Prototype V2: play with an LGM to co-create machine generated images based on image prompts selected and curated from the Tate’s digital collection Artist and Art website.
The current prototype called V2 is my exploration into generative AI tools that are open-source and fairly accessible to a lay person without a paywall.
The tools are assessed in terms of their potential to be used as part of a workshop for instance for participants to engage with their first predictive AI platform, whilst using images from the Tate’s digital collections. on their Art and Artists portal. I am currently experimenting with an LGM (Large Multi-view Gaussian Model) for High Resolution 3D generation created by a team of scientists at Beijing University, Nanyang Technological UNiversity and Shanghai University. The platform enables the generation of 3D models either through textual prompts, whereby a six second footage of a three dimensional image will be rendered. Or the user can upload an image of their choosing onto the platform and the LGM will generate a 3D object from this image. The model is impressively efficient at making three dimensional two dimensional objects and is particularly successful in generating images of sculptures for example as three dimensional objects.
The use cases become even more interesting when we take objects that are fully two dimensional such as paintings. I tested the system with David Hockney’s “A Bigger Splash” (1967), one of the most searched works on Tate’s digital collections site. The result was a floating cube, with its four outwards facing sides showing increasingly abstract renditions of the original painting. THe first side would have an image that is almost identical to the original, the following face will have only the pool and the splash with the springboard and a palm tree, the following imaged will increasingly abstract figurative elements to the point that only brushes of colour and structuring movements of the painting appear.
The 3d object has a certain comic effect which contributes to the enjoyment of using this tool that can produce surprising new visual objects. But beyond that this re-imagined picture can be thought as enlightening us in two ways:
-through the glitch or hallucination, the machine shows us how it sees this flat image when it is trained to be offered three dimensional objects. The machine approaches the image through clusters of pixels and thus is able to generate an image that holds essential qualities of Hockney’s Bigger Splash, whilst also failing to generate a flat painting as a three dimensional object.
-this process of abstracting the painting and reconstructing its individual elements (the pool, the house, the splash, the palm tree) can also offer opportunities for the user to engage creatively with the collection because it is fun to use (the outcome of trying to generate Sarah Lucas’ Self-Portrait with Fried Egg is comically uncanny). But it also offers new ways of seeing these works of art through the eyes of the machine and perceiving the paintings in new ways. For instance with the Hockney example, the increasing abstraction of this figurative painting can prompt us to look at the painting’s individual elements in more focus (the tree, the house, the pool) or wonder how the arrangement of colours and shapes builds an appealing image. It can also make us think how seeing this image might miss the contextual information that might make this work historically important. Just as much as it can make a work interesting by abstracting curatorial mediation of the work’s materiality and historicity.
The V2 prototype troubles the usual arrangement of actors, environments and infrastructures of the museum. It troubles the autonomy of the work of art as held in the collection, it troubles the curatorial authority of mediation, it troubles the institutional space of the gallery and website as the sole site of representation and reception of the work, it troubles the passive relation of spectator to active artist. In short this example holds a first potential for an ignorant use of generative AI to use the collection in unexpected and surprising new ways that generate new visuality for sure, and which we need to push further to see if it can be used to generate collective modes of knowledge production about the digital collection and the AI black box itself.
Resources
Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng & Ziwei Liu, LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation >>https://me.kiui.moe/lgm/
LGM Model on Huggingface >> https://huggingface.co/spaces/ashawkey/LGM
Tero Heikkinen, Petri Kaverma & Denise Ziegler, Prototyping as research methodology (Finnish) >> https://taju.uniarts.fi/handle/10024/6021
Also in: Terro Heikkinen, Petri Kaverma & Denise Ziegler, The Postresearch Condition: Five Earn Working Groups, pp.27-29, in The Postresearch Condition, edited by M. Slanger, 2021 Metropolis M, MaHKUscript Series, Utrecht >> https://www.hku.nl/getmedia/d35e65f1-7a29-4afd-ae7b-8627028f21b3/Publication_The-Postresearch-Condition.pdf
Tate Digital Collection >> https://www.tate.org.uk/art
Tate, Practice as Research >> https://www.tate.org.uk/research/research-centres/tate-research-centre-learning/practice-as-research