GPT Image 2 Yacht Poster Generation Workflow

2x2 grid, 16:9, do this for 4 unusual luxury yachts that are real: INPUT: $ SUBJECT, [$MOVE1, $ MOVE2, $ SYS1, $ SYS2] (optional; else auto-seed) // Phase 1: If movements/systems not given, sample with subject hash. IF not provided: MOVES = random_choice(Poster_Movements, 2, seed=hash($ SUBJECT)) SYS = random_choice(Graphic_Systems, 2, seed=hash($ SUBJECT)+1) IDENTITY = NAME_FROM(MOVES, SYS) // Build a compositional field Φ that preserves the subject's silhouette and spatial structure. Φ = preserve_silhouette( edge_map($ SUBJECT) ) // gradient of similarity to original F = ∇Φ // flow lines guide typographic grid Curl = ∇×Φ // rotation informs collage/tear edges Div = ∇·F // density maps to halftone/screenprint layers // Poster rendering (left half): // - Apply movement 1's colour palette and movement 2's texture/print logic. // - Graphic systems modulate the build: e.g., if SYS1 = "Torn Paper Collage", displace F with tear vectors; if SYS2 = "Risograph Halftone", apply stochastic dot patterns where |Div| > threshold. // - Typography placed along F streamlines, hierarchy derived from Div magnitude. // - Authentic ink effects: registration misalignment, paper grain, subtle yellowing. // - Silhouette constraint: the subject's outer contour must match $ SUBJECT within 92% IoU. / Render 8K, horizontal.