
原始文档
GPT Image 2 · en
Scene:
Compose a square beverage campaign image built around ritual, vessel, light, and atmosphere. The frame should feel like it belongs to a premium drink brand that understands daily habits: ceramic warmth, glass condensation, steam behavior, and tactile surfaces all need to read as observed and real rather than generic lifestyle filler. Keep the same tabletop geometry and campaign polish, but shift to twilight tones, reflective glass, and a crisp rim light that catches bubbles and ice edges.
Subject:
The hero beverage must feel consumable and culturally grounded, with believable proportions, realistic liquid behavior, and supporting tools or props that reinforce the ritual without overwhelming the drink itself. Show a tonic-style coffee, tea, or nonalcoholic evening drink being poured over clear ice in a premium glass, supported by one minimal garnish.
Important details:
Credibility comes from timing and materials: crema density, ice clarity, tea translucency, condensation trails, and the way a hand or vessel sits in space. Keep the mood intentional and campaign-ready, with enough atmosphere to suggest a lived routine while preserving product focus. Preserve the exact camera height, lens feel, prop family, tabletop scale, palette discipline, and brand logic established by 01-morning-bar-anchor. Preserve the reference composition and prop restraint while letting carbonation, condensation, and glass reflections carry the new time-of-day story.
Use case:
Coffee Rituals series study 4/9, built for Prompt Atlas coverage of a food and beverage commercial visual system.
Constraints:
No readable text, no watermark, no logos, preserve exact 1:1 aspect ratio, no plastic café props, no latte-art gimmick overload, no fake steam clouds, no unreadable cup slogans, no random breakfast clutter, no artificial neon excess, no nightclub scene, no cocktail umbrellas, no bar-label branding.使用方法
- 复制完整原始提示词,并保留其中的参数。
- 按需求替换主体、文字和构图要求。
- 在兼容的外部工具中使用,核对当前使用条件并检查结果。
开始前需了解
- 先确认是否需要参考图。标注兼容模型,不代表该示例由这个模型生成。
- 核对名称、数字与画面文字;不同模型版本的结果可能不同。