Multi-Limbed Sculptures - Why?

As part of our initial creative pitch, we used free AI generators DALL-E 2, Craiyon and Stable Diffusion to ‘co-design’ our sculptures for the show. We knew we wanted sculptures of of us with screens mounted to them to display our AI avatars. However, combining the idea into a single effective prompt proved quite difficult. These are some example outputs (July 2023) from DALL-E 2 using prompts like: “A full image of detailed cast of a person in white plaster sitting on a box plinth with a metal stand protruding out of their arm with a TV mounted to it”.

As we moved on with our project ideation, we decided to situate the figures facing each other in a conversational pose - without screens. This simpler request proved easier to accomplish a more desirable output - and it enabled us to test a variety of different styles and material combinations. The following examples were generated by Stable Diffusion (July 2023).

As a quick visualisation tool, AI image generators can be really great. Doing this helped Bill and I decide which direction would be most suited to the focus of our project. While we quite liked some of the more abstract and modernist renderings, we decided to take a more mass-produced approach using mannequins as the base for our design. Not only would this reduce production time, it would also signal the way in which a lot of AI avatars are built over existing/set base model architectures.

As we tested different prompts, we started to notice recurring issues with AI tools effectively generating limbs. Here are some fairly ‘usual’ examples from Stable Diffusion (July 2023) . [We tended to use Craiyon and Stable Diffusion quite a lot in the early design phase, due to limited free monthly DALL-E 2 credits].

In these examples, you can see the way extra limbs are often added to the composition. It is a fast moving terrain, but even in more the more recent DALL-E 3 (a marked improvement), you will often find an extra arm included in the image. Here is a Bing Chat (DALL-E 3) example from 29th November. Since, Christmas is fast approaching, I used the prompt: create an image of a family christmas photo.

Note the clones in the background and extra hand holding the ‘Happy Holidays’ sign…almost, but just not quite right….

We embraced the mannequin idea and tried to find the right prompt to incorporate screens into the design. Here are some trials from Stable Diffusion (July 2023).

In the end, we decided to use the multi-limbed mannequins as our base design, but create our own mock-up of the screen design in an image editor, to ensure it worked with the idea of a public/interactive setting.

Initial draft design using Stable Diffusion and augmented in an image editor.

Final draft design using Stable Diffusion and augmented in an image editor. The aim is to use mannequins as a base and then customise them with life cast elements (face, hands, feet etc.). This should produce quite a strange and unsettling visual experience that captures the uncanniness of current AI renderings.

And that is why we have multi-limbed sculptures - cos’ AI currently loves throwing in an ‘extra’ here and there.