Proto-Agonist: Forever 😢~Assets~

2017.

Investigations on the generative adversarial network (GAN) models for learning graphic assets generation from retro game database and internet found images.

︎︎︎ ︎ NeurIPS Machine Learning for Creativity and Design, Art Gallery. (Link)

︎︎︎ ︎ AI Grant Fellow

From 2017:

“This series of digital images and video works were generated by DCGANs (Deep Convolutional Generative Networks), by training with found images online (Google image search results, Tumblr, DeviantArt, indie game developer forums, etc). These source images were created, collected, and annotated by active and surrounding participants of anime and JRPG(Japanese Role-playing Games) fandoms. “proto-agonist” discovers and imagines archetypes and in-between entities of RPGs from a dataset of RPG maker 2D character sprites. “wind&&blood&&glitters (FX array)” is a DCGAN’s attempt to imitate generative process of video game special effect animations and simulation, limiting its reference to old school 2D effects animation sprite images. The network for “, forever 😢~memories~” were trained on anime sky background images and it “walks” through the space spanned from possible modalities of the atmospheric apparatus, which is recurrently used within anime narratives.Prosumer dynamics that can be observed within game and anime fandoms highlight the cultural spirit of this era. Participants of these subcultures are technologically adept, and also aggressive in utilizing and appropriating online spaces for their activities — shared consumption and re-mixing of their cherry-picked cultural fragments. Going through their econo-ecological cycles accelerated by the contemporary infrastructure, (sub)cultural images finally resolves their space and accumulate within the archipelago of online platforms and databases, waiting to be re-discovered and then re-mixed into strains with brand-new labels.

What does it actually mean to participate in such processes with a peculiar degree of devotion? This situation creates an interesting ground for AI. Not only can cutting-edge AI technologies accelerate individual processes in this self-referential feedback loop, but they also provide an arena to extract a compressed version of the meta-narrative that our reality breathes within. Deep neural networks particularly provide a great format for this intellectual and poetic challenge, because it is itself a network model for information processing, which can be a versatile metaphoric structure for existing systems. Adequacy of anthropomorphic metaphors for generative AI, especially the antagonistic metaphor of man versus machine, are skeptical for most of practical and theoretical discussions, but may be useful as a critique of this particular econo-cultural case.”

︎ wind, blood, glitters: DCGAN generated game effect sprites.

︎Dataset: old school J-RPG FX animation sprites collected from game developer and fan art communities

Proto-Agonist: DCGAN generated character sheet. ︎

︎Dataset: old school J-RPG character sprites collected from game developer and fan art communities.

︎Forever 😢~Memories~: DCGAN generated anime sky.
︎Dataset: collected images from Tumblr, tagged "anime" "sky"

alm chung 정앎

info

︎