Want to generate images with AI on your iPad? In this tutorial, kwgeek shows you how to use apps like Draw Things or Guernika to work directly on your Apple tablet.
Whether it’s on the App Store, Instagram, or Reddit, it’s hard to go online these days without encountering images generated by AI. A few months ago, the only way to create one was through a web interface like the Dall-E Mini. Countless AI models are now freely available to generate images of all kinds.
Last fall, when things like StableDiffusion were all the rage, image generation was not at all user friendly. It requires an Automatic1111-style user interface and as powerful a graphics card as possible.
If you don’t have a suitable GPU, you can also run your Python-based scripts through Google Colab, a cloud-based framework with virtually unlimited computing power. Midjourney can also be paid, but the painting style is more particular.
How did StableDiffusion come to be on the iPad?
Over time, the computing power required to generate images through AI has diminished. Stunning images can now be produced entirely on the iPad, rather than using a high-performance computer or the cloud. We tested two different apps on the iPad Pro M2, each with their pros and cons: Guernika and Draw Things.
Both applications are just GUIs (Graphical User Interfaces) that use templates to generate images. They all immediately offer a list of image templates available for download. There are different versions of StableDiffusion and Midjourney’s open source alternative, Openjourney, everywhere. Draw Things, on the other hand, offers an extensive list, from Disney to Super Mario to cartoons.
Both Draw Things and Guernika allow you to use your own templates, which are always shared as .CKPT files. This acronym stands for “checkpoint” and designates the “reflective” state of a machine learning model, frozen in a file. For every model imaginable, there are countless resources on the web.
What are the important parameters to consider?
In addition to the models, you’ll find a range of settings that largely overlap in the two apps. Prompts, steps and “coaching scales” are of course especially important.
There are various adjustment possibilities and upgraders in order to achieve the desired result with Draw Things! / © kwgeek
Hints are textual instructions for the AI model to generate images. At first glance, this may seem simpler than it is, because to get good results with the style you want, you often have to use complex hints. Additionally, negative hints allow certain image styles, perspectives, or things to be excluded. But we’ll come back to that later.
These steps represent the number of operations required to generate the image. There will be steps, the better the quality but the longer the build time. Depending on the model used, different numbers of steps yield good results. The guidance scale determines how far the model can deviate from your cues. Low values allow the AI to have more “artificial degrees of freedom”.
For Draw Things and Guernika, you can also use other input images. For this, there is a “Strength” slider that determines how close the result is to the original image. Since each generated image is unique, Draw Things offers the possibility to generate up to 100 images at a time. Guernica even offers an infinite mode.
In Guernika, you’ll find most of the settings similar to Draw Things. However, it is not (yet) possible to adjust the resolution. / © kwgeek
Finally, only Draw Things allows you to set the output resolution. Guernika currently only outputs images of 512×512 pixels, although the developers have promised an update soon to allow more flexibility. Draw Things also allows you to add an AI converter to further increase the resolution. You’ll be able to create high-resolution images of 12 MP and beyond.
On our iPad Pro M2, image generation takes anywhere from 20 seconds to 5 minutes, depending on the settings chosen. It has been found convenient to examine the relevance of very low-resolution cues, and then, if the results are promising, to immediately generate a larger series of high-resolution images.
How to write a good prompt?
Writing good prompts is a science in itself that must first be learned and mastered. It’s not for nothing that we’ve already found our first job on a professional website as an “instant engineer”, the person who writes the instructions for the AI.
Your tips aren’t up to date? You can get inspiration from keyword combinations that other users have had good results with. On a site like PromptHero, you can click on an image and see which template or checkpoint and a prompt to create the image.
You can also ask ChatGPT to write detailed hints for you for some reason. I invite you to try it out.
What do you think of AI-generated art? Do you find the current progress with Midjourney, StableDiffusion, and others fascinating or frightening? Is this a topic you’d like to see more often on kwgeek?