Compositing AI Tools: A Primer

Nettrice Gaskins
3 min readJan 10, 2023
Nettrice R. Gaskins. “Ọya,” 2023. Created using Midjourney, Deep Dream Generator, and Photoshop.

The text prompt I started with was an art term: sfumato. Sfumato comes from the Italian language and is derived from fumo (“smoke”, “fume”). I was thinking about its use in famous paintings such as da Vinci’s “Mona Lisa.” The painting gives an elusive and illusionistic rendering of the human face, as well as atmospheric effects. Midjourney generated several images but only one represented the term:

Only one of these images illustrates “sfumato.”
Midjourney rendering from the thumbnail.

As Oya opened her mouth fire (spewed)) from her lips… — via Yoruba Religious Concepts

The result took me in a different direction. I remembered that Ọya, also known as the goddess of weather, can call forth lightning, storms, tornadoes and earthquakes. She also breathes fire. The image from Midjourney was a start but I wanted more, so I moved to Deep Dream Generator, which produced this image:

Deep Dream Generator rendering from the Midjourney source image.

Again the key text prompt was “sfumato” and DDG now provides several training models (styles) to choose from, including “Artistic.” What DDG did differently is that is transformed most of the hair into dark smoke, and made the fire and burning embers in the hair more realistic.

Text-2-Dream in Deep Dream Generator.

The final image was missing one thing: lightning. I used free stock images of lightning to create a composite image in Adobe Photoshop. A composite image is a collage created from several images. Thus, the process of generating, then layering multiple images is like making a digital collage.

Pexel’s Free Stock Photos

I never referenced a specific artist in my text or image prompts. Rather, the purpose was to illustrate or represent sfumato, then the Yoruba goddess Ọya. The next day I imagined what would happen if water was used to douse the flames and this was the result:

Nettrice R. Gaskins. “Ọya 2,” 2023. Created using Midjourney, Deep Dream Generator, and Photoshop.

I followed the same process as before using a different image from Midjourney. It was a variation on the prompt. It is possible to have an image in your mind and create many variations until you get one that is close to the mental image. It is also possible to avoid reproducing a specific artist’s work and use free stock images… in the same image composite.

Generative AI tools (minus Deep Dream Gen)

AI tools such as Midjourney, Deep Dream Generator and even Adobe Photoshop are changing frequently, to include more features (ex. filters) and training (style) models. My goal is not to learn to use all of them but master a couple and follow the evolution.

--

--

Nettrice Gaskins

Nettrice is a digital artist, academic, cultural critic and advocate of STEAM education.