Image Quality

#AI Image generation is not perfect

Let me briefly talk about the much discussed image quality issues for AI generated art.As with any emerging technology, AI image generation does not always deliver precise results. Quality of images rendered is a function of the AI imaging tool or algorithm,and the prompt given, which include the style, keywords, and the quality of rendering defined. The previous chapter provides an overview of different algorithms used to render or generate AI images. The user inputs (prompts) are covered in Chapter Three.

#Look Ma! I married a dog

When it comes to image quality,in particular, limbs, face and fingers of humans are often disfigured or missing. The poor rendering of these body parts has been a sore point for the early version of many AI imaging tools. Things are getting better, but generally speaking, we are not quite there yet. There are workarounds to help overcome this problem, atleast partially, but that is beyond the scope of this book. 

Some minor imperfections can be fixed using photo editing programs. The shoulder strap of the woman in the red dress for example, or the cowboy with an extra hand. In other cases the image style can be modified, say from photorealistic to comic art. At other times can be hilarious. See the example of a man, woman and a black dog below, or the levitating woman. Many times the images might not be usable at all.

The woman with three hands and one leg is one such example.

This image comes with 4 limbs. (Original Prompt: Woman wearing a blue dress and dancing on a stage)

(original prompt: Woman sitting on a chair and reading from ipad)

This shoulder strap does not exist

Talk to the Hand? Okay.. but which hand of the Witch? 

The cowboy who comes with a spare hand

Parting Words: As the algorithms get more robust, and millions of images are generated every month using AI, I believe the image quality issues will slowly weed out for most part. The use of negative keywords will also help alleviate some of these issues.