AI ART🎄 Technique for higher quality random images (Patreon)

When you're doing random images with Stable Diffusion, a quick way to get better results is by using the "Use as input" button that you see over the generated images. The idea is that you start with some smaller random shots until you find something you like in terms of composition. When you do, you pick that image and use it as an input and ask the AI to make larger images based on that.

STEP 1: GENERATE RANDOM IMAGES

You can start by doing a text prompt and generate multiple images that are 512x512px or smaller. If you go bigger than that with just a text prompt, the AI will end up doing a lot of weird things, you'll see extra legs, arms, etc. That's because the AI models learned from 512x512px images and when you go bigger than that the AI 'loses track' of what's going on around the image and might add extra arms or legs.

I generated these 6 images for example:

I liked the style of the 4th one and wanted more shots like that in better quality. This was the original 512x512px that was generated:

STEP 2: GENERATE HIGHER QUALITY IMAGES

So I:
- clicked on "Use as Input" over it
- changed the image size to be much bigger, I picked 896x896px
- set the Prompt Strength slider to 0.67. The bigger it is, the more freedom the AI has from the original image (more variation but also more chances for artefacts, extra limbs etc). You might need to adjust this depending on your image for better results

This is an example image it generated, with Prompt Strength at 0.67:

You can spot the increase in details more easily in the fur areas or the hair. You can hit Upscale and get even more details.

With this technique you can get higher quality random images without wasting a lot of time doing lots and lots of large images until you find a shot/composition that you like.