0.1–0.3 preserve · 0.4–0.6 balanced · 0.7–1.0 reimagine

⚠ Face / identity preservation requires Kontext or GPT Image. Current model uses the reference loosely for style/composition only.