Generate multi-model anchored concepts, drop old Comedy Store refs, and trim the concept gallery

docs/screens/concepts/hollywood_boulevard/README.md

@@ -44,3 +44,15 @@ mild — treat this as the overcast target, not a decay reference.
 ## Across the boulevard
 
 ![Ovation Hollywood frontage across the boulevard](hollywood_boulevard_across_street.png)
+
+## Facade — local-reference anchored, two models
+
+![El Capitan facade anchored on a local research frame, Gemini](el_capitan_facade_gemini_day.png)
+
+![El Capitan facade anchored on a local research frame, GPT image](el_capitan_facade_gpt_day.png)
+
+Both frames are anchored on the same local-only research photo
+(`tmp/research/elcapitan/01.jpg`, gitignored) with `--reference-unchecked`, one with
+`google/gemini-2.5-flash-image` and one with `openai/gpt-5-image-mini`. Provenance and model are
+in each frame's `.refgen.json` sidecar. Local reference only; the photo never enters the repo and
+these frames are art direction, not assets.