We’ve been writing a lot lately about AI-based image generators. And, we are going to have two artists (one with a notable career in art, the other who is a developer that creates art but doesn’t think of himself as an artist) speaking this week at Synthedia, the synthetic media conference we are hosting.
The reasons for this are threefold: these technologies are maturing quickly, they have sparked strong consumer interest, and new innovations are arriving each week. While generating an interesting image is ridiculously easy, it is not always easy to get just the image you want. You are only in control of the words your enter into the prompt; the AI image generators take it from there. You cannot control how the solutions interpret your words.
Trial, Error, $$$
This situation leads to a lot of trial and error along with a new skill set of prompt writing. You might think that trial and error is no problem because you type and get an image in the response. That might be partially true if the solutions were free.
All of the solutions have some free credits to start, but you can burn through them very quickly, changing a word here, another there, upscaling the quality, and so forth. A recent analysis suggests that the cost after your free credits are used up is about $0.13 per Dall-E image and $0.05 per image for about 200 images per month with a $10 monthly subscription to Midjourney. Midjourney also has an unlimited account option for $30 per month. The founder of PromptBase says it costs about $10 in credits to come up with a high-quality prompt that works well repeatedly.
You would be surprised how quickly you can use up dozens of credits on a few images. Then there is the time involved. Frequent use will make you more efficient and produce fewer prompt errors, but there definitely is an art to the process. And who doesn’t want to save time?
AI Software to the Rescue
We live in an age of convenience. If you don’t know how to write the code for a common function, you can find many examples quickly and for free on Stack Overflow. If you want the process to be even easier by eliminating the search, copy, and paste steps, start using GitHub’s Copilot AI. It will suggest you add the code before you even search.
It’s not surprising that this convenience has come to prompt writing. PromptBase lets you purchase pre-written prompts for $1-$3. However, that means you are shopping from a catalog. If you have something truly novel or specific, you need to write the prompt. Phraser will help you do it.
Phraser’s first feature walks you through the prompt-writing process for several of the popular AI-image generators such as Midjourney, Dall-E, Stable Diffusion, and Craiyon. Whether or not you have some experience with prompt writing, Phraser will help you avoid forgetting a key element to include. That may save you a couple of errors in your trials. If you are new to the space, Phraser is also training you on prompt writing best practices in the process.
Phraser will also show you “similar images” of what you might expect using that prompt. If you like the direction, you can copy, paste, and generate the images. If not, you can refine your prompt.
Search for Images
A new Phraser feature allows you to follow a similar but simpler process to see previously generated images using a prompt. In this case, you can just type away without answering all of the questions required for a full prompt. Phraser will show you images with similar keywords generated by others, who owns the copyright, and on which platforms they originated.
This is another tool you can use to reduce your trial and error iterations. My expectation is this will save you time and money.
It is interesting that the “skill” and “IP” behind AI-generated images are embedded in the prompt. This sentiment was echoed by Jason Allen, the artist that won the Colorado State Fair art contest last month. “If there’s one thing you can take ownership of, it’s your prompt,” Allen said.
Great prompts are assets in this view. I could see a cottage industry of people using Phraser to generate high-quality images and then listing the prompts for sale on PromptBase. Maybe it will cost them less than the $10 average and save them a lot of time in the process.
Use of AI text generation recommendations for creating AI-generated images would seem to be inevitable, not quite dystopic, and beneficial. It’s the machines controlling the machines with a human intermediary. I’m sure a lot of people are concerned that machine-to-machine connections with human disintermediation are also inevitable. Either way, the endless march toward convenience, time savings, and interest in offloading expertise to systems suggests tools like Phraser will become more sophisticated and widely used.
Let me know what you think either by commenting here or on Twitter.