There are lots of different AI models that offer the option to generate an image for you (Dall-E, Midjourney or Stable Diffusion, to name a couple). Dall-E creates fairly simple images - we see better results with Stable Diffusion and Midjourney. Midjourney is quite complex to use, so for this article we'll use Stable Diffusion as an example.
A simple way to use Stable Diffusion without creating an account is available here: https://replicate.com/stability-ai/sdxl.
Follow the steps below to get your perfect artwork:
- You can start by filling in a prompt. This is where you'll give a description of what you want to have displayed. It's recommended to be super specific with this, so to clarify if you want it to be a photo or a drawing, what sort of colours should be used and what kind of 'situation' you want to see.
- It's important to update the resolution of the image. The minimum aspect ratio for artwork is 1400 by 1400 pixels. However, these models aren't trained to generate images of this size. So if you fill in 1400 x 1400 as the aspect ratio, the results won't be too great. If you fill in 1024 x 1024, you'll most likely get better results and you can always upscale your image after downloading it (more on that here).
- Keep in mind that these models can sometimes generate quite odd things, if you ask them to generate specific text or humans (you'll get hands with too many fingers or weird characters that are meant to be letters). So you might get some odd results, but who knows, it might turn out pretty cool.
- Text can become an issue for your generated image. To pass the release checks that certain stores have (Apple for instance), the text on the artwork needs to be aligned with the release metadata. See more information on what makes good artwork in our separate guide here. So it would be our advice to add the text of your release artwork after the image has been generated. You can use Photoshop, Paint, Canva or the free online option Photopea.com to make changes to your generated image, and to add your artist name or title.
- You can also give a negative prompt, to specify what you DON'T want to see in your generated image. So you could for example input "text, letters, captions, names, titles, stickers, labels, words, logos, barcodes or watermark", so the model will try it's best to exclude this (it might fail to do so).
- If you're happy with your settings, you can simply press 'Run'. You might see some cryptic text first, but after about 30 seconds you will get to see your generated image.
- You can use the same settings to generate another image, if you're not happy with the first option that is shown. Or you can play around with the settings to get to your preferred image. Then, the only thing left to do is download the image, and upload it in Make Waves for your release.
Alternatively, you can use Stable Diffusion by creating an account on dreamstudio.ai, but this system is quite buggy in our own experience. Most of the steps described above are the same.
If the image that you've downloaded is smaller than 1400 x 1400 pixels, you can use the same website to upscale it: https://replicate.com/mv-lab/swin2sr. You can simply upload the image and enlarge it, without it losing its quality. Download the image and use this version for your release in Make Waves.
Questions? Email [email protected]