v0.2: per-word emphasis, true per-aspect re-layout (switch ratios and the same design recomposes — nothing crops), six show templates, Layer One logos pinned, auto-formatted headlines, expanded motifs + art styles, and paste-to-add guests with auto cut-out & placement.

Phase 2: the Context box feeds an image model (gpt-image-1 / Flux / Ideogram) via a Cloudflare Function; the result drops into this engine and gets the same editorial treatment so it never reads as raw AI.