How I use AI as a creative tool

Written By Sarah Daly

I am a painter who integrates artificial intelligence (AI) into my creative workflow, combining traditional techniques with advanced machine learning tools to push the boundaries of artistic expression. My approach is collaborative, where AI acts as a generator of ideas, while the final piece is shaped by my artistic judgment. I have been using AI as part of my creative process since 2018.

The Process: Combining Traditional Art and AI

My process begins with a visual database I’ve curated, which includes detailed shots of my own paintings, along with images of textures, patterns, and contextual elements such as maps, marble surfaces, plants, and architectural features. These inputs are chosen based on their relevance to the project and their potential to inspire new visual directions.

Below are a selection of the process and source images I used to create a recent commission.

I primarily use Deep Style, a neural style transfer technique that leverages pre-trained convolutional neural networks (CNNs) such as VGG-19. This technique extracts two sets of features:

  • Content features from one image, which represent its structural and spatial elements.

  • Style features from another image, which capture its textures, colors, and patterns.

Using optimization algorithms, Deep Style blends these features to generate new images that retain the content of one input while incorporating the stylistic elements of the other. This process involves defining and minimizing a loss function, which measures the deviation between the desired and generated images.

Iterative AI Collaboration

For each project, I generate an initial series of AI outputs using Deep Style. These are evaluated for aesthetic potential, balance, and how well they align with the thematic goals of the piece. The initial outputs often serve as the base layer of the painting, over which I manually refine and build additional layers.

The iterative process continues as I generate further variations with Deep Style, tweaking parameters like:

  • Layer selection: Adjusting which CNN layers influence content and style to emphasize specific features.

  • Style weight and content weight: Balancing how much influence each input image has on the result.

  • Resolution adjustments: Scaling outputs to ensure compatibility with the final composition.

While AI contributes to generating ideas, its outputs are not the final artwork. My role involves analyzing the generated visuals for patterns, contrasts, and movements that resonate with the painting’s intended direction. Some AI outputs are discarded for lacking coherence or balance, while others are incorporated and reinterpreted through traditional painting techniques.

The Role of Traditional Methods

Once I finalize the AI-generated base layers, I begin painting on linen canvas using oil paints, turpentine, and beeswax. The tactile nature of painting allows me to apply layers of texture, adjust color luminosity, and create depth that AI tools cannot replicate. I take new images of the paintings as it progresses and add these to my visual database. I create new AI-generated layers as I am working on the painting, and these inspire - but do not dictate - the next steps. The process evolves as a visual conversation. Each painting takes around eight weeks to complete, with every step blending AI outputs and manual intervention.

‘We saw, along the quay…’ Sarah Daly (2023) Oil on canvas

Tools and Experimentation

In addition to Deep Style, I experiment with other AI platforms, including:

  • DALLE2: For generating conceptual imagery based on textual descriptions.

  • Night Cafe and MidJourney: For creating highly stylized interpretations of source material.

  • Deep Dream: For enhancing patterns in existing images through feature amplification.

These tools offer diverse capabilities, enabling me to explore different facets of AI-aided art. However, I continually refine their outputs, ensuring that the final work aligns with my vision and aesthetic standards.

Why Use AI?

AI enables me to explore perspectives and visual combinations that would be impossible to generate manually. It acts as a creative partner, expanding the possibilities within my practice while leaving critical aesthetic decisions to me. For me, the process is about discovery and collaboration rather than automation.

‘We saw, along the quay…’ Sarah Daly (2023) installed

 

Interested in art that tells your story?

Let’s have a conversation.

Sarah Daly
Previous

Creating ‘Emerald in the Deep’
Next

Thinking about art that tells your story?