“Text based generation alone is not enough if you want to control how your end result ends up looking. This is where more advanced methods like using ControlNet (shared in the workshop) can be helpful in directing generated visuals to stick to a reference more closely. Generation process still feels more of a slot-machine type of deal where you cannot fully predict what you will get.”
“All the various components have to “talk” to each other in a set structure while the output they collectively can produce veers more on the uncontrolled/unpredictable territory and this polarity was interesting. What is actually inside each of these models and checkpoints seems like a mini black box where the end user can only make use of them but apparently not be able to modify or inspect their content. It also involved coming to terms with not being able to fully control what I want to use in terms of datasets as these models are already trained on content which I have no feasible way to find out.”
— on how working with models/checkpoints to create images affected their perception of the entangled agencies involved in artmaking
“It was an interesting idea to try and incorporate a self trained visual dataset of AI into the story of an artwork and use TD for movement or interaction but the platform or hardware constraints may be an issue. But will take time to explore a bit more.”
— on what ways they (fore)see ComfyUI and TouchDesigner adding to their creative practice?
entangled agencies module 2 post-workshops form
3 respondents