image is too small to read, click this link to the Google drive doc: https://docs.google.com/document/d/13TAKxJkwpDtujo6MACvsVMyhxeLwfFCMbME52Y68xfc/edit?usp=sharin

Below are links to all the AI Generator sites.

screenshot by author

Where To Find Each Image Generator

Midjourney

To access Midjourney will mean a few extra steps — but I think it’s worth it.

  1. First, set up a Discord account — I logged in through a a Google account for convenience and speed. You’ll have to confirm your email address.

2. Next open Midjourney — click Join the Beta.

You will be offered an invite — accept.

3. On the left of the screen you will see “newbie” rooms. Enter one.

4. You’ll be in a screen of constantly changing prompts and images. You can take inspiration from many of these prompts.

Next, type /imagine and this box will pop up:

5. Type in your prompt.

Please note, when you hit enter after typing in your prompt you will be instructed to accept TOS and then have to re enter the prompt.

If there is a way to accept the TOS first and avoid this rigmarole I haven’t figured it out.

Images usually take 30 seconds to one minute to appear in your gallery, and about 10–20 seconds to appear in the newbie room.

6. You can make your images in a private chat with the Midjourney Bot — people can still see them, but you no longer have to see a room full of constantly shifting images. Just click the Midjourney Bot and start a conversation. I usually just type “test”, and you will be taken to a private room.

7. You can also use images to assist you with making new images. For example, I used a photograph of my own face and the prompt “in a superwoman costume” to produce these:

To do this, you have to use a URL that ends in jpg, png or similar. So, right click on the image and click “open image in new tab”.

I found my picture online at my old Amazon author page — no longer linked to books because I unpublished them both.

8. Click on your name at the bottom left of the page to change your settings — you can make yourself invisible online if you wish.

9. To see your created images in your own gallery copy paste this link then go to gallery:

https://www.midjourney.com/app/

AI Images have taken off like wildfire and provide many creators with an imaginative avenue to improve their work.

Love them or hate them, they’re not going anywhere. You may as well enjoy the ride.

Links to the other image generators:

Midjourney
Ai Image Generator
Image Generation
Copyright
Zobisquatch
Recommended from ReadMedium
avatarDharshanya Venkataramanan
What is your favorite tool?

And how do you use it?

2 min read