Free online AI image generators
I'll review these six:
Each can be used to produce a limited number of images for free. Other online image generators include Midjourney, Imagine, DreamStudio, Night Cafe and dozens more.
Cost
Although each of these sites is free for limited usage, they all carry an environmental cost which each of us must evaluate for ourselves.
Monthly subscriptions are a more immediately tangible cost, and a link to the subscription page for each site is provided in the reviews below, when possible. My reviews are based on free accounts. Another cost consideration is that no special software or computer hardware is required, because the intensive computing is done in the “cloud."
Features
Some AI image sites let you upload your own photo as a starting point for the AI image, others don’t. Some let you choose a variety of image aspect ratios, other are limited to square images. I think of these as “features.” Here’s a list of AI image generator features:
Prompt fidelity: are the images generally faithful to the description you provide in the text prompt? Let's try "an octopus wearing a Hawaiian shirt"
Artist styles in prompts: can you influence the AI image by including the name of an artist? (For example, “a painting of a cat in the style of Van Gogh”)
Image prompts: can you upload your own photo along with your text prompt? I really enjoy "improving" my photos with AI. Check out these examples.
Aspect ratios: can you specify different image shapes and orientations?
Text: does the AI reliably produce text which you include in the prompt? Suppose we ask for “a wine bottle with the words 'Sip some LOREM IPSUM' on the label”
Upscale: most AI images are initially produced at a resolution of around 1024 X 1024. Does the AI include an option to increase the resolution?
There are plenty of other features which various AI generators promote. For example, some can generate a series of images with a character of consistent appearance, which would be useful in making a comic book. I don’t use these features, so I won’t comment on them in the reviews.
Reviews
Free usage: about 8 per day (one image per prompt)
Paid accounts: Subscription Plans | Deep Dream Generator
Prompt fidelity: poor
Aspect ratios: yes
Image prompts: yes, no strength options
Artist prompts: yes, many are built-in
Text: poor
Upscale: yes, fair quality
Comment: only one image per prompt, instead of the usual four. Often ignores most or part of the text prompt. Fun to use with image prompts, however.
Free usage: about 7 per day
Paid accounts: Leonardo.Ai subscriptions
Prompt fidelity: very good
Aspect ratios: yes
Image prompt: yes, with strength options
Artist prompts: yes, with limitations - fewer restrictions with a paid account
Text: very good
Upscale: yes, very good
Comment: my current favorite, strong performance all around, lots of fun to use.
Free usage: unlimited (?)
Paid accounts: very confusing. You use “Microsoft rewards,” I guess.
Prompt fidelity: very good
Aspect ratios: no, square only
Image prompts: no
Artist prompts: yes
Text: fair, seems to be getting better
Upscale: no
Comment: old in AI imaging years, usually makes good images, but lacks many of the features found at newer sites.
Free usage: 5 to 10 per week, with long waits for each image
Paid accounts: Ideogram Subscription Plans
Prompt fidelity: very good
Aspect ratios: yes
Image prompts: no (second-level paid accounts only)
Artist prompts: yes
Text: yes, very good
Upscale: no
Comment: Excellent performance, particularly with text, but their free accounts are highly restricted and annoying to use.
Free usage: "each user is limited to a certain number of daily generations"
Paid accounts: not offered
Prompt fidelity: excellent
Aspect ratios: yes
Image prompts: no
Artist prompts: yes*
Text: yes, very good
Upscale: no
Comment: One of the best free AI generators, along with leonardo.ai.
Free usage: 250 "Generative credits" per month
Paid accounts: Generative credits
Prompt fidelity: fair
Aspect ratios: yes
Image prompts: yes**
Artist prompts: absolutely not - Adobe rejects all proper names in prompts
Text: nope
Upscale: yes
Comment: Generally decent, middle of the pack performance and features.
Whisk
I must mention Google Lab’s Whisk, which is new, different and goofy. It doesn’t use text prompts at all; instead, you upload your photo and select from three options: Plushie, Sticker or Enamel Pin. It will then use your photo to very freely imagine what the subject would look like as one those three. Take a look at these examples. Then try it yourself at Whisk - labs.google/fx.
Conclusion
Leonardo.ai is my all-around winner because of its excellent performance, desirable features and generous free usage terms. Image.FX is equally good overall but doesn't offer an image prompt option. Bing deserves an honorable mention since it consistently produces good images, but it lacks features which most other AI image generators now offer.
Links to more examples
Text in AI images
Take a look at this (non-existent) Lansing Photo Meetup merchandise to see how AI images can incorporate text which is supplied in the prompt. The images were made in leonardo.ai and ideogram.ai, both of which do a good job with text.
Artist’s styles in prompts
See what Bing thinks a painting of a cat by several famous artists would look like. I also made a series of Christmas cards by famous artists in leonardo.ai.
These are a few of my favorite things made in Deep Dream Generator, leonardo.ai, bing, ideogram.ai, ImageFX and Adobe Firefly. In some cases, I've added text in Photoshop.
* "ImageFX features precautions to protect minors so certain queries that could lead to outputs of children will not be generated. Additionally, certain queries that could lead to outputs of prominent people will also not be generated." (I thank ImageFX for my all-time favorite prompt rejection)
** Adobe Firefly doesn't use image prompts as other AIs do. First you need to click Generate with nothing in the prompt line, in order to bring up a page with various options. You can then upload a “composition” image reference and/or a “style” image reference, and adjust the strength of each. These may be the same or different images. Takes some getting used to, but can produce interesting results.