How recipes
are crafted.

Honest about the pipeline — because dishonesty about an AI-assisted process is the part that fails everyone (readers, search engines, and the recipes themselves).

Each recipe is produced by an AI-assisted authoring pipeline (Claude Opus 4.7) on top of the documented Fujifilm catalogue: 17 film simulations, 13 camera bodies, and the published parameter ranges for each setting (highlight, shadow, colour, sharpness, noise reduction, clarity, white balance shifts, dynamic range, grain). The model writes a recipe candidate; the editorial gate decides whether it's published.

Parameter validity

Every numeric setting must fall inside the range that Fujifilm cameras actually accept. Highlight and Shadow are integers in [−2, 4]. Colour, Sharpness, and Noise Reduction in [−4, 4]. Clarity in [−5, 5], omitted for cameras that don't have a Clarity control. White-balance shifts in [−9, 9]. The gate parses the recipe and rejects anything outside.

Camera × film-simulation compatibility

Each recipe declares a list of compatible Fujifilm bodies. The gate rejects mismatches: Classic Negative isn't on the X-T3, Nostalgic Negative only ships on later X-Trans V (and selected IV) bodies, Reala Ace arrives only on X100VI / X-T50 / X-E5 / X-M5 natively and via firmware on X-T5 / X-H2 / X-H2S. A recipe that asks for a Clarity adjustment on a body that lacks Clarity is rejected.

Anti-thin prose

Each recipe's body must reference at least three of its own specific parameter values and their photographic effect. A boilerplate "this recipe is great for…" paragraph that could be pasted onto any other recipe is rejected. A short list of generic phrases ("stunning results", "take your photos to the next level", "elevate your photography") triggers immediate rejection.

Anti-duplication

A new recipe is compared against every published recipe by three measures: exact slug collision, a near-identical parameter set against any existing recipe with the same film simulation, and word-shingle similarity of the prose body. Near-twins are rejected — every published recipe earns its own URL by being genuinely different.

It isn't a quiz-mill that publishes whatever the model says. Rejection is normal. Each batch publishes some recipes and rejects others — the rejected ones get logged with the gate's reason, and the next batch's seeds get sharpened to avoid the same failure.

It also isn't a personal-experience archive. We don't pretend a photographer named Alex shot this recipe at golden hour on a beach in Lisbon. Recipes are described in editorial brand voice — knowledgeable, restrained, useful — because that's the truth of how they're produced.

Search engines in 2026 ask whether AI-assisted sites are honest about their workflow. The answer matters for indexing and for reader trust. Telling you exactly how recipes are made — and what gets rejected — is the most useful version of that disclosure.