
Originaldokument
GPT Image 2 · en
Scene:
A square fine-dining hero shot photographed from directly overhead on a deep black marble table with soft veining. A single matte white porcelain plate sits slightly off center with generous negative space around it, lit by a precise top light with a faint 45 degree fill from the upper left. The atmosphere is quiet, premium, and editorial, like a Michelin restaurant campaign image made for a luxury magazine spread.
Subject:
At the center of the plate, build a small architectural tasting course around a quenelle of chilled creme fraiche, a restrained mound of glossy black caviar, tiny cubes of translucent kohlrabi, a few gold-leaf accents, and delicate edible flower petals in pale violet and white. Add a thin ring of clear herb oil and several micro dots of dark reduction so the dish reads as intentional constellation-like plating rather than random garnish. The portion should feel expensive and precise, not abundant.
Important details:
Preserve true culinary realism: caviar beads must look separate and hydrated, creme fraiche should hold a soft spooned texture, porcelain should remain clean, and sauce dots must have crisp edges. Keep the garnish logic sparse and balanced, with every component readable at a glance. The visual language is minimalist French fine dining with immaculate spacing, premium ingredient knowledge, and commercial food-photography polish.
Use case:
Michelin Plate Hero Shot single study 1/8, focused on a luxury caviar plate with disciplined top-down composition.
Constraints:
No readable text, no watermark, no logos, preserve exact 1:1 aspect ratio, no cutlery, no hands, no table clutter, no oversized portions, no messy sauce smears, no impossible garnish placements, no extra plates.Verwendung
- Den vollständigen Original-Prompt einschließlich Parametern kopieren.
- Motiv, Text und Komposition an die eigenen Anforderungen anpassen.
- Ein kompatibles externes Werkzeug nutzen; Zugangsbedingungen und Ergebnis prüfen.
Vor dem Start
- Prüfen, ob ein Referenzbild nötig ist. Kompatibilität belegt nicht das tatsächlich verwendete Modell.
- Namen, Zahlen und Bildtexte prüfen; Modellversionen können andere Ergebnisse liefern.





