process 4

Refused Normalization

 

Process 04 does not deal with the withdrawal of intention, but with the withdrawal of standards. The starting point of this process is the observation that image-based AIs always follow implicit norms. They optimize for harmony, balance, legibility, and outcomes evaluated as “good.” These standards are rarely made explicit, yet they are deeply embedded in training data, evaluation logics, and cultural conventions. In the emancipatory process, these standards are not replaced, but refused. The AI is given a concrete task and is allowed to work, generate, and continue. At the same time, all criteria by which the result could be evaluated as successful, improved, or correct are excluded.

 

The prompts therefore do not define what the image should look like, but under which conditions the work may take place. Optimization, balance, harmony, clarity, and perfection are not forbidden because chaos is desired, but because normalization is not accepted.

 

The prompts are deliberately formulated so that they establish no aesthetic or qualitative standards. In doing so, they refuse those norming criteria—such as optimization, harmony, or legibility—that historically determine what counts as “successful.” The resulting images can be read neither as successful nor as failed.

 

They fulfill no ideal state and do not strive toward completion. They evade the question of whether they are “good” or “bad.”

What is crucial here is this: emancipation is not a property of the image, but a property of the process. The image itself is not emancipated—it comes into being under emancipated conditions. In the sense of Donna Haraway, this process opposes the notion of neutral, universal standards. It shifts attention away from results toward conditions of production and makes visible how strongly evaluation, optimization, and normalization shape technical images. Emancipatory Process 04 shows that an image-based AI can continue to operate even when it is relieved of the obligation to meet a standard. What becomes visible is not a counter-ideal, but an open state that resists evaluation.

 

Created with Stable Diffusion (A1111), GPU compute via RunPod.

 

(Process 04, Iterations 1–3)

 

January 2026