There are many AI Text-to-Image Generators available, some more expensive than others. Many are new to the market, and they’re constantly learning and updating (Dall-E2 released a new beta feature update last week: Outpainting, while Midjourney just added the “Remaster” feature to their supped up chatbot.) As an artist, I can see the value each of them bring as individual tools, and that they have strengths and weaknesses that make some more suited than others to specific tasks.
NightCafe’s Stable Diffusion algorithm and OpenAI’s DALL-E2 are both currently in open Beta: pricing for them varies, with each company using different models for now. Expect for all of this data to be outdated soon, as the market learns what users are willing to pay. I’m anticipating DALL-E will have an unlimited pricing plan soon to compete with other programs, or they’ll be too expensive for viability once the initial buzz wears off.
Dall-E is currently charging $15USD for 115 credits (each credit buys the user 4 (four) renders of a prompt, averaging $0.033 cents per image.)
Midjourney has been getting a lot of attention lately, and for good reason. Don’t let their terrible webpage fool you: the eponymous research lab behind the Midjourney bot have created a really powerful, and so far relatively inexpensive, chat-bot-on-steroids.
Midjourney has the capacity to render and refine images at or above the level of the competitors from within a Discord server. Their pricing is planned to cap users’ time spent generating the images (which they say take about a minute) and their $10/month plan gets you 200 minutes, meaning about $0.05 per image.
NightCafe recently released their newest algorythm “Stable Diffusion” which is doing some impressive things. Their credits only buy a single image, purchased in packs or monthly subscriptions with pricing ranging from $0.20 — $0.053 per image. NightCafe does offer ways to earn free credits, including daily login bonuses. The other cool thing about their model is that you can scale how many renders, how much time, etc. in their application so that you only pay for the capacity that you want.
At under a penny each for most of the options, it may seem like a cheap trick, but it adds up. Using these programs can be addictive in a fun way, but it does take some learning about how to use them properly to get what you want.
When you’re paying to play, bad outcomes are not fun. Simply put, prompt engineering is when you instruct the app in a specific way that you know it understands, like you’re translating your command to the machine’s native language. For example, instead of saying you want a photo “old, brown” you would say “sepia.” For more about this, see my article about Prompt Engineering and Jobs of the Future.
I want to mention there are fully free generators that do pretty great with some prompt engineering, my personal favorite is Wombo Dream. While this app can be hit-or-miss, there is a button to roll it again and at the price of free, you have to love it. I highly recommend practicing your prompt engineering in a free app like this, and once you’re getting close to the results you want, spend a couple credits in a premium app to get fantastic outcomes.
TLDR:
If you want premium quality, nothing beats DALL-E 2. If you want high-quality with a reasonably priced subscription fee for a power user, get Midjourney. If you want to mess around for free, I’d use Wombo Dream. If you’re willing to do a daily login to get free premium access, check out StarryAI and NightCafe.
