Printing images from Open AI
How often are everyday digital images actually printed? Very rarely. Prints at A2 size even more rarely.
Creating AI images to put on Instagram or Facebook can be quick and easy. Images for printing at A2 size are a different matter. Physically, a mobile phone image may be about 7.5 centimetres wide. An A2 print is up to about 40 centimetres wide and 57 centimetres high. Currently, it is not possible to get anything like a decent quality image simply from an AI-generated image. That is where the work begins.
STAGE 1: Obtain an AI image
The process starts with a 'prompt', asking the AI tool to create the image you described. To describe something, you must first have something specific in mind. The generated image may not be right first time, requiring an amended description and a new prompt. It can sometimes require a dozen or so requests to get the result you want. Sometimes you simply have to give up and try something else.
STAGE 2: Adobe Photoshop
Adobe Photoshop is then used to open the AI image and adjust colours, contrast and general tone – altering parts of the image that are too dark or too bright. Often some manual adjustments are made with digital brushes to correct errors, typically the wrong number of fingers, incomplete fingernails, malformed toes and so on. This process is similar to manual painting.
STAGE 3: Upscale
At this stage the images are typically 1024 pixels wide: nowhere big enough for a decent A2 print. They need to be 'upscaled' to around 4,600 pixels wide. This is done using software made for the purpose. I tend to use Topaz Gigapixel or Topaz Photo. Those programs offer choices of upscaling models to add detail, additional realism, sharpness, grain, and to reduce the 'digital artifacts' that can appear in an upscaled image and make it seem more textured than it should be. Experimentation and test prints are usually required to get things right.
STAGE 4: Back to Photoshop
Once upscaled, an image is then re-opened in Photoshop. It is now much larger than before, enabling further adjustments to be made more accurately. Often, it is back to fingers and toes. If they can't be corrected manually it may be necessary to go back to the beginning and ask Open AI to create the original image again. It is sometimes easier to place hands behind the figure's back to remove them from the picture, or to request shoes on the figure's feet to avoid excessive 'toe work'.
Several different versions of the same image, each adjusted slightly differently, can be stacked together on different layers, making layers partly transparent to combine the effects of the different versions.
A final adjustment is usually to manually remove the digital 'halo' – the unnatural bright or dark borders that surround part of an image where a dark element is next to a light element, or vice versa. This can take several hours for a single image.
STAGE 5: Print the image
The image is printed on high quality A2 professional print paper at high resolution, again involving test prints and making any necessary adjustments, usually to colour, contrast and overall tone.
STAGE 6 (possibly): Realise you aren't satisfied and start all over again.
