home site : liqueenie.com

2020 - 

DEEP DREAMSSLOW FAKES

    Work in progress.

Its semiotic transition from an ‘error’ to a ‘dream’, however, incubates a rhetorical proposition that can potentially bridge artistic and computer art languages in new light.

After a successful launch of the Moon Disaster Project, the MIT Centre for Advanced Virtuality, as the mastermind behind it, warned about the smudging of fictional and realistic fronts and the potential end of trustworthy journalism [Panetta et al, 2020]. The current applications of deepfakes situate mainly around fake news and fake pornography. Amongst the 14,698 pieces of deepfake videos identified on the internet at the end of 2019, 96% are non-consensual pornography [Giorgio, 2019]. As against the Deepfakes genus which is often criticised to manipulate truths, DeepDream generates images that are instead hailed for its dynamic plasticity.

Image: MIT Center for Advanced Virtuality.

DeepDream, created by Google engineer Alexander Mordvintsev in 2015, is a computer vision program that utilises a convolutional neuronal network and a pareidolic algorithm to generate a specific tactility of images that are often seen as hallucinogenic. The code prescribes a trained deep network to identify visual layers in the input image, then modify the image to enhance desired patterns through iterative activations. As against the Deepfakes genus which is often criticised to manipulate truths, DeepDream generates images that are instead hailed for its dynamic plasticity. 

Open-source library: link

Image: Google AI.

As its distinctive output style is what defines its essence, the program has mostly seen as an end rather than a means. Exposing as-yet overcome limitations of machine learning in realistic visual generation, it is largely seen to be a window to the science of artificial neural networks, whilst yet to be internalised within mainstream visual culture and artistic mediums.

 

A similar self-reflexive path of photography may offer us a sneak peek to potential futures of DeepDream. Identifying a set of visual strategies that specify layers of visual constructs, octave, a.k.a. size of selected visual constructs, learning rate and tiling frequency for the trained neuro network, the tailored output becomes a polymorphic silhouette. What is the worth of such a blend of multiplicity and ambiguity? 

The traditional canon of portraitures was unprecedentedly disrupted by the invention of photography in the 19th century. How about today - how do machines see us? How do we see ourselves through machines? Hallucinating images by DeepDream recalled deeply intricate sentience amongst us, and especially towards contemporary political leaders. 

Slow Fakes (2020- ).

A set of 6 generative images by DeepDream. 

If the casualty of fake materials that stemmed from deepfakes is attributed to its snowballing effect powered by light speed distribution, perhaps a silver lining lies in suspending the tempo. What is hinted is perhaps an artistic gesture: to distort the destructive power of deepfakes through manipulating their global circulation speed.

What are the visual stories that travel slow? The Fluxus movement emerged in the 1960s particularly invested in mail art to navigate the ground-based communication networks that thrived before the internet boom. This humble art form celebrates a kind of egalitarian relationship that trespasses the capitalist regime of elitist gallery structures. It puts forward a resistance to the contemporary rule of thumb that worships the God of speed, and offers an additional path to engage a broader (albeit possibly narrower) network of actors whose agencies are not predetermined

Image: Simon Cutts, The World Exists to be Put on a Postcard, 2013.

Slow Fakes (2020- ).

Custom-printed postcards of DeepDream portraiture, 12.7x17.8cm, 32pt uncoated, naturally textured Mohawk Superfine papers.

What are the visual stories that travel slow? The Fluxus movement emerged in the 1960s particularly invested in mail art to navigate the ground-based communication networks that thrived before the internet boom. This humble art form celebrates a kind of egalitarian relationship that trespasses the capitalist regime of elitist gallery structures. It puts forward a resistance to the contemporary rule of thumb that worships the God of speed, and offers an additional path to engage a broader (albeit possibly narrower) network of actors whose agencies are not predetermined

Slows Fakes is a series of postcard art that experiments with the DeepDream programme to create tailored images for postcard art and explores the analogue distributive network of Deepfakes imageries. Through slowing down its real-time exponentiality, the work hopes to spark further imagination between new imaging technology within our current social-political contexts.

Think about the postmen who might flip through the postcard, the neighbours of the postcard recipient who may be attracted to the visuals on a voyeuristic motive, and the acquaintances of the recipient to whom the delivered postcard is shown and shared; these are organic feedback loops that exist outside the monitored and coded realm, rippling through in-personal interaction and real-life encounters.

DeepDream Image of El Zeft’s stencil of a riot-ready Queen Nefertiti mask (2012) during the Egyptian Revolution.

Soft Fakes (2020-).  A DeepDream collage printed on synthetic silk, 139x190cm, with uneven scissor cut on edges.

In addition to the alteration of image dissemination speed, what are further potentials of manifesting generative images, liberated from screen-based image networks to entail tangible agency? Soft Fakes intend to suggest alternative formats for deepfake and DeepDream imageries to engage audiences apart from journalistic and entertaining objectives.

A series of masks that prevailed in different global activism waves such as the Egyptian Revolution and Extinction Rebellion in the last decade are stitched together to imagine a new layer of consciousness. The collage fabric ostensibly shields an unknown organism gasping beneath, evoking a sense of vulnerability in the age of multiplicity.

The narrative of activism is deliberately selected in this intervention to match with the thematic investigation of ‘dream’ in this prompt, which is situated between the authentic root of archives and the fictional agency of memories. This artistic staging, in particular, questions what might be lost in the mesh of ideological and historic bricolage.

Soft Fakes (2020- ). Work in Progress.

A 2-minute video excerpt of installation documentation, DeepDream imagery and sonic archives. 

This work aims at an in-depth unpacking of artistic, technological and political notions that interweave through Slow Fakes and Soft Fakes. One of the implications lies in how fictitious materiality and ambiguous aesthetics could potentially empower individuals through provocative language that might spark new conversations in social-political debates. Besides, the physiological and psychiatric aspects of ‘dreaming’ might be worth further exploration to move forward this series of generative art with neural networks.   


Every arrival of new visual forms has an intricate journey ahead until occupying a unique position within the existing art canon. The politics of images are never confined within the visible, as the production process is as important as its consumption and circulation which yield an ecology that could have political implications. Is the art powered by generative images able to offer a hint of regeneration? If one could reclaim the rights of imaging techniques inspired by machine-learning advancements, deconstructions in artistic and creative approaches might help to generate new meanings, offering a mediative space that slows down the tempo and transfigures the contested deepfake spectacle.            (To be Continued)

Acknowledgement: the MIT Media Lab Experiments in Deepfakes; the Pilot grant from the ACT/Council for the Arts at MIT (2020).