Day 2: Creating “Art” with AI 

Welcome to Day 2! Today we’re going to explore using AI images and how we consider this artwork.  

Today’s post is authored by Mark Robinson, Learning Technologist at London College of Fashion, UAL. 

Artists and creatives are likely to be particularly interested in visual Generative AI tools, but it is important for all users to be curious and cautious. The rendered image created by AI is just the final stage of a process which involves data collection, processing and labelling, and incorporation into diffusion models. So, the AI outputs need to be viewed through a critical lens considering hidden processes we are not always aware of. 

The mathematician Marcus du Sautoy wrote about AI and creativity in The Creativity Code: Art and Innovation in the Age of AI (2019) and mentioned the ‘Lovelace Test’ named after the first computer programmer and possibly first AI sceptic, Ada Lovelace. Du Sautoy’s work looks for something ‘new, surprising and valuable’ when judging creativity. This might be a helpful perspective when considering GenAI. 

Artists Mat Dryhurst and Holly Herndon have created their own AI model available online and built from their own data set of images. As text to image prompts are entered, the outputs, which always resemble the artist, are then incorporated back into the training data. So, the use of the tool is instilled within the production system with control of online images determined by the artist’s use of AI rather than Big Tech.  

Images produced by AI tools could be characterised as ‘statistical averages’ having a similar quality to the stock image. Artist Hito Steyerl has described these as ‘mean’ images, referencing these methods of production as well as other connotations of the phrase ‘mean images’. 

Many of us have created images through prompting, producing a variety of outputs which frequently share a common aesthetic. Fully engaging with AI tools can require some technical knowledge, such as developing data sets and models, but we’d like to begin with an activity starting with an image before prompting. We are encouraging a curious and critical lens on GenAI tools. 

Activity 

Part 1 

Let’s take an optimistic approach and think about text to image generative AI as part of a process which we can use creatively and iteratively. 

  1. Make a quick sketch or take a photo of something in your immediate surroundings as a starting point, such as on your desk or nearby. For example, you could capture an image from outside your window or draw an item on your desk.   
  1. Head to Deep Dream Generator and select the Sign Up option. You may find this on the upper right-hand corner. You may need to verify your account access through an email confirmation. 
  1. Select the Generate option from the upper menu once logged in. Select the option Visual Prompt.  
  1. Upload your sketch or image as a starting image to Deep Dream Generator. Apply a narrative text -to-image prompt on the file to render an image. This might be a prompt such as ‘a painting in pastel colours, pointillist style, joyful’. 
  1. Try a few variations of the process to see if the images are interesting, novel or productive. Might they pass Du Sautoy’s ‘Lovelace Test’? 

Alternative Tools 

You might want to try another alternative, though some require paid subscriptions. 

  • Adobe Firefly (free accounts can be created and images uploaded within text-to-image prompting but you may have access at UAL or through your institution.) 
  • Midjourney (paid subscription required) 

Here are our examples: 

First, a sketch rendered by Deep Dream Generator.  

The sketch was uploaded as a visual prompt and outputs from 2 different image models (daVinci 2 and Artistic2 ) created with the prompt: pastel coloured, abstracted painting a sense of melancholy. 

Here’s a photo of a staircase in a UAL building. Although the GenAI created the image of the staircase, the tool also added a figure for reasons only known to itself, perhaps misreading or ‘hallucinating’. This example was created using Midjourney. 

A photo from my desk processed by Deep Dream Creator with the prompt: pointillist style painting in pastel colours, abstracted image, very simplified.

Part 2 (optional) 

Have a look at the data privacy of your chosen platform. Is it clear about the use to which your uploaded images might be put? Will the tool share images or use images to train their data sets?  

What is said about any ethics underpinning the GenAI tool creation and use? (This will be discussed more during the next 12 days.) 

Here is a link to Dream Generators privacy policy and for reference. 

Discussion 

Join us in the Teams space to share your responses and the images you’ve generated. If you don’t have access to the space, email us at teachingexchange@arts.ac.uk for the attention of Hannah. 

  • Do you feel that you have produced an image which is visually interesting?   
  • Did your images pass Du Sautoy’s ‘Lovelace Test’?  Is there something ‘new, surprising and valuable’? 
  • Might you use GenAI images in activities such as teaching or working through ideas for visual projects? 
  • Is there any sense of co-creation with GenAI? 
  • How does this activity and the process of it make you think about the humanistic qualities in creative practice? 
  • Could you discover what might happen with any images you uploaded?