* Rewrote 7 generic presets so they no longer assume a face exists in
the input — removed words like `face`, `portrait`, `pores`, `skin`,
`expression` that were causing Flux Kontext to hallucinate faces onto
non-portrait inputs (e.g. warship / missile images). Each preset now
preserves whatever subject is present and includes an explicit
negative against adding new subjects, faces, or characters.
* Added `Ghibli (Clothes & Hair)` — intentionally face-specific, applies
Ghibli illustration only to clothing/hair while locking the face as a
real photograph and the background as the original.
* Added `Zombie (Game Style)` — AAA-game zombie skin/decay/blood while
preserving original colors and lighting (avoids the desaturated
teal/green grade Kontext applies when game references are mentioned).
* Added sample workflow at workflow/Transform-Image.json showing the
intended Load Images -> Prompt Preset -> CLIP -> sampler wiring.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
