#t3t2

VIP FREE| Yeezy Style | Top&Bottom | #combination2

VIP:

Top: Crop Dress (body BB2-#comb2 (#combination2) ▯ 16 rec (Mesh by me)

Bottom: Yeezy slippers (body BB2-#comb2 (#combination2) ▯ 11 rec (3t2)

Free:

Deco: 2 types cats ▯ 7 rec (4t2)

Credits: easonfys | complex

DOWNLOAD ALL | HERE

BLIP Captions/ Text -- BGE Model Match -- Play structure = Agent Preference

The agents navigate through urban segments and select the city elements that match their own tendencies. These selected fragments are reorganized and combined to generate new spatial behaviors and relationships within the city.

Language

Art

Law

War

Myth and Ritual

Through matching five leading figures from the architectural school with the spatial characteristics of Rotterdam, whilst connecting them to five databases—art, war, theology, law, and language—I have constructed five ‘algorithmic avatars’:

1. The Situationist — Constant Nieuwenhuys

Representative theories: New Babylon, The Player.

Algorithmic preferences: floating networks, scaffolding, fluid spaces that break down boundaries.

CLIP prompts: ‘Floating networks, labyrinthine scaffolding, playful continuous structures, anti-capitalist spaces’

2. The Provocateur — Rem Koolhaas (OMA)

Representative theories: Culture of Crowding, Bigness, Programmatic Friction.

Algorithmic preferences: Massive modern commercial spectacles, glass megastructures, vertical stacking.

CLIP prompt: ‘Massive metropolitan scale, programmatic friction, glass skyscrapers, radical modernist spectacle’

3. The Data-Stacker —— Winy Maas (MVRDV)

Representative theories: Data landscape, giant bonsai, pixelated assembly.

Algorithmic preferences: Exaggerated symbolic containers, high saturation, voxelised collages.

CLIP prompt: ‘Voxelised pixel architecture, extreme colourful data landscape, hyper-pragmatic pop-art containers’

4. The Structuralist — Piet Blom

Representative theories: The city as a forest, cellular proliferation.

Algorithmic preferences: Non-orthogonal geometry, compact residential pods, polygonal repetition.

CLIP prompt: ‘Cellular proliferation, non-orthogonal geometric clusters, dense mutated residential forms’

5. The Rational Machine — J.A. Brinkman (Nieuwe Bouwen)

Representative theories: Nieuwe Bouwen, absolute rationality, machine aesthetics.

Algorithmic preferences: hardcore industrial ports, steel frameworks, factory buildings, cranes.

CLIP prompt: ‘Hardcore industrial machinery, rational steel grids, functionalist port infrastructure, pure mechanical aesthetic’

5 Key Database Settings

These 5 databases serve as the ‘rules of the game’ that govern Agent behaviour.

Art: Expressiveness, sculptural forms, vivid colours, expressive visual tension.

War: Defensive bunkers, post-apocalyptic ruins, heavy brutalist armour

Theology: Sublime verticality, monumental scale, divine light and shadow

Law: Strict grids, rigid boundaries, ordered institutional repetition

Contest_Agon

Order_Rules

Ritual_Ceremony

Spatial_Separation

2,504 sets of coordinates and landmarks for cityscapes in Rotterdam

Som 41*41, 5000 iterations

rotterdam_som_landmark_hotspots_41x41.png

To build a systematic urban corpus, Rotterdam’s landmarks can be categorised into four spatial dimensions

1. Modernist Centre & Spectacle

Markthal: MVRDV

Kubuswoningen: Piet Blom

Depot Boijmans Van Beuningen: MVRDV

De Rotterdam: OMA

Rotterdam Centraal

2. Waterfront & Bridges: The core structures connecting the north and south banks serve as the city’s ‘connectors’ and ‘pathways of movement’ across physical boundaries:

Erasmus Bridge (Erasmusbrug)

Willem Bridge (Willemsbrug)

Kop van Zuid Peninsula

3. Industrial & Port

Europoort

Maeslantkering

RDM Rotterdam

4. Historic & Residential

Within the highly modernised urban fabric, these surviving fragments of history provide a striking contrast in scale:

Delfshaven: one of the very few old districts in Rotterdam that was not destroyed during the bombing, preserving traditional Dutch canals, windmills and red-brick gabled buildings.

Van Nelle Factory: A masterpiece of the ‘New Building’ style from the 1920s, featuring extensive glass curtain walls and steel structures that emphasise light and air; it is listed as a UNESCO World Heritage Site.

Sint-Laurenskerk: The only surviving medieval Gothic building in the city centre, it stands as a unique visual anchor amidst the surrounding modern skyscrapers.

War footage is first translated into BLIP captions and embedded into a shared semantic field through the CLIP text encoder. A SOM map is then constructed to organize the war archive into distributed cognitive regions, where semantically related scenes gradually cluster together during training. Through this process, the War Agent develops its own internal tendencies, forming different behavioral zones within a single memory system. Rather than functioning as a fixed classifier, the agent operates as a dynamic field of preferences shaped by the structure of the dataset itself.

When a new text or image is introduced, it is transformed into a CLIP embedding and positioned within the agent’s memory space. The agent evaluates the relationship between the new input and its learned tendencies, returning the closest, middle, and farthest memories as a way of revealing how the war archive interprets movement, conflict, territory, and spatial behavior.

The urban fragments are then extracted as segmented objects from video frames, such as vehicles, infrastructures, or moving elements. Each fragment is independently captioned and embedded into the same semantic space as the agent’s memory. Instead of selecting complete scenes, the War Agent navigates through these fragments and chooses the objects that most closely align with its own tendencies. These preferred fragments are later reorganized and aggregated, generating new spatial relations and behavioral structures within the city.

All paintings are first converted into BLIP captions and embedded into a shared semantic space using the CLIP text encoder. A SOM map is then generated, organizing the dataset into cognitive territories through a neural grid structure. During training, similar paintings gradually gather into different clusters, forming internal tendencies within a single Art Agent. These clusters act as different personality regions inside one cognitive system. When a new image or text is input, the system first converts it into a CLIP embedding and compares it with the agent’s memory space. The agent then returns the closest, middle, and farthest items from its memory, revealing how the dataset itself interprets the new input.

Each dataset is trained as an independent agent, so different datasets produce different personalities. Because each agent is formed from a different semantic structure, they develop different tendencies, preferences, and ways of interpreting images and space. The agents then navigate through urban fragments and select the city elements that best match their own tendencies. These selected fragments are reorganized and combined to generate new spatial behaviors and relationships within the city.