Lying on deep surfaces.
Triptych in Context Protocol
Context Protocol constitutes a series of deep explorations, structured in three triple prototypes, oriented to transcend certain limits of static and dynamic image-based media, and to penetrate and cross the dominant tipping point in particular contexts managed by synthetic entities. The ambition and inherent risk of this project is based on the convergence of two elements: the power of intuitive expression, intimately intertwined with rigorous analysis, and an understanding of the nature, origin and eccentric limits of three deep learning models, trained by the community and selected according to their degree of specialization:
- WrongTurnXL_ WTSDXL7.7.safetensors (SD XL)
- LightRaysEnd_ contrast.3.7.safetensors (SD XL)
- RealisticReal_ V5.1VAE.safetensors (SD 1.5)
For the revision and updating of textual content, the process is conceived in a collaborative manner, with the following recent model:
The generative process is carried out offline and locally, using own means, without relying on commercial platforms. First of all, the possible implications derived from the training of the different models are considered. The focus of this project is not on the “headline effect”, on the novelty or trend of this actuality, but on the most basic expressive power, combined with the effort to understand how to disrupt certain inertias and constraints, both in relation to the practice of moving image and to the consequences and responsibility of immersion in contexts of generative synthesis. The analytical process of immersion and expression unfolds recurrently in a regressive forward trajectory, adopting an intuitive reverse engineering methodology, which deliberately circumvents the linear and standardized progression characterized by technological advancement.
Context Protocol consists of two phases:
- Context-Protocol (M)
- Context-Protocol (S)
The main body of the project, Context-Protocol (M) is composed of three distinct moving image modules:
The secondary body, called Context-Protocol (S), is composed of three modules of documentation and fable based on static imaginaries and contents that originate from the main body of the project. These modules are:
- First Triptych (S) - in preparation
- Fifth Triptych (S) - in preparation
- Seventh Triptych (S) - in preparation
Lying in deep surfaces is not lying.
Context Protocol explores the intersections between context, technology, and subjectivity, using deep learning models as both tools and conceptual interlocutors. This process leads to an ambitious and urgent inquiry: the development of a manual counterpoint that thoughtfully dialogues with the generative in an exercise of human deep learning.
A hypothetical human deep learning would be grounded in a fertile paradox: assimilating the analytical efficacy of artificial neural networks—their capacity to process layers of data and recognize hidden patterns—to redirect it toward the construction of subjective context, a territory where machines still stumble. If machine deep learning optimizes prediction based on massive correlations, the human variant delves into what data does not yet record: semantic ambiguity, subjective memory, and cultural weight. This approach does not seek to replicate artificial processes but appropriates its layered structure to apply it to organic creative processes. Analogous to how a generative model is trained on millions of images, human creativity "trains" its perception through critical observation, the accumulation of references, and intuitive reworking. The human advantage lies in the ability to weave improbable connections, generating narratives where the technical and the biographical-experiential can coexist.
Protocol de Context is rooted in this hybridization: using machine logic not to automate but to enrich manual creation with a stratified discipline of intuitive attention, error, and reinterpretation. This trilogy acts as an exploratory prototype of the poetics inherent to three distinct community models of generative deep learning (Stable Diffusion 1.5 and XL), as well as their ethical, conceptual, and technical limits. Each piece not only visualizes formal or symbolic outcomes but also attempts to resolve hidden logics within these systems: the tension between randomness and control, the notion of "memory" in datasets, and the illusion of authorship in a process mediated by layers of mathematical abstraction. This initiative highlights a paradox: the more generative models are technically stabilized, the more urgent it becomes to question humanity’s place in an increasingly automated creative ecosystem.
Lora Tensorska
+ DeepSeek-R1-Distill-Qwen-32B-abliterated
